Magnetoencephalography—theory, instrumentation, and applications

to noninvasive studies of the working human brain
Matti Hamalainen, Riitta Hari, Risto J. Ilmoniemi, Jukka Knuutila, and Olli V. Lounasmaa
Low Temperature Laboratory, Helsinki University of Technology, 02150 Espoo, Finland
Magnetoencephalography (MEG) is a noninvasive technique for investigating neuronal activity in the living human brain. The time resolution of the method is better than 1 ms and the spatial discrimination is,
under favorable circumstances, 2 - 3 mm for sources in the cerebral cortex. In MEG studies, the weak
10 fT-1 pT magnetic fields produced by electric currents flowing in neurons are measured with multichannel SQUID (superconducting quantum interference device) gradiometers. The sites in the cerebral
cortex that are activated by a stimulus can be found from the detected magnetic-field distribution, provided that appropriate assumptions about the source render the solution of the inverse problem unique.
Many interesting properties of the working human brain can be studied, including spontaneous activity
and signal processing following external stimuli. For clinical purposes, determination of the locations of
epileptic foci is of interest. The authors begin with a general introduction and a short discussion of the
neural basis of MEG. The mathematical theory of the method is then explained in detail, followed by a
thorough description of MEG instrumentation, data analysis, and practical construction of multi-SQUID
devices. Finally, several MEG experiments performed in the authors' laboratory are described, covering
studies of evoked responses and of spontaneous activity in both healthy and diseased brains. Many MEG
studies by other groups are discussed briefly as well.

CONTENTS
I. Introduction
A. General comments
B. The human brain
C. Basics of magnetoencephalography
D. Detection of neuromagnetic
fields
E. Magnetic fields from the human body
II. Neural Basis of Magnetoencephalography
A. Constituents of the brain
B. The neuron
C. Ion mechanisms
D. Postsynaptic potentials
E. Action potentials
III. Neural Generation of Electromagnetic Fields
A. Quasistatic approximation of Maxwell's equations
B. Primary current
C. Current dipole
D. Integral formulas for V and B
E. Piecewise homogeneous conductor
1. Integral equations for V and B
2. Spherically symmetric conductor
3. Realistically shaped conductor models
IV. Inverse Problem
A. General
B. Lead
field
C. Value of different measurements
D. Comparison of magnetoencephalography and electroencephalography
E. Combination of magnetoencephalography with tomographic imaging methods
Probabilistic approach to the inverse problem
1. The general solution
2. Gaussian errors
3. Alternative assumptions for noise
4. Confidence regions
5. Marginal distributions
The equivalent current dipole
1. Finding dipole parameters
2. Estimation of noise
3. Validity of the model
4. Confidence limits
Multidipole models
Reviews of Modern Physics, Vol. 65, No. 2, April 1993

414
414
416
416
418
420
421
421
422
423
424
425
425
425
426
426
427
427
427
428
428
430
430
431
431
432
433
433
433
434
434
434
435
435
435
436
436
436
437

1.
2.
3.
4.

Formulation of the problem
Selecting the correct model
Linear optimization
Finding dipole positions
a. The multiple-signal-classification approach
b. Variable-orientation dipoles
c. Fixed-orientation dipoles
d. The combined case
e. Applications and limitations
I. Current-distribution models
1. Minimum-norm estimates
2. Regularization
3. Minimum-norm estimates with a priori assumptions
J. Future trends
V. Instrumentation for Magnetoencephalography
A. Environmental and instrumental noise sources
B. Noise reduction
1. Shielded rooms
2. Gradiometers
3. Electronic noise cancellation
4. Filtering and averaging of neuromagnetic data
5. rf filters and grounding arrangements
C. Dewars and cryogenics
D. Superconducting quantum interference devices
1. SQUID basics
2. Thin-film fabrication techniques
3. Sensitivity of the dc SQUID
4. Optimization of flux transformer coils
5. Examples of dc SQUIDs for neuromagnetic applications
6. Integrated thin-film magnetometers and gradiometers
7. High-r c SQUIDs
E. dc SQUID electronics
1. Electronics based on flux modulation
2. Direct readout
a. Additional positive feedback
b. Amplifier noise cancellation
3. Digital readout
F. Multi-SQUID systems
1. Optimal coil configuration
2. Sampling of neuromagnetic signals
Copyright ©1993 The American Physical Society

437
438
438
439
439
440
440
440
440
440
441
441
443
443
444
444
444
444
445
447
447
448
448
449
449
450
451
452
453
454
454
455
455
457
458
458
459
460
460
460
413

414

Hamalainen et a/.: Magnetoencephalography

3. Information conveyed by the magnetometers
4. Practical aspects
5. Examples of multichannel systems
a. T h e 24-channel gradiometer at the Helsinki
University of Technology
b. The Siemens 37-channel " K r e n i k o n " systems
c. The BTi gradiometers
d. The P T B 37-channel magnetometer
e. The Dornier two-Dewar system
f. 19-channel gradiometers
g. The 28-channel "hybrid" system in R o m e
6. Whole-head coverage
a. The "Neuromag-122" system
b. The C T F instrument
c. Japanese plans
7. Calibration of multichannel magnetometers
8. Probe-position indicator systems
9. D a t a acquisition and experiment control
VI. Examples of Neuromagnetic Results
A. Auditory evoked responses
1. Responses to sound onsets
2. Responses to transients within the sound stimulus
3. Mismatch fields
4. Effects of attention
5. Audio-visual interaction
B. Visual evoked responses
1. Responses from occipital areas
2. Responses to faces
C. Somatosensory evoked responses
1. Responses from SI
2. Responses from S2
3. Painful stimulation
D . Spontaneous activity
1. Alpha r h y t h m
2. M u and tau r h y t h m s
3. Spontaneous activity during sleep
E. Magnetoencephalography in epilepsy
1. Background and data analysis
2. Example: boy with disturbed speech understanding
F. Other patient studies
VII. Conclusions
Acknow ledgments
Appendi X
References

461
462
462
462
463
464
465
465
465
466
466
466
467
468
468
468
469
469
469
469
470
472
473
474
475
475
476
477
477
477
478
478
478
479
480
481
481

about the functioning of the human brain, and it is expected that M E G will become increasingly more important in the near future as clinical applications emerge. In
many M E G measurements, an area in the brain that has
been activated by a stimulus, such as a sound or a picture, can be located by analyzing the detected magnetic
fields. A typical signal, evoked by a tone burst, is shown
in Fig. 1. In addition, spontaneous brain activity has also
been studied successfully.
In this introduction, we shall briefly consider the basics
of magnetoencephalography. Section II of this review
explains the neuronal basis of M E G . In Sees. I l l and IV
we discuss the theory of the method in detail. Section V
contains a thorough description of SQUID instrumentation as applied to magnetoencephalographic measurements. Finally, in Sec. VI, we present several illustrative
examples of our own M E G studies and review briefly
some of the work from other laboratories.
The M E G method is based on the superconducting
quantum interference device or SQUID, a sensitive detector of magnetic flux, introduced in the late 1960s by
James Zimmerman (Zimmerman et aL, 1970). The first
SQUID measurement of magnetic fields of the brain was
carried out at the Massachusetts Institute of Technology
by David Cohen (Cohen, 1972). He measured the spontaneous a activity of a healthy subject and the abnormal
brain activity of an epileptic patient. Evoked responses
were first recorded a few years later (Brenner et aL, 1975;
Teyler et aL, 1975).
SQUID magnetometers are used also in studies of magnetic fields emanating from several other organs of the
human body, notably the heart. The whole area of
research is referred to as biomagnetism. In the context of
estimating the distribution of electrical sources within

484
485
486
486
486
487

N100m

100 fT/cm
I. INTRODUCTION
A. General comments
The human brain is the most complex organized structure known to exist, and, for us, it is also the most important. There are at least 10 10 neurons in the outermost
layer of the brain, the cerebral cortex. These cells are the
active units in a vast signal handling network, which includes 10 14 interconnections or synapses. When information is being processed, small currents flow in the neural
system and produce a weak magnetic field which can be
measured noninvasively by a SQUID magnetometer,
placed outside the skull, provided that thousands of nearby neurons act in concert. This relatively novel method
of recording, called magnetoencephalography (MEG),
has already produced several pieces of new information
Rev. Mod. Phys., Vol. 65, No. 2, April 1993

P50m

P200m

p
0

~1
100

200 ms

FIG. 1. Typical magnetic response as a function of time, measured close to the subject's auditory cortex. The signal was
evoked by a 50-ms tone, illustrated by the hatched bar on the
time axis. 150 single responses were averaged. The signal consists of the following parts: a small deflection (P50m) at about
50 ms after the stimulus onset, a prominent peak (NIOOm) at
100 ms, and yet another peak (P200m) at about 200 ms. P indicates that the corresponding deflection in electric potential
(EEG) measurements is positive at the top of the head; N, that it
is negative; and m refers to "magnetic."

Hamalainen et a/.: Magnetoencephalography
the body, these studies are often called magnetic source
imaging (MSI), and sometimes magnetic-field tomography or current-flow imaging.
Present understanding of brain functions is based
mainly on experiments using small mammals. Spontaneous neural activity can be recorded by microelectrodes
inserted into well-defined brain areas. These sites may
also be stimulated selectively. The effects of brain damage on behavior and on the cerebral electrical activity of
the animal also provide important information. Although extremely useful, animal models can never substitute for studies on humans when functions characteristic
to man are of interest. H u m a n brain activity may be
recorded directly during surgery on patients suffering
from drug-resistant epilepsy or cerebral tumors. However, for ethical reasons, such studies are restricted to small
areas in the vicinity of the damaged tissue.
Exploration of the human brain is of the utmost intellectual interest: the whole humanity depends on our
minds. Although a great deal has been learned about
cerebral anatomy and physiology, the fundamental question of how the brain stores, retrieves, and processes information is still largely unknown. Neural computers,
which will handle information in a way similar to the
brain, are becoming useful tools of analysis and decision
making. Recent developments in this new field of computing (Kohonen, 1988, 1990) are thus an important
source of inspiration for studies of the functional principles applied by the human brain and vice versa. In addition, many interesting questions in basic and clinical neuroscience can be investigated by magnetoencephalography.
The main emphasis in M E G recordings so far has been
on basic research. Clinical applications are now in their
early stages of development. To understand the dynamics of brain activity, one can monitor the temporal and
spatial changes in the active cortical areas. It is also possible to study in detail the reactivity of various brain regions to changes in the stimulus sequence. Responses to
speech sounds form an especially interesting branch of
research. Language is a typically human function, and
the complete noninvasiveness of M E G allows studies of
brain mechanisms associated with speech production and
perception in healthy persons under a wide variety of
conditions.
Many important imaging methods of the human brain
are available today (Pechura and Martin, 1991). Anatomical structures can now be investigated precisely by
means of computer-assisted x-ray tomography (CAT; see
Lauterbur, 1973 and Hinz, 1988) and by magneticresonance imaging (MRI; see Damadian, 1972, Lauterbur, 1973, and Hinshaw and Lent, 1983). Both these
techniques provide high-quality but static pictures of living tissues. Functional information about the brain can
be obtained with single-photon-emission computed tomography (SPECT; Knoll, 1983) and with positronemission tomography (PET; see ter-Pogossian et al.,
1975, Kessler et a/., 1987, and Jaszczak, 1988). A newly
developed echo-planar technique allows functional imagRev. Mod. Phys., Vol. 65, No. 2, April 1993

415

ing with M R I as well, with one-second time resolution
(Belliveau et ah, 1991; Stehling et al., 1991; Ogawa
et al., 1992). All these methods permit studies of the
brain without opening the skull, but the subject is exposed to x rays, to radioactive tracers, or to time-varying
and strong static magnetic fields.
Electroencephalography (EEG), the measurement of
electric potential differences on the scalp, is a widely applied method of long clinical standing. M E G is closely
related to E E G . In both methods, the measured signals
are generated by the same synchronized neuronal activity
in the brain. The time resolution of M E G and E E G is in
the millisecond range, orders of magnitude better than in
the other methods mentioned above. Thus with M E G
and E E G it is possible to follow the rapid changes in
cortical activity that reflect ongoing signal processing in
the brain; the electrical events of single neurons typically
last from one to several tens of milliseconds.
A very important advantage of M E G and E E G is that
they are completely noninvasive. One only measures
brain activity as a result of sensory stimuli such as
sounds, touch, or light, or even when no stimuli are
given. The very rapid development of the M R I technique has provided an ideal medium for linking the functional M E G information to anatomical data. Several
groups are now working towards routine use of
magnetic-resonance images in conjunction with M E G .
The interested reader may wish to consult other review
articles or books. The September 1979 and 1992 issues of
Scientific American are devoted to brain research. The
books by Kuffler et al. (1984), Kandel and Schwartz
(1991), and Thompson (1985) are general introductions to
the human brain; the text by Thompson is, in particular,
to be recommended for physicists. The books edited by
Williamson et al. (1983), Grandori et al. (1990), and by
Sato (1990) are more specialized treatises on biomagnetism and magnetoencephalography. Review articles on
M E G have been written, for example, by Williamson and
Kaufman (1981, 1989), Romani et al. (1982), Hari and IImoniemi (1986), Romani and Narici (1986), Hoke (1988),
Hari and Lounasmaa (1989), and by Hari (1990, 1991).
Specialized information is available in the proceedings of
international conferences on biomagnetism, edited by
Erne, Hahlbohm, and Lubbig (1981), Romani and Williamson (1983), Weinberg et al. (1985), Atsumi et al.
(1988), Williamson et al. (1989), and by Hoke et al.
(1992).
Magnetocardiography (MCG) is methodologically
closely related to M E G : instead of the brain, the heart is
being investigated (Nenonen and Katila, 1991a, 1991b;
Nenonen et al., 1991). Much of the theory and practice
of M E G , discussed in Sees. I l l , IV, and V of this review,
applies to M C G as well. However, the volume conductor
geometry is much more complicated for the thorax than
for the head, whereas the brain is functionally far more
complicated than the heart. New uses of multi-SQUID
devices are to be expected now that more reliable and
higher-sensitivity sensors have become available, and at a
lower cost than before (Ryhanen et al., 1989). For exam-

No. so that it fits into the cranial cavity formed by the skull. For example. The human brain Figure 2 is a drawing of the human brain viewed from the left side.. which receives tactile stimuli from the skin. lips) or for which accurate control of movement is needed (fingers). Vol. Figure 4(a) illustrates a localized source. If the primary source and the surrounding conductivi- Rolandic fissure Motor cortex Somatosensory cortex Frontai lobe Parietal lobe frfTj" Occipital lobe Visual cortex Cerebellum Brainstem FIG.g. Most regions of the cortex have been mapped functionally. The area in the frontal lobe just anterior to the Rolandic fissure contains neurons concerned with integration of muscular activity: each site of the primary motor cortex (Ml) is involved in the movement of a specific part of the body. the returning volume currents. while the primary visual cortex (VI) is in the occipital lobe at the back of the head. large areas of cortex are devoted to those body parts which are most sensitive to touch (e. Human brain seen from the left side. and occipital.: Magnetoencephalography 416 pie. The primary auditory cortex (Al) is in the temporal lobe buried within the Sylvian fissure. This process is associated with a primary current source related to the movement of ions due to their chemical concentration gradients (see Sec. April 1993 . Tesche et ah. temporal. the cerebral cortex. 2. C. while the Sylvian fissure is almost horizontal. 1989) and magnetic monopole detectors (Cabrera.Hamalainen et a/. and the magnetic-field lines around the dipole. The magnetic field is generated by both the primary and the volume currents. parietal. are divided into lobes by two deep grooves. Rev. The cortex has a total surface area of about 2500 cm 2 . which is a 2 . 1983 and Ilmoniemi et aL. This so-called volume current completes the loop of ionic flow so that there is no buildup of charge. with some of the anatomical features identified. Phys.. the primary somatosensory cortex SI. passive ohmic currents are set up in the surrounding medium. The left and right halves. SI and M l on the left side of the brain monitor and control the right side of the body and vice versa. 1983) have been used for some time already. they respond in a more complex way to external stimuli. 2. 1982. In M E G one is usually concerned with the uppermost layer of the brain. Mod. There are four lobes in both halves of the cortex: frontal. folded in a complicated way. 65. geomagnetic instruments (for reviews. Figure 3 il- lustrates representations of the body surface on SI and M l (Penfield and Rasmussen. II). In addition. The Rolandic fissure runs down the side of both hemispheres. in turn. The brain consists of two hemispheres. approximated by a current dipole. B. separated by the longitudinal fissure. 1950). Some of the important structural landmarks and special areas of the cerebral cortex are indicated. is located posterior to the Rolandic fissure. Most of the remaining regions are known as association areas. see Clarke. Basics of magnetoencephalography A sensory stimulus initially activates a small portion of the cortex.4 m m thick sheet of gray tissue.

and [aw FIG. The net external field is then zero. 65. The two halves are separated by the longitudinal fissure. Mod. 5.: Magnetoencephalography < ^ 417 F. The picture illustrates a transsection of the brain in a plane parallel to the face [see Fig. (b) Example of the topographic field map calculated from the measured MEG signals. M E G measures mainly activity from the fissures of the cortex. Modified from Penfield and Rasmussen (1950). Fortunately. From noninvasive measurements of the MEG or EEG field distributions. Vol. Examples of volume currents (dashed curves) and magnetic-field lines B (solid curves) produced by the primary current are shown as well. the cerebrospinal fluid. which often simplifies interpretation of the data. The head was approximated with a fourcompartment sphere consisting of the brain. The locations of the left and right primary auditory cortices Al in the upper surface of the temporal lobe are shown as well. III. Schematic illustration of idealized magnetic-field and electric-potential patterns produced by a tangential dipole (white arrow). the resulting electric potential (EEG) and magnetic field (MEG) can be calculated from Maxwell's equations. It follows from the linearity of Maxwell's equations that once we possess the solution for the elementary current dipole. These "homunculi" are based on direct cortical stimulation of awake humans during brain surgery. Representations of different body parts on the somatosensory (SI) and motor (Ml) cortices. April 1993 Magnetic field Electric potential FIG. In certain finite conductor geometries the volume current causes an equal but opposite field to that generated by the primary current. the active area in the brain can be determined by a least-squares fit to the data. Rev. 13(a)]. only currents that have a component tangential to the surface of a spherically symmetric conductor produce a magnetic field outside.__ Lower lip I—Teeth. 2. The simple geometrical construction for locating the equivalent current dipole in the brain is also illustrated: the dipole is midway between the field extrema (compare with Fig. (a) Current dipole (large arrow) in a homogeneous conducting medium. No. which has been widened here for clarity.Hamalainen et a/. the fields of more complex (a) / N V » '«> sources can be obtained readily by superposition. Therefore. the skull. soma- 4 £ » _ ~' S ' / j \v FIG. Figure 5 illustrates the magnetic. 3. all primary sensory areas of the brain—auditory. and the scalp.. This forward problem will be discussed in Sec. radial sources are thus externally silent. .and electric-field patterns due to a current dipole in a spherical head model. Phys. ty distribution are known. 4. For example. gums. Magnetic field of a current dipole. 5).

if the . Significant magnetic noise is caused.D. Because of the poor electrical conductivity of bone. because the noise level is lower in them than in rf SQUIDs (Clarke. In interpreting M E G and E E G data. April 1993 o rf SQUID * o dc SQUID FIG. The current dipole is a popular source model in M E G research. schematically illustrated in Fig. Hermann von Helmholtz showed in 1853 that this problem has no unique solution. 6. 65. The magnetic field. For example. therefore. i. an axial first-order gradiometer consists of a pickup (lower) coil and a compensation coil. Detection of neuromagnetic fields Neuromagnetic signals are typically 5 0 . changing the impedance around (rf SQUID) or across (dc SQUID) the loop. The magnitude. The outcome of this procedure is commonly called the equivalent current dipole (ECD. The magnetic signals from the brain are extremely weak compared with ambient magnetic-field variations (see Fig. i.e. 6. The only detector that offers sufficient sensitivity for the measurement of these tiny fields is the SQUID (Lounasmaa. For a single current-dipole source. is 10 nA m. use source models. Mod. by fluctuations in the earth's geomagnetic field. The sensitivity of the SQUID measuring system to external magnetic noise is greatly reduced by the proper design of the flux transformer. Figure 4(b) illustrates a typical topographic field distribution. 1966. with two junctions. IV. Ryhanen et al. V. This is because electrical potentials measured on the scalp are often strongly influenced by various inhomogeneities in the head. We shall discuss the theory and practice of SQUIDs as applied to M E G measurements thoroughly in Sec. 0 7 fWb (see Sec. 1976). dc SQUIDs.e. The weak links of the SQUID alter the compensation of the external field by the circulating current.9 1982a). this is accomplished by supercurrents that flow along the surface of the ring so as to precisely cancel any deviations from the quantization condition..D). a device normally used for bringing the magnetic signal to the SQUID. such as current dipoles.5 0 0 fT. in contrast. The dipole is halfway between the field extrema. A typical strength of a dipole. and both methods have their own advantages. Clarke et al. or special estimation techniques to interpret the data (see Sec. the irregular currents in the skull and on the scalp are weak and can be ignored as contributors to the external magnetic field. one part in 109 or 108 of the earth's geomagnetic field. This system of coils is insensitive to spatially uniform changes in the background field. Therefore.G). 8). 2. by moving vehicles and elevators. which on the chest is two to three orders of magnitude larger than the signals from the brain outside the head. Romani et al.y 1989). No. Rev. and position of the source can be deduced unambiguously from the map (Williamson and Kaufman. D. is mainly produced by currents that flow in the microscopically relatively homogeneous intracranial space. Figure 5 shows that the M E G and E E G field distributions are mutually orthogonal. M E G has better spatial accuracy than E E G . Schematic illustration of an rf and a dc SQUID. a few millimeters under favorable conditions (see Sec. caused by the synchronous activity of probably tens of thousands of neurons.I). The electrical activity of the heart also generates a field. 1974. Quantum-mechanical phase coherence of charge carriers in a bulk superconductor gives rise to magnetic-flux quantization in a solid superconducting ring: the magnetic flux <P through the loop must be an integral multiple of the so-called flux quantum. <&0=h/2e = 2 .418 Hamalainen et a/. for example. IV. V. In determining the locations of source activity in the brain. television. In an external magnetic field. These weak links limit the flow of the supercurrent and are characterized by the maximum critical current Ic that can be sustained without loss of superconductivity. A schematic illustration of the usual experimental arrangement for M E G measurements is shown in Fig. direction.: Magnetoencephalography tosensory. by radio. The Josephson (1962) junctions are indicated by crosses. Vol.. interrupted by one or two Josephson (1962) junctions. making accurate determination of the activated area difficult. The SQUID is a superconducting ring.. and by the omnipresent power-line fields. provided that the dipolar assumption is valid. but responds to inhomogeneous changes.. and microwave transmitters. see Sec. Thus rejection of outside disturbances is of utmost importance. the source currents are limited to a small region of the brain. Data obtained by these two techniques complement each other. thus making it easier to exploit flux quantization for the measurement of the magnetic field. the map of the radial magnetic field Br has one maximum and one minimum.D). In practice the optimal solution is found by fitting the theoretical and measured field patterns in the leastsquares sense. 7(b)]. Phys. The magnetic flux <P threads the superconducting loop of the SQUID. with the deduction of the source currents responsible for the externally measured field. one is dealing with the electromagnetic inverse problem. and visual—are located within fissures. are preferred. This can be detected (dc SQUID) by feeding a current and measuring the voltage or (rf SQUID) by inductive coupling (see Sec. V. 1981. IV). One must. It is used to approximate the flow of electrical current in a small area.D. 7(a). 5). which are identical in area and connected in series but wound in opposition [see Fig. at a right angle to the line joining them (see Fig.

spontaneous brain activity. Processing of incoming information by the brain can be studied by recording responses to sensory stimuli. and three layers of aluminum. we use coordinate systems that are fixed to each sensor unit separately.1 1 10 100 1000 F r e q u e n c y (Hz) FIG.B.D. 9. Phys.>vQoise • Heart ( Q R S ^ 9 n e t l c • Eye (steady) ^ s ^ ^ a rhythm" • Evoked fields ^ c SQUID /V^Brain noise" Dewar noise 10° IO"2 0.l). In many experiments. The subject has her head inside a 122channel neuromagnetometer (see Fig. such as the a rhythm (see Sec. For still better rejection of 10 12 |x io 10 1 1 U — Earth's steady field r fe 108 \£ 10 6 CD "O g 1 4 8 102 ° C/D Lung \ . The shielding factor is 10 3 -10 4 for low-frequency magnetic fields and about 105 for fields of 10 Hz and above. Auditory stimuli are presented via plastic tubes and earpieces (not shown) from loudspeakers outside the shielded room. April 1993 FIG. or incoherent background events are sources of noise. During actual measurements the subject is usually left alone and the double doors are closed. (a) Typical experimental arrangement during a MEG measurement.001 Thermal noise of the body J_ 0. thus producing a net change in the output. 27). 65. The bottom of the helium dewar.. it will cause a much greater change of field in the pickup loop than in the more remote compensation coil.01 _L 0. which attenuate very well the high-frequency band. signal of interest arises near the lower coil. V.6). while the upper coil compensates for variations in the background field. Peak amplitudes (arrows) and spectral densities of fields due to typical biomagnetic and noise sources. with the flux-transformer pickup coils near its tip. The subject lies on a nonmagnetic bed with his head supported by a vacuum cast to prevent movements.c) Two superconducting flux transformers employed in brain research: the axial gradiometer (b) measures Ai?z/Az (approximately Bz at the lower loop) and the planar gradiometer (c) ABz/&x (approximately dBz/dx). eo ~ . which are effective for shielding at low frequencies of the external magnetic noise spectrum (particularly important for biomagnetic measurements). Flux transformer (b) detects two field extrema. Laboratory particles >. View of the interior of the magnetically shielded room at the Helsinki University of Technology (Kelha et al. For localization of the active brain area. 4(b) and 27]. ? N . while with (c) the maximum signal is recorded above the dipole (see Fig. Vol. In effect. 1982). (b. 2.. Rev. 8. Mod. V. in this way the evoked responses emerge from the background noise.: Magnetoencephalography external disturbances. The room is a cube of 2. Detection of cerebral magnetic fields. Thus signals resulting from many successive stimuli must be averaged.F. M E G measurements are usually performed in a magnetically shielded room (see Fig. the lower loop picks up the signal. 7. 9 and Sec. . measurements must be made over many sites. 41 and Sec.I). VI. No. typically at (b) Liquid helium (c) Flux transformers 419 3\ FIG. one on each side of the dipole [see Figs.Hamalainen et al. is brought as close to his head as possible.4-m inner dimensions with three layers of /x-metal. Note that when referring to individual magnetic sensors.

420

Hamalainen et a/.: Magnetoencephalography

20 to 60 points separated by about 3 cm. At each location the stimulus must be repeated 20 to 500 times in order to obtain an adequate signal-to-noise ratio. This is a
potentially tedious and time-consuming procedure that
also endangers the reliability of the data: the subject cannot be expected to remain in the same state of vigilance
throughout a long measurement session.
Indeed, the main drawback of the M E G method in the
past was the long time needed to gather data for a topographic field map [see Fig. 4(b)] from which the activated
cortical site could be deduced. An instrument that
records the whole magnetic-field pattern at once is clearly very useful. Several M E G groups and companies are
already constructing and using such multi-SQUID magnetometers. A photograph of the 24-channel neurogradiometer in Helsinki is shown in Fig. 10; this instrument
employs planar gradiometric flux transformers, whose
construction was schematically illustrated in Fig. 7(c).
An example of M E G data recorded with this system is
shown in Fig. 11. A typical auditory evoked response
was detected as a function of time over the left temple.
The main peak occurs approximately 100 ms after the

yx^^^^^w^
^v*^^,
9
0 100 ms

^^V V ^

10

jcc^^^r^^^

" 100 fT/cm

\xr*y\**&*

FIG. 11. Typical auditory responses evoked by short tones.
The approximate location of the magnetometer is depicted on
the schematic head. The magnetic-field gradient as a function
of time is illustrated at 12 locations (see Figs. 10 and 37) for two
orthogonal directions at each site, dBz/dy above and dBz/dx
below. The Cartesian coordinate system is individually determined at the site of each sensor unit [see Fig. 7(c)]. Two traces,
both corresponding to averages of 66 individual responses, are
superimposed in each case to show the repeatability of the
responses. The stimulus has been given at t==Q (time scale
below the data of unit 8). The arrow on the head shows the location of the equivalent current dipole in the auditory cortex
during the largest peak of the response.

stimulus onset. The locations of the largest signals imply
that the equivalent current dipole is beneath unit 8, in the
auditory cortex at the upper surface of the temporal lobe,
pointing down (see Sec. V.B.2).
E. Magnetic fields from the human body

FIG. 10. The 24-channel neurogradiometer in Helsinki (Kajola
et aL, 1989; Ahonen et ah, 1991) at a typical measurement distance above a model of the brain. During an actual measurement the instrument is, of course, in a helium dewar and the
brain is inside the skull! The measuring head of the gradiometer and the tip of the dewar are concave so that they fit as closely as possible to the subject's head.
Rev. Mod. Phys., Vol. 65, No. 2, April 1993

The strongest electrophysiological signals are generated by the heart and by skeletal muscles. Typically, the
amplitude of the QRS peak, corresponding to contraction
of the cardiac muscle, is several tens of picoteslas. However, the clinically more important signals from the cardiac conducting system, which are superimposed on the
field due to contraction of the muscle, are less than 1 pT
in amplitude (Fenici et al., 1983).
From the normal awake brain, the largest field intensity is due to spontaneous activity. The so-called a
rhythm, observed over the posterior parts of the head, is
1-2 pT in amplitude (see Sec. VI.D.l). Abnormal conditions, such as epileptic disorders, may elicit spontaneous
spikes of even larger amplitude (see Sec. VI.E). Typical
evoked fields following sensory stimulation are weaker by
an order of magnitude or more; i.e., their strengths are
only several tens or hundreds of femtoteslas. The ultimate limit for the sensitivity of a biomagnetometer is
determined by the thermally induced magnetic noise in
the body tissues. This background is, however, only on
the order of 0.1 f T / V H z (Varpula and Poutanen, 1984).

Hamaiainen et air. Magnetoencephalography
The brain itself produces magnetic fields that are often
irrelevant to the experiment being conducted (see Fig.
12). This background activity limits the signal-to-noise
ratio in measurements made with the best SQUID magnetometers. Of course, during studies of spontaneous
brain activity, this subject "noise" is actually the signal!
In contrast to instrumental noise, the background activity is correlated between the different magnetometer
channels and can, at least in principle, be taken into account in the data analysis. If the noise source is different
from the signal source, the two can be separated in multichannel recordings.
The spectral density of the cerebral background activity is typically 2 0 - 4 0 f T / v ' H z below about 20 Hz, decreasing towards higher frequencies and showing peaks
at the spontaneously occurring rhythms of the brain, e.g.,

around 10 H z (Maniewski et aL, 1985; Knuutila and
Hamaiainen, 1988).
There are many other magnetic signals originating
from the human body (see Fig. 8). Ionic currents in the
eye give rise to a static 10-pT field, which also has a
changing component during eye movements and blinks.
The fields due to inhaled or digested magnetized particles, often observed in the lungs of welders or iron foundry workers, may be several nanoteslas in amplitude.
II. NEURAL BASIS OF MAGNETOENCEPHALOGRAPHY
In this section, the structure of the human cerebral
cortex and some neuronal functions are described in such
detail that the reader may gain an understanding of the
brain's electrical activity and the ensuing electromagnetic
field.
A. Constituents of the brain

(a)

0

20

40
60
Frequency (Hz)

80

100

0

20

40
60
Frequency (Hz)

80

100

FIG. 12. Background activity of the brain over the left temporal lobe for a subject with his eyes open; an axial first-order
gradiometer was used: (a) spectral density of the noise spectrum with (upper curve) and without (lower curve) a subject; (b)
normalized coherence functions measured between channels
separated by 36.5 mm (solid curve) and 73 mm (dashed line).
The dot-dashed curve shows the coherence between channels
without a subject. The prominent peak at 50 Hz is caused by
the power-line frequency. From Knuutila and Hamaiainen
(1988).
Rev. Mod. Phys., Vol. 65, No. 2, April 1993

421

The principal building blocks of the brain are neurons
and glial cells, the latter being more abundant by a factor
of 10. The glia are important for structural support, for
the maintenance of proper concentrations of ions, and for
the transport of nutrients and other substances between
blood vessels and brain tissue. Specialized glial cells, the
oligodendrocytes, form an insulating myelin sheath
around certain nerve fibers, thereby speeding signal
transmission. Neurons are the information-processing
units. Their cell bodies and dendrites are concentrated in
the gray matter, the largest part of which is the cerebral
cortex forming the surface of the brain. In addition,
there is gray matter in subcortical nuclei, like the
thalamus, but these structures do not directly interest us
here because M E G signals are mainly due to cortical
currents.
The interior of the brain is largely occupied by nerve
fibers; this tissue is called white matter because of the
bright appearance of the myelinated axons (see Sec. II.B),
which form connections between different cortical areas
as well as from the cortex to other brain structures and to
the periphery. There are also many links between the
two hemispheres, the most conspicuous of them being the
corpus callosum.
Anatomy has implications for neuromagnetism in
three ways. First, local, cellular-level structures determine how neuronal electrical events produce macroscopic current sources. Second, the electrical conductivity as
a function of location on a macroscopic length scale is
needed for the solution of the forward problem (see Sec.
III.E.2). Third, knowledge of the brain's anatomy can be
utilized in the solution of the inverse problem (see Sec.
IV.E). In particular, if we assume that the source
currents are confined to the gray matter, the threedimensional localization problem becomes essentially
two-dimensional, thereby improving the accuracy. In
practice, the distribution of gray matter in the brain, as
well as the shape of the cranium, can be determined with

422

Hamalainen et a/.: Magnetoencephalography

FIG. 13. Coronal section (a) at the level of the ears and sagittal (longitudinal) section (b) near the head midline of MRI's made on one
of the authors; many important brain structures are clearly seen. The images were acquired with the Siemens 1.0-T MRI system at
the Helsinki University Central Hospital.

magnetic-resonance imaging. Two magnetic-resonance
(MR) images of a human head are illustrated in Fig. 13.
The detailed structure of the head is quite complicated.
The brain is surrounded by membranes which, although
thin, may have to be taken into account when precise
analysis of the cerebral magnetic field is attempted. In
addition, the electrical conductivity of brain tissue is
highly anisotropic: the white matter conducts 10 times
better along an axon fiber than in the transverse direction. At the present level of analysis, we largely ignore
these complications; in fact, we usually consider the
whole intracranial volume as a homogeneous conductor.
Refinement of the volume conductor models presented in
this review will be useful only when there are tools to
measure the details of conductivity and to take them
properly into account in the analysis.
B. The neuron
Figure 14 shows the visual cortex of a rat in cross section. Only a small percentage of the neurons is seen because a selective stain has been applied to the tissue.
Neurons can send electrical impulses, so-called action potentials, to other neurons nearby or to distant parts of the
brain (see Sec. II.E). A neuron (see Fig. 15) consists of
the cell body (the soma), which contains the nucleus and
much of the metabolic machinery, the dendrites, which
are threadlike extensions that receive stimuli from other
cells, and the axon, a single long fiber that carries the
nerve impulse away from the soma to other cells. The
two principal groups of cortical neurons are the pyramiRev. Mod. Phys., Vol. 65, No. 2, April 1993

FIG. 14. Transverse section through the gray matter of rat cortex; the thickness of the sheet, from top to bottom, is in humans
about 3 mm. A prominent pyramidal neuron (H) is illustrated
in the middle and two stellate cells (F) nearby. The tissue was
prepared by the Golgi method, which stains about 1 % of the
neurons. This picture was drawn by Ramon y Cajal in 1888.

K + .Hamalainen et a/. inhibitory synapses often attach to the soma. Phys. see Sec. The intracellular potential is increased by input through the excitatory synapses. No. Vol. Rev. redrawn from Ramon y Cajal. The neuron fires an action potential when the potential at the axon hillock reaches a certain threshold level. but decreased by inhibitory input. like other cells. The dendrites and the soma have typically thousands of synapses (connections) from other neurons. The channels are ion selective. The insulating membrane divides the tissue into intracellular and extracellular compartments with different ion concentrations. These molecules diffuse quickly through the gap and some of them attach themselves to receptors on the surface of the postsynaptic cell.D and Fig. Cortical neuron. II. This difference is maintained by special protein molecules on the membrane that pump selected ions against the concentration gradient and also serve as passive ion channels. so that they tend to be perpendicular to the cortical surface. depending on the receptor that is activated. April 1993 . 15. Most excitatory synapses are on the dendrites. the intra- Dendrites . 14). 15(a). The most important is the N a . Schematic illustrations of three synapses are shown in Fig. 2. C. the increase of the resting voltage due to this current is only a few mV.: Magnetoencephalography dal and the stellate cells (see Fig. is surrounded by a membrane. this is important for M E G studies. 1979). and Cl~ ions) changes the membrane potential in the second cell. 65. When a pulse arrives along the axon of the presynaptic cell. special transmitter molecules are liberated from the synaptic vesicles into the 50-nm-wide synaptic cleft. (b) pyramidal neuron.Axon hillock Axon FIG. opening ion channels 423 through the membrane. Ion mechanisms The neuron. the resultant direction of the electrical current flowing in the dendrites is also perpendicular to the cortical sheet of gray matter.K pump. Mod. In a quiescent neuron. The former are relatively large. Since neurons guide the current flow. 16(a)]. only certain types of ions may pass through the membrane. which moves three Na"1" ions out and two K + ions into the cell in one duty cycle. (a) Schematic illustration of a pyramidal neuron and three magnified synapses (modified from Iversen. Although the pump gives rise to a current through the membrane. their apical dendrites from above reach out parallel to each other. As a result. a 10-nm-thick liquid-crystal bilayer of phospholipids. The ensuing flow of charge (mainly N a + . this event is called the postsynaptic potential [PSP. the receptor molecules change their shape..

but not in amplitude. The resulting current flows into the cell if sodium channels open. Mod. Vin and VQxt are the intracellular and extracellular potentials.7 0 mV . d = 1 /xm. (2). April 1993 c r i n = l £l~lm~\ and A F = 25 mV. and e denotes the electron charge. Since rs =4/(Trd2crin). the Nernst equation is obtained: kBT V=Vit In (1) Here. about a million synapses must be simultaneously active during a typical evoked response. The current / through the synapse can be calculated from the change of voltage AV during a PSP: I =AV/(Xrs). the permeability of the membrane for specific ions is altered and the potential in the vicinity of the membrane changes. and F ( N a + ) = + 5 2 mV. with strength Q =IX. Action potentials increase in frequency. kB T is the thermal energy.and extracellular concentrations of C l are about 20 and 120 mmol/liter. From a distance. In the inactive state. Because the permeabilities P ( K + ) and P(C1~) are large. Postsynaptic potentials When transmitter molecules released to the synaptic cleft (see Fig..4 8 mV.: Magnetoencephalography 424 cellular concentrations of N a + and K + are about 20 and 140 mmol/liter. with the intensity of the stimulus.e. Rev. respectively. When the permeabilities change as a result of synaptic excitation or inhibition. we find that Q « 20 fA m for a single PSP. Inserting typical values. in the latter case.. When these three most important ion types are simultaneously taken into account. the resting potential is between the equilibrium values for K + and Cl~~. The intra. Usually. in addition to the pump currents. F ( C l . i.) = . whereas the corresponding extracellular concentrations are about 140 and 5 m m o l / liter. From this. The equilibrium voltage across the cell membrane is determined by the requirement that the diffusive and ohmic currents. 65. respectively. the voltage over the membrane becomes J knT -In ^ ( K ^ ^ K ^ ^ N ^ ^ N a ^ ^ ^ ^ c r = . Schematic illustrations of (a) a postsynaptic potential and (b) an action potential as a function time. the simultaneous activation of as (a) Action Postsynaptic potential 100 mV 10 mV 1 ms 10 ms (d) (c) 40 > 40 > 0 E E -40 -80 L 0 -40 -80 0 200 Time (ms) L 0 200 Time (ms) FIG. No. In the former case. (1) and (2)]. respectively (Scott. the cell is hyperpolarized and thereby inhibited.Na-+i>(Cr)Cext.K++P(Na+)Cin.F ext [see Eqs. Phys. where d is the diameter of the dendrite and crin is the intracellular conductivity. 2. 16(a)]. In a cortical neuron. The strength of the current source decreases with the distance from the synapse. the diffusion coefficient P ( K + ) for K + through the membrane is much higher than P ( N a + ) .8 9 mV. which is stronger in (d) than in (c). The equilibrium potential is determined by thermodynamics: the concentration C of each ion type tends toward thermal equilibrium between the two compartments according to the Boltzmann expression C~exp(-\e\V/kBT) . Q =ird2(TinAV/4. X is typically 0.2 mm. the current-dipole moments required to explain the measured magnetic-field strengths outside the head are on the order of 10 n A m . the permeability of potassium dominates that of sodium. When K + ions diffuse out of the cell. the PSP looks like a current dipole oriented along the dendrite. Voltage V = Vm . or in the course of an action potential. we obtain the equilibrium membrane voltages (inside with respect to the outside) at the body temperature T = 310 K: F ( K + ) = .Hamalainen et a/. (2) ^(K+)Cin. 16. The length constant of the exponential decay is X = (gmrs ) _ 1 / 2 . be balanced for each type of ion. Applying the Nernst equation to concentrations mentioned above. 1977). . This causes an electric field and a current along the interior of the postsynaptic cell. where gm and rs are the conductance of the membrane and the resistance of the intracellular fluid per unit length. D. the interior becomes more negative. Vol. it flows out if potassium or chloride channels are activated.1-0. Since there are approximately 105 pyramidal cells per mm 2 of cortex and thousands of synapses per neuron.Cr This is Goldman's (1943) equation. the voltage across the membrane follows Eq. the cell is depolarized and there is an excitatory PSP [see Fig. 15) arrive at the postsynaptic cell. Therefore.

the area would be 5 times larger (Hari. been detected both electrically and magnetically in peripheral tissue. if the relevant distance is taken to be the length constant X = 0 . Wikswo and Roth. From these two sets of data. This change of potential triggers the neighboring region. Since the two dipoles are opposite. 1975. NEURAL GENERATION OF ELECTROMAGNETIC FIELDS A. First. During this stage the interior of the cell is positive for a short time [see Fig. No. Henry. When such a voltage pulse travels along an axon. there is a solitary wave of depolarization followed by a repolarization of longer spatial extent. than for action potentials. which last tens of milliseconds. with opposite polarities on the two sides of the nerve. 1984). On the other hand.C]. but this has not been demonstrated so far. temporal summation of currents flowing in neighboring fibers is more effective for synaptic currents. Thus E E G and M E G signals are produced in large part by synaptic current flow. (1989). they form a current quadrupole. Compound action fields after electric stimulation of the right median nerve at the wrist. The measured transmembrane potential closely resembled that calculated from the observed magnetic field. The membrane potential provides the energy needed to propagate the unattenuated pulse. When action potentials were triggered in the nerve. Kraut et al. The response is monophasic at most locations. 1989). Action potentials Signal transfer along an axon is based on the ability of the membrane to alter its permeability to N a + and K + ions..Hamalainen et al. Erne et al. Furthermore. Figure 17 shows that the noninvasively recorded CAFs near the elbow peak at 6 . 1985). A dipolar field. 16(c)]. III. Assuming this estimate over an effective thickness of 1 mm. respectively.. This is also illustrated by a more realistic estimate based on measured current densities. The measurements were made twice (two superimposed traces) at both sites with a 7channel gradiometer. When the excitatory input becomes stronger. produced by synaptic current flow. II. et al. 1985). a dipole moment of 10 n A m would correspond to 40 mm 2 of active cortex (Chapman et al. the action potential can be approximated by two oppositely oriented current dipoles. 1988. et al. 15(a)] reaches the firing threshold of about —40 mV. .7 ms after median nerve stimulation at the wrist. it was possible to determine the intracellular conductivity. decreases with distance as 1/r 2 . 1985. Wikswo. When the conductivity a and the electric current generators in the brain are known. Mod. E. 2 mm (Kuffler et al. Later. Abraham. and Hentz. J and p are the total current density and the charge density. but the frequency of firing increases [see Figs. 2. the amplitude of the action potential remains the same. Over straight portions of a nerve fiber of uniform thickness. Vol. Quasistatic approximation of Maxwell's equations The previous section described structural details of the brain as well as neuronal electrical activity. and about 1000 single responses were averaged. more slowly than the l/r 3 -dependent quadrupolar field. Hari. The course of the median nerve along the arm is shown schematically. the permeabil- FIG. a biphasic magnetic signal of about 1 ms duration was detected. The change is due to the opening of voltagesensitive channels as a result of an approaching action potential [see Fig 16(b) and Sec. Action potentials have. the action potential thus travels along the axon with undiminished amplitude like the wave of falling pieces in a chain of dominoes. Phys. activation of larger areas is necessary because there is partial cancellation of the generated electromagnetic fields owing to source currents flowing in opposite directions in neighboring cortical regions. 1984). They also might contribute significantly to high-frequency electromagnetic fields outside the skull. An action potential is initiated when the voltage at the axon hillock [see Fig. 65.. Rev. In practice. however. 17. the magnitude of each dipole is about 100 fAm. April 1993 425 Wikswo et al. They used the sciatic nerve in the hip of a frog. 1985. et al. 1985..: Magnetoencephalography few as one synapse in a thousand over an area of one square millimeter would suffice to produce a detectable signal. the fiber was threaded through a toroid in a saline bath.J = — dp/dt can be used to calculate the electric field E and the magnetic field B. Their separation depends on the conduction velocity v and is about 1 mm in an unmyelinated cortical axon where v^l m / s . Hallstrom. which have a duration of 1 ms only. We can simplify the treatment of Maxwell's equations from the outset by noting two facts. Maxwell's equations and the continuity equation V .. Recently it has become possible to record magnetic compound action fields (CAFs) associated with electrical impulses propagating along human peripheral nerves (Wikswo... Hallstrom. 1985. Wikswo. 16(c) and 16(d)]. which is approximately dipolar. the latency becoming longer as the measurement site is moved closer to the central nervous system (CNS). From Hari. the magnetic field of an action potential propagating in a single giant crayfish axon was recorded as well (Roth and Wikswo. (1980) reported the first measurements of the magnetic field of a peripheral nerve. 100-250 n A / m m 2 (Freeman. 1990).

r e ) . If the events are considered on the cellular level. a current dipole is used as an equivalent source for the unidirectional primary current that may extend over several square centimeters of cortex. We start from Maxwell's equations: V-E=p/e0 . cellular electrical contain mostly frequencies below 1 kHz. Vol. In addition. Here cr(r) is the macroscopic conductivity. we shall employ it extensively.: Magnetoencephalography 426 ity of tissue in the head is that of free space. (11) This definition would be meaningless without reference to the length scale. 65. the quasistatic approximation appears justified.e. (11) is illustrative in that neural activity gives rise to primary current mainly inside or in the vicinity of a cell. we can adopt the quasistatic approximation. and / . B. electric field can be expressed with a scalar potential. II. 2. it is sometimes convenient to consider Q as a line element of current / pumped from a sink at r x to a source at r 2 . April 1993 A current dipole Q. In particular. the membranes. guide the flow of both intracellular and extracellular currents. need not be taken into account in the calculation of B.Hamalainen et a/. the symbol for a charge dipole. X c = 6 5 m. whereas the volume current flows passively everywhere in the medium. The letter Q is used here to avoid confusion with p . (6) and (7). the primary current is largely determined by the cellular-level details of conductivity.. e is tivity of the material. J = <jE+aP/3f . although the conversion of chemical gradients to current is due to diffusion. being good electrical insulators. J y (r) = <r(r)E(r). viz. i. Everything else is the primary current 3P: J(r) = F ( r ) + (7(r)E(r) = F ( r ) . we find 2irfe/a = 2X 10~ 3 « 1. it is customary to speak about the impressed rather than the primary current. VXVXE=~v(VXB) 3* C.e. we generally deal with that are below 100 Hz. Therefore. In neuromagnetism.. B=-VV . e ^ l O ^ Q .E). Primary current It is useful to divide the current density J ( r ) produced by neuronal activity into two components. With cr = 0 . i. i. which means that in the calculation of E and B. (9) Solutions of this equation have spatial changes on the characteristic length scale kc = 7rffi0a(l +j2Trfe/a)|~1/2 . The division in Eq. For example. is a widely used concept in neuromagnetism. (6) In a passive nonmagnetic medium. By finding the primary current. Q at TQ can be thought of as a concentration of J ^ r ) to a single point: J'(r) = Q 8 ( r . Second. The volume or return current. is passive. this so-called displacement current. Current dipole = . In other words. Let us first see why this is so. VXB=/i0[aE + (e-60)3E/3r+603E/3f] . dB/dt must be small. J is the sum of ohmic volume current and the polarization current. the whole cortex is modeled as a homogeneous conductor. From Eqs. JU=JU 0 . Let uniform and let us consider electromagnetic at frequency / (j = V — 1): the permitfrequencies phenomena a and e be phenomena E = E 0 (r)exp(y27r/Y) . since V X E = 0.100 Hz. (8) With Eqs.. (5) VXB=/x0(J + 603E/ar) . Mod. Phys. much longer than the diameter of the head. With the above parameters.e. Nevertheless.a ( r ) V F ( r ) . (7) where P = (e — e 0 )E is the polarization. Although the current dipole is usually a model for a layer of primary current. This implies that the contribution of 3B/3f to E is small. It should be emphasized that Jp is to be considered the driving "battery" in the macroscopic conductor. (4) and (6). 3 ft-1 m " 1 (value for brain tissue). the Rev. we locate the source of brain activity. This does not mean that we should forget time-dependent phenomena altogether. For the quasistatic approximation to be valid. (3) VXE=-3B/3r . approximating a localized primary current. It is the result of the macroscopic electric field on charge carriers in the conducting medium.M o l T ( ^ E + e3E/3f) ot = -j2Trfti0(a+j2Trf€)E . In the quasistatic approximation... (12) where S(r) is the Dirac delta function. No. (10) The use of V considerably simplifies derivations of formulas for electromagnetic fields. 2rrfe/a « 1 . . In E E G and M E G applications. e 0 3E/3£. cellular-level details are left without explicit attention. 3 E / 3 f and dB/dt can be ignored as source terms. it is necessary that the time-derivative terms be small compared to ohmic current: | e 3 E / 3 ? | « | c r E | . (4) V-B=0 . the capacitive current through the cell membrane is significant in determining the properties of the action potential (see Sec.

1983): J 0 = fjp(r)dv. crE is replaced by an equivalent fictitious current VVa.a -a VFandVX(CTVF) = VaXVF. R = r — r'. . The unit vector normal to the surface Sy at r' from region i to j is denoted b y n ^ r ' M s e e F i g . in an infinite homogeneous conductor the magnetic field is obtained directly from Jp. J 0 = Q.e.^ . D. in general. (13). the second is due to Jv. (11). has no direct physical meaning. Since VaXVV = — V X ( F V a ) . (16) Because the source of the magnetic field is the total current J. AIT 47T j% m where and J X V ' ( l / 2 0 = ( V ' X J ) / i * .L a p l a c e law. Integral equations for Vand B -dv' where r is the point where the field is computed.V ' X ( J / i * ) into Eq. their boundaries by dGi9 their conductivities by at. .. V a is nonzero only at the boundaries. R R/R3=-V(l/R) With proper boundary conditions. a comparison of Eqs. For the primary current distribution (12). 18). i. (6). . Q is closely related to the dipole term J 0 in the current multipole expansion (Katila. in Eq. (14). Taking the divergence of Eq. (17) From Eq. both Jp and J E contribute. we find exd/dx+eyd/dy+e2d/dz V' = exd/dx'+eyd/dy' + ezd/dz' . (11) into Eq. (13) and transforming the volume integral of V ' X ( J / R ) into a surface integral.Hamalainen et a/.F . and (15) shows that B(r) = . Once V is known. and the surfaces between Gt and Gj by Stj. E. FIG. / = 1. Mo dv'-f /in r B 0u(r) = -7 5LJJ ATT (19) 1 = 1 rV'aXV'F dv' R G R Jp(r')X-^rdv' R3 (20) is the magnetic field produced by Jp alone. Inserting Eq.J[(Jp+VV'a)X~dvf 4TT R3 .. and it is possible to write the second term of Eq. Q = / ( r 2 —r 2 ). (16). *'>-£/ V . Each region Gt has uniform conductivity o> Unit vectors normal to the surfaces are denoted by n^.P(r') within the brain. We notice that there is no contribution from Jv if V'cr = 0. we obtain V-J/>=V-J-hV-(CTVF) . we find for a current density that approaches zero sufficiently fast when r' goes to infinity B(r) Mo B(r) = B 0 (r) and 4ir (18) 1. Inserting the identities V= 427 VX(KV(1/J?)) = VFXV(1/JR) and (15) f VXudv= f dSXu . m . Piecewise homogeneous conductor (13) = V'(l/R) .. where If the conductor is assumed to consist of homogeneous parts. (16). the continuous counterpart of the Biot-Savart law which applies to closed wire loops. The first term in this equation is the direct contribution of Jp to the magnetic field. and the primed symbols refer to quantities in the source region. . However.J = 0 in the quasistatic approximation. 18. J R V ' X J ' V ) _. since the divergence of a curl vanishes. Integral formulas for l/and B The forward problem in neuromagnetism is to calculate the magnetic field B(r) outside the head from a given primary current distribution . which. Multicompartment conductivity model. Thus we obtain V-(aVF) = V . (16) as a sum of surface integrals over boundaries. The regions of different conductivity will be denoted by Gt. .: Magnetoencephalography then. In the quasistatic approximation. This is the A m p e r e . With the vector identities (14) -dv' W i t h J = F . (13). this equation can be solved for V either analytically in special cases or numerically with finite element techniques. it is straightforward to compute B from Eq..

we must calculate V on S. 4TT J A-TTCTQ where d S ' = n(r')dS". April 1993 (26) Furthermore. Note that datd') r-r '1-3 (r-r')-dS. the field B outside can be obtained without explicit reference to the volume currents. /=i ±V>2F_FV. Vol... (18) directly. in fact. viz.y at r \ 2. (<f>V2if>-if. B can be derived from the magnetic scalar potential: .means that the gradient at the boundary is taken inside region /. No. It is now evident that in the calculation of the magnetic field. (18) to obtain the result (Geselowitz.. (22) the contribution of volume currents to the radial field component R J R •dS. the volume currents can be replaced by an equivalent surface current distribution B J R ^hp(T')X^-erdv' (31) R If the source is a current dipole Q at r ^ .. the unit conductivity a0=l/(flm) is needed to get the dimensions right. 1967) V(r)a(r)= tfr=B(r)-er Mo r Br = 4-7T r On the left-hand side we make use of V' 2 (27) dv' we obtain an integral equation for V(r): -(c7. Of course. Making use of the limiting value (Vladimirov.)F(r')n/. viz. they are only a mathematical tool for simplifying the solution of the forward problem. since V X B = 0 outside the conductor. In order to compute B from Eq.-a. Since the current density crVV-n is continuous across Sy.V2<t>)dv = f G (^V^-^V^)-dS we find lim f V(r')-^ r n -*re5„ Tn-T' --dS. but. Phys. Br can be calculated simply from Eq. (13) with J replaced by lp. 2. is the solid angle subtended at r by the surface element dS. Mod. (23) where V. (26) approach a point on Stj.7 rn — r = -2irF(r)+/ Hr'^-dS^ . .-(r') / R In order to obtain the integral equation for V on the surfaces SiJy we let r in Eq. Spherically symmetric conductor 2 "</. but it turns out that one can derive a surface integral equation for the electric potential which involves V on the boundaries only. V0(T) is the potential due to Jp in an infinite homogeneous medium with unit conductivity.y. Rev. since (r-r')Xn(r')-er = (r-r')X- -4ir8(R) (25) =0 (30) Therefore. Combining the terms from all Sij9 we arrive at (Geselowitz. R R J a^ (21) R F0(r)=-—i— f J'(r')-V'4-«fo' R B(r) = B0(T) + ^2. ATT and of Eq.'(*i-aj)fSiV(r')^Xd$'ij . By applying Green's second identity.2JL R R l R ~ cr i R dv' If the conductor is spherically symmetric. These are often called secondary currents. (31) reduces to B = Mo Q X r o * e r (32) <T0V0(T) — rK-^iLwi^-ds. we find f V'VX^rdv'= * JGt f V'VXV'-^rdv' JG. b ij (28) R (ai+aj)V(r)=2a0V0(r) 2TT (29) where r€ES i . (23) reduces to (24) ij 'J R = B(r)T/|r| vanishes.. 1971) (22) where the prime signifies summation over all boundaries.Hamalainen et a/. 65. J G on each interface 5 / y . we first note that in Eq.: Magnetoencephalography 428 Here.. 1970) 1 = G VV± '-F r ATTOC. the right-hand side of Eq. we could start with Eq. (22). Eq. n ( r ' ) is a unit vector perpendicular to 3GZ and pointing out from Gt. To show this for spherical shells..j.

V will be attenuated on the scalp and the potential pattern due to a dipole inside will be more widespread than in a homogeneous sphere. Vk A k (r'WS/. April 1993 f A v0(r')ds. (29) over one of them.r ) and 1 8k = V F ( r . Integrating.9Aln_ (Barnard et al.e z 4TT |r-rQ|3 (35) Therefore. This means that the method is optimal for detecting activity in fissures. The result from the latter publication is B(r) = .W . Since on a closed triangulated surface with nt triangles we have «/ = « / / 2 + 2 vertices. Derivations of B. This is true for any axially symmetric current in an axially symmetric conductor (Grynszpan and Geselowitz.: Magnetoencephalography B = — fi0VU. the number of equations in (36) drops to about one-half and the computations become. (36) where 7=1 The column vectors gl and the matrices H' 7 are in this constant-potential approximation defined by F(r. No.9 1991). 3. each surface Si9 here indexed with a single subscript for simplicity. B must be computed numerically. (1985) and by Sarvas (1987). Alternatively. 1983).re)2 2. In practice. 1990.. (33) This shows that even though the tangential field components are affected by the volume currents in a spherically symmetric conductor. significantly faster. and r = | r | . Furthermore. H$ are often approximated by replacing ftA. Rev. were presented by Ilmoniemi et al. based on Eqs. Since Br=Q9 U = 0 and thus also B = 0. Its computation in general requires the evaluation of a series expansion in spherical harmonics (Arthur and Geselowitz. If the primary current is radial.(r) on each A^ by its value at the centroid clk of the triangle. in rectangular coordinates using Eq. we find 1 U(T)~- — f °° Br{tr)dt . An important point is seen from Eq. (38) l respectively. van Oosterom and Strackee. 1973). a set of nt linear equations is obtained for Vlk at triangles belonging to surface /.. Vol. 2. On a sphere. V(T)= Vlk. Note that the properties of B cited above for the spherically symmetric conductor are also valid for a horizontally layered conductor [a = a{z)]9 since this is a limiting case of a sphere. 1968. The electric scalar potential V on the surface of a spherically symmetric conductor is affected by the conductivity profile. viz. The exact gains in accuracy and speed are yet to be determined. 1973). O '+0-. Horacek. Thus Bz generated by a current dipole located within a horizontally layered conductor is given by 2 A4 a HlJ. if a poorly conducting layer (skull) is placed between two good conductors (brain and scalp). Lynn and Timlake. is tessellated into nt suitable triangles. (22) and (29). Since V-B = 0. 1967. r e ) = (r~ 1 fl 2 + fl"1aT + 2a + 2r)r — (a + 2r +a~~l2L't)TQ . the potential is now continuous over St.Hamalainen et a/. one can use a numerical integration formula to improve the accuracy (Nenonen et al. A source in the center of the sphere will produce no magnetic field outside: in Eq. The linear-potential approach highlights two different . the normal derivative dU/dr = —Br//j. (Barnard et aL. 1970. therefore. (33). Br = 0 if tQ = 0..(r') is the solid angle subtended by Aj at r . nl-. Here a J and af are the conductivities inside and outside surface Si9 respectively. 65.= 2TT J ~at . but we expect that the computational burden can be significantly reduced while retaining at least the same level of precision. .0. a = | a | . de Munck.7 1 — 4TT 8 ^ F(r.f r 2 — r f i . In the boundary element method (BEM). on each triangle A^ and integrating Eq.. Another difference from B is that both radial and deep sources contribute significantly to V. \xlk is the area of the /cth triangle A^ on St. they can be computed without knowing the conductivity profile cr = <j(r). The matrix elements H$ depend only on the conductor geometry and thus only the source term glk needs to be recalculated for each source configuration. we can designate the potential values at the vertices of the tessellated surface as unknowns in the discretization (Urankar. Phys.. 1967. (32). Br vanishes. (31). JP(T')=JP(T')T'/\T. U is harmonic and uniquely determined by its normal derivative on the surface of the conductor and by the requirement that it vanish at infinity. . For example.r g ) . A . k = 1. If the potential is allowed to vary linearly within A^. M E G is sensitive only to the tangential component of the primary current. Mod. ^ Mo Q X ( r .. .TQ) = a(ra . which is more time consuming than the computation of B from the simple analytical expressions. Realistically shaped conductor models If the regions Gt of the piecewise homogeneous conductor are arbitrarily shaped. de Munck. and ftA. Assuming further that the potential is constant.. in a spherically symmetric conductor. 1992). (34) 429 1988). The accuracy and speed of BEM can be improved by using higher-order approximations for V(r) on each A^. °i + * /T (37) : and with a = (r — T Q ) .

G R (42) . we suffer from the errors caused by replacing the original smooth surface by a polyhedron. (26) will vanish. modeling t h e head by a poor conductor (the skull) between two much better conductors (the brain tissue and the scalp) requires modifications in the standard numerical approach in order to produce accurate results. resulting in considerable savings of computing time if the potential of several different sources must be evaluated. The resulting equations can be solved by an iterative method such as the Gauss-Seidel algorithm. Many of our results are. and (2) solution of the integral equations for W(v). and by Silvey (1978). The neuromagnetic inverse problem is to estimate the cerebral current sources underlying a measured distribution of the magnetic field. Phys. we shall have to leave enough degrees of freedom for the potential on each surface element t o reproduce the high spatial frequencies present in the fields of nearby sources. B can be obtained from Eq. A (40) If there is only a single interface. and estimation theory.V ' 4 " d S ' ATT s R v-vf4-<fo'=o. It was shown by Helmholtz (1853) that a current distribution inside a conductor G cannot be retrieved uniquely from knowledge of t h e electromagnetic field outside. 65. V0 in Eq. meaning that the HiJ are replaced by a^H^'-j-e^J where e. Once V on each St is available. General (39) y conductivity approaches zero. the solution of Eq. giving B(r) = -^-Jp Jf n ( r ' ) X V ' . numerical mathematics. which implies that Eq. However. which imposes constraints on the sources that our measurements might reveal. i.y 1967). Vol. (22) by direct integration: B(r) = B 0 (r) + ^ . valid for other applications as well. although we shall keep constantly in mind that neuromagnetic fields arise in t h e brain. The electric potential is defined only up to an additive constant. if Jp is solenoidal ( V . or both. it is sufficient to use a brain-shaped homogeneous conductor in the analysis of M E G data. Even with a more complex potential variation on A^. (20).\ Magnetoencephalography 430 aspects in the calculation of Vlk. Second. is a unit vector with ni components and T denotes t h e transpose. et aL. the isolated problem yields accurate results. since K 0 ( r ) = ^ J/ n < r ' ) . which relate the potential values at adjacent elements on the surface Sj (Heller.. when one is dealing with a bounded homogeneous conductor. ^ir$ (4i) Air J G R and thus V=0. 1990). The first suggested solution for this numerical instability was the use of the Richardson extrapolation (Richardson and Guant.e. With this technique. (18). Meijs. Hamalainen and Sarvas (1989) used t h e isolated-problem approach. V=0 as a consequence of Eq. the linear equations can often be solved explicitly. electrically silent (E = 0 outside G). Simulations presented by Hamalainen and Sarvas (1989) show that. Meijs and Peters (1987) found that numerical errors in computing V may be excessive because the ratio in the conductivity of the skull to that of the brain and t h e scalp is less than 0. This approach relies on the assumption that the correct potential distribution can be obtained by estimating an asymptote on the basis of two solutions with different grid densities.1. £ {aj-af) ™ j = \ \ k=l f V(r')fjXdS'j A k . however.= o . we must have an accurate enough description of the geometry. (36) has no unique solution. However. April 1993 IV. No. For more complete treatment on these methods. 1987). They also demonstrated that the currents in the skull and on the scalp contribute negligibly to B and proposed that. Later. This ambiguity can be removed by deflation (Barnard et al. 2. INVERSE PROBLEM A. The deflated equations have a unique solution. (36) is straightforward.F = 0). in a spherically symmetric conductor. A n example of the opposite case is a current loop. which decomposes the potential on the interfaces as V(r)= W0(r)+ W(r). There are primary current distributions that are either magnetically silent ( B = 0 outside G). We shall use several standard results and terms from linear analysis.. where W0 is the solution in the case of a perfectly insulating skull and W(T) is a correction term that vanishes as the skull's Rev. which is the solution for the brain-shaped homogeneous conductor. which is electrically silent but which produces a magnetic field. However. which also applies to Eq. we recommend t h e textbooks by Golub (1989). If \JP\ is constant over a closed surface S inside a homogeneous subregion of G and Jp is normal to S. Press et ah (1992). as was demonstrated by Oostendorp and van Oosterom (1989). Mod. First. the numerical calculation of V(r) is divided into two steps: (1) computation of W0(r). This problem can be alleviated by improving t h e accuracy of the matrix elements H$. Now B can be computed from Eq. A simple example of a magnetically silent source that produces an electric field is a radial dipole in a spherically symmetric conductor. in practice.Hamalainen et al.^ S ' ATT S R = ir-Jpf 477 J V'xv'4-^ . We shall limit our discussion to this task and shall not elaborate on the physiological interpretation in this section. there is no guarantee that the assumption is valid for this particular problem. In general. (36). Peters. 1927.

III. z = 125 mm. is readily obtained if one is able to compute the magnetic field B = B ( Q . However. if Vt is a potential difference between two electrodes. This requires knowledge of the conductivity distribution a{t) so that the effect of volume currents can be properly taken into account.: Magnetoencephalography If the magnetometer consists of a set of planar coils with normals n 7 . r */(Q^G>=2 (43) JLt is called the lead field. dBz/dx. without any reference to the actual conductivity profile a(r) = a(r). Mod. and a tangential derivative of the radial field component. As was shown in Sec. the two tangential components also can be obtained analytically. as was shown in Sec. (45) From this relation. from the computational point of view. JC( depends on the conductivity cr = a(r) and on the coil configuration of the magnetometer. Because of the nonuniqueness.D. Even the limited problem may be difficult to solve in a satisfactory way. Each signal coil is located at x —y = 0 . the area of coil j . 65. (b) Bx. as defined by Eq.n y > 0 yields a positive signal at the output. caused by any dipole Q at any position rQ. j = 1. 1976.L a p l a c e law. ( r ) . there is no evidence to support this conjecture. Tripp. Fig. One of the first derivations of the magnetic field due to current dipoles in conductors of simple shapes was presented by Cuffin and Cohen (1977). it describes the sensitivity distribution of the /th magnetometer (Malmivuo. we find Thus this source produces no electromagnetic field outside the conductor.E. Rev. (46) fsB(Q9TQ)-njdSj 7 = 1 B. The integration extends over Sj. Phys. r ' ) . Small experimental errors can produce large inaccuracies in the solution: the problem is often ill-conditioned.re)=X/(re)-Q. They concluded that Br should be preferred in practical measurements because of the zero contribution from volume currents. . For Q at tQy J /? (r) = Q8(r — rQ). Similarly. so that a field satisfying B . J where the directions of the normals n^ have been chosen to take correctly into account the winding sense of the coils. III. . As an illustration of the form of X . sampled on a surface whose radius is 90 mm. and (c) of a gradiometer measuring dBz /dx. all three components of X^TQ) be found for any tQ. Two-dimensional projections of the lead field JC(T) of point magnetometers measuring (a) Bz. we must confine ourselves to finding a solution among a limited class of source configurations. both B and E are linearly related to J p . Vol.. Value of different measurements Br can be calculated for a spherically symmetric conductor directly from Jp with the A m p e r e . if bt is the output of a magnetometer. m. all field components can be utilized easily and that there is no preferred component. Thus. (43). so that by measuring just the radial field one could justify the use of the simple sphere model in a wide range of actual conductor shapes.* 4 \ \ > • * 4 * * 4 4 * * 4 i 0 < * i 4 i 4. We assume a spherically symmetric conductor and display the lead field distributions on one shell. This means that. Inserting this dipolar primary current distribution into Eq. Methods for handling this restricted inverse problem are the main subject of this section.Hamalainen et a/. 19. The lead field. No. April 1993 .2. W ' > + r t i t * f -4 t4 K FIG. we can find the corresponding electric lead field Jlf such that Vt = f JLf(t>V(T)dv C. (44) . However. there is a vector field jCt{r) satisfying . ^ v * . . we obtain 2»/(Q. 1983). 2.. (43). Lead field bi^fjCiirhPMdv 431 can (b) (c) * • f * „ ' * + * % * + + * * * i t \ tt 4 ' t t * • f H » / > 4 t tX t tX * 1 i 4 * •*»*«-»Ji4«' y ' ' » . This would be the case if nonidealities in the conductor geometry would distort the tangential components more than Br. 19 shows the lead fields of magnetometers on the z axis which are sensitive to the radial field component Bz> the tangential field component Bx.

5 mm. As a result. see Fig. although they were assumed to be pointlike in the data analysis. In their study. at present. several important differences as well as similarities of the two methods may be stated with confidence. (5) N o attempt to fit the sphere model to the local curvature of the head was reported. A recent example about the controversial aspects of comparisons between M E G and E E G is the discussion evoked by a paper by Cohen et al. The critique of the study of Cohen and co-workers can be summarized as follows. whereas telemetric and long-term E E G recordings are possible. such as dBz/dx [see Fig. An 8-mm average error for M E G and a 10-mm error for E E G source location were reported. more expensive than that for EEG. In the spherical model. Both techniques measure weighted integrals of the primary current distribution. a less extensive set of sensors is needed to pick up the essential field data from a local source (see Fig. since the gradient \//(dBz/dx)2-\-(dBz/dy)2 (z is the sensitivity axis of a detector) peaks right above a current-dipole source. Instead of measuring just one or two components of B. 27). i. D. JCI'V — O. Comparison of magnetoencephalography and electroencephalography Although no final evaluation of the capabilities of M E G and E E G can be given at present.. which senses an off-diagonal derivative. An example is the so-called 2D coil (Cohen. orthogonally oriented at 12 locations. whereas they have to be taken into account in the analysis of E E G data. interpretation of E E G signals will require more precise knowledge of the thicknesses and conductivities of the tissues in the head. 1992). More comprehensive discussions of this topic are available (Anogianakis et al. and on experiments with several technical deficiencies on the other. the signal-to-noise ratio for M E G was reduced considerably. Thus M E G and E E G are formally on an equal footing. (1990). therefore. (3) The instrumentation necessary for M E G is more sophisticated and. A multichannel gradiometer with pairs of such coils. JLt = 0 at the origin. Their differences are the following: (1) The lead fields jLt and JLf are different.B. This remaining uncertainty can be fully accounted for by the experimental inadequacies mentioned . A definite comparison between the ability of M E G and E E G to locate the sources thus appeared possible. 1989. respectively. (6) in the current-free space outside the source region. 2. has been developed (Kajola et al. (2) The dipoles were 16 mm long. simulations with realistically shaped conductor models (Hamalainen and Sarvas. Therefore. (4) Twice as many signals were averaged for E E G than for M E G . 1992. (1) Nearly radial current dipoles. the subject has to be immobile during the M E G measurements. Two of the test sources happened to be tangential. Later. Phys. Vol. No. This result is. In the spherical model. of course. Ahonen et al. based on simulations and theoretical calculations on one hand. a somewhat open question. were used as test sources. provided that the sphere model is fitted to the local radius of curvature in the area of interest.. April 1993 (4) M E G measurements can be accomplished more quickly. 37). 1979). (2) Jlf is affected by the conductivities of the skull and the scalp much more than jCt. 1991. (6) N o error estimates were given for the "real" source locations determined from the roentgenograms.432 Hamalainen et al. (3) The magnetic field was inappropriately sampled. 7(c)]. (1990) measured the electric field with 16 scalp electrodes and the magnetic field with a singlechannel rf SQUID magnetometer from 16 sites outside the head. The precise position of each source was determined from roentgenograms. This affects E E G results as well. 65. In addition. M E G is sensitive only to the tangential component of Jp. Cohen and co-workers had the rare opportunity to form artificial current dipoles inside patients' heads by feeding current through depth electrodes that had been inserted into the brains of several epileptic patients to monitor their abnormal cerebral activity. which gave further advantage to E E G in the comparison. 1989) showed that nonidealities in the conductor shape produce equal distortions in all three components of the magnetic field. Rev. These authors claimed that M E G is only marginally more accurate than E E G in locating cerebral electrical activity. both E E G and M E G provide a projection of the primary current distribution on the respective lead fields. On the other hand. one can also design short-baseline coils to measure their derivatives. optimal for E E G but barely detectable by M E G . The paper has been criticized by Hari et al.: Magnetoencephalography It has been shown (Hari and Ilmoniemi. The locations of the test dipoles were calculated on the basis of M E G and E E G measurements. The reported error in the magnetic site determinations was only 5. Mod. whereas E E G senses all primary current components. This alone may cause an uncertainty comparable to the total reported error.e. Cohen et al. 1989) and actual measurements have shown that this arrangement is highly effective. concentric inhomogeneities do not affect the magnetic field at all. The absolute accuracies of both E E G and M E G source determinations are. 1986) that the sphere model gives good results even if the actual conductor is spheroidal. van den Noort et ah. since no electrode contact to the scalp needs to be established. As was shown in Sec. the magnitude of JLt falls off more quickly than Jlf near the center of the sphere. IV.. but also tends to obscure possible differences between the capabilities of M E G and E E G . (1991) and by Williamson (1991) on methodological grounds.. to be expected from the fact that any component of the field can be computed from the distribution of any other component by integrating Eq. Furthermore. Simulations (Knuutila and Hamalainen.

Provided that the coordinate systems for M E G and M R I have been aligned. One can foresee the combination of these imaging methods with M E G and E E G at several levels. Example of source locations superimposed on coronal (a) and sagittal (b) MR images. Furthermore. F. Unfortunately. E E G and M E G are complementary in the sense that measurement with one technique will not always reveal everything that can be found with the other method. With the outline of the structure of interest at hand. the best results are obtained by combining information from both techniques. however. Phys. Positron-emission tomography (PET) gives information about metabolic activity with a spatial resolution of about 4 mm. The white lines indicate transsection levels of the companion slice. for example. All source locations are within a sphere of 15 mm in diameter (Hamalainen. 2. on the M R F s (George et al. some human intervention is needed to produce correct results. Vol. 1991). 1991). Meijs. the results reported by Cohen et al. For example. found by M E G . The white dots show positions of equivalent current dipoles corresponding to the 100-ms deflections of auditory evoked responses in independent repeated measurements. 20. Mod. 1989. U p to now. With the assumption that M E G mainly reflects the activity in the tangential part of cortical currents. one could. FIG.. (1988) in locating dipoles within a sphere and a model skull. 1988). Therefore. No. one meets serious practical difficulties.Hamalainen et a/. Furthermore. Suk et al. At present. similar model experiments are much harder to perform for E E G . Schneider et al. at least in principle. a priori information and modeling inaccuracies need to be incorporated into the analysis. 65. these authors reported a similar precision in determining the site of a cortical source. The accuracy of the inverse problem solutions can be improved by bringing in complementary information to restrict the set of possible source configurations. High-quality anatomical images allow the use of individually shaped realistic conductor models. reliable automatic segmentation of the shape of the brain from M R F s is not trivial. one can superimpose the locations of brain activity. E. Rev. extract the geometry of the cortex from M R images and use the result as a constraint in the source estimation procedures.. M R F s have been used this way only in special cases (see. Combination of magnetoencephalography with tomographic imaging methods Computer-assisted x-ray tomography (CAT) and magnetic-resonance imaging (MRI) provide accurate images of the brain's anatomy with millimeter resolution. because the conductivity distribution of the skull and scalp is difficult to implement in a model head. It is then possible to compare the locations of source estimates with the actual sites of ana- 433 tomical structures (see Fig.. 1989. 1989. This is in line with the 3mm maximum error found by Yamamoto et al. but the time resolution is tens of seconds.: Magnetoencephalography above. Probabilistic approach to the inverse problem 1. April 1993 . it is relatively easy to construct the corresponding triangulation needed for boundary element modeling. (1990) actually support the superiority of M E G in locating a tangential dipole source. Here. 20) or in patients with patho logical findings within the brain. Therefore. Hamalainen. The general solution A study of the restricted inverse problem involves estimation theory: the unknown current sources must be found on the basis of noisy and incomplete measurements.

the exponential density fb(Vb) = c QX o P -2 to/1/*/ (52) where ai=E{7]2i}. (51) i. the computation of the magnetic field of a current dipole in the sphere model. e./-g/(x)|/a/ . where A is a matrix. With this solid foundation we can then more easily formulate the underlying assumptions and establish the theoretical validity of the practical methods used in source analysis. x 0 can be obtained directly by multiplying the measurement bobs by the pseudoinverse of A (Albert. Phys. According to the G a u s s . Under these assumptions. the criterion corresponding to Eq.g ( x ) ] r 2 . 1989). The Bayesian confidence Rev. for example. g ( x ) = Ax. we know the conditional density v(b o b s |b).( 47 ) becomes i/. where S(x) is again the Dirac delta function and.e. (51) is now ^(x) = 2lbobs. which is assumed to be zero. the least-squares estimate is well established even if g(x) is nonlinear. Vol. we would also like to determine a confidence region for x.. the errors of the instrument are independent of input. Before and after the measurement. which gives rise to deviations from the correct values.x(x) = CoPx(x)ffb(bobs-b)e(b\x)db (48) Assuming now a model function b = g(x) which accurately describes the forward problem. We shall assume a measurement with known noise statistics. it is instructive to give a mathematical discussion of the relevant concepts. 1978). Equation (47) nicely separates the sources of information giving rise to x[?x(x). On the other hand. The corresponding probability density functions will be denoted by px(x) and ipx(x). at least in principle. i. 1982. straightforward to include modeling uncertainties such as an imprecisely known conductor geometry. ipx(x) corresponding to the measurement b o b s is ^ ( x ) = C 0 p JC (x)exp{ . First. If we want to find the M L E . through the conventional least-squares search.l [ b o b s . With px constant. If. bobs=b-\-rjb and 7]b is described by a known probability density function fb{rjb).e. giving the chances for b to occur for a given x. x 0 = A t b o b s . for any b . under the assumption of Gaussian errors. With these assumptions one can show (Tarantola.e.e. i. e. v anc Ec ^obs\^~fb^Vb) * l. is found by minimizing the weighted sum of squares 5'(x)-[bobs-g(x)]^-1[bobs-g(x)]. it is readily seen that the maximum-likelihood estimate (MLE) x 0 for x. allows us to incorporate more general or alternative assumptions at any stage of the computation. i. px(x) can be assumed constant in this volume and zero elsewhere.g. 2. 65. We then obtain an estimate for the parameters x that is robust against outliers in b .1 [ b o b s .e. i. disturbed by noise.. (53) i which is often called the L r n o r m criterion. If the neural source is modeled by a current dipole. which started from the general equation for the posterior density. We assume that the unknown current distribution can be identified by a finite-dimensional parameter vector x and the magnetic field to be measured by a vector b. (50) If the only a priori information about x is that it is confined to a given region. since the forward model is described by the probability density 0(b|x).. in addition. The treatment presented above shows that. Gaussian errors 4. A general probabilistic approach to problems of this kind has been introduced (Tarantola and Valette. 1987) and also applied to the neuromagnetic inverse problem (Clarke.: Magnetoencephalography 434 Before describing any source estimation methods. it is. is denoted by b o b s . The forward model is described by 0(b|x). the M L E minimizing S ( x ) is also the minimumvariance linear unbiased estimator for x. Alternative assumptions for noise The above derivation." although the approach of Tarantola (1987)...M a r k o v theorem (Silvey. we might consider a long-tailed probability density for the measurement errors. Eq.Hamalainen et at. the Marquardt (1963) algorithm. April 1993 . i>x(x)=-CQpx(x)fb(bohs-g(x)) (49) 1=E{r]br]b} and by the mean E{t]b}.. Tarantola. 1987) that the posterior density is i>x(x) = C0px(x)fv(bohs\bMb\x)db. the statistics of x are described with prior and posterior distributions. we can take 0(b|x) = 8(b — g(x)). 1972).. the parameter vector x 0 can be found using some iterative numerical technique. Second. consequently. (47) where CQ = l/[ f px(x) f v(bohs\b)6(b\x)dbdx] is a normalization constant. No. if g(x) is linear. Mod. If g(x) is a nonlinear function. This method is sometimes called "Bayesian.. corresponding to the maximum a posteriori density. For example. if we expect that there will be outliers in the measured data. 3. the M L E is commonly called the equivalent current dipole (ECD). The actual observation. 2. the measurement noise. Confidence regions If 7]b is normally distributed.g ( x ) ] ) . provided that g(x) is a linear function of x. (47). its probability distribution is completely determined by the covariance matrix In addition to the M L E . on which the following treatment is based. respectively.g. is accounted for by v(b o b s |b). the a priori knowledge of x is embedded in px(x). Finally. is more general.

. The marginal density for xk then becomes 1 |a (x-x0)| <4Va V A Va ^VVQ- . . Marginal distributions Instead of using the maximum-likelihood estimate and the Bayesian confidence set to characterize tpx(x). The problem can be overcome by studying some marginal densities of tf^x(x). The ellipsoid can be characterized by its volume. 1978). xh n pTi ^ m (•**) = 2 435 Vlirat (xk exp x 0k ) 2a\ (59) l where crk=\/(Q )kk. (60) distribution so G.. If g(x) is nonlinear. In this case the probability densities are computed via integration over a subset xk . 1987) that these limits are given by r r r the remaining variables xk =f " 1 * (56) For instance. on a set of parallel planes. 1 <p < n. t/>m is a function of three variables and one can easily plot its contour lines. The estimates of the location.yk. . If the source model consists of k current dipoles. .Hamalainen et a/. of the components of x. . . If g(x) is linear and the errors are Gaussian. . Vol. . 9xm. Then ipm(xl) = t/>m(xl. 1978). which is given by the product of the diagonals of A.yuzl.: Magnetoencephalography region is defined as the smallest collection of values of x making up a given total posterior probability (Silvey. The points corresponding to the highest probabilities then make up the confidence region. . In practice one is often interested in finding the extremal values of some linear combination a = aTx of the model parameters in the confidence ellipsoid. (54) where J is the Jacobian of g(x) at x 0 : Jij=dgi/dXj . 1987) that the/^-percentage confidence region for x is an x 0 -centered m-dimensional ellipsoid given by (x-x0)rQ(x-x0)<< (55) where c^ is the p-percentage point of the ^ " d i s t r i b u tion and Q = J : r 2 : ~ 1 J = V r A ~ 2 V . Moreover. Mod. and orientation of the dipole were based on a simple geometrical construction utilizing the shape of the field map. (57) In practice. giving \xt —x0i\ <c^nyQiYl. One possibility of finding this region is by appropriately sampling ^ ( x ) and then sorting the values in decreasing order. the form of I/JX(X) is generally very hard to imagine. amplitude. one could determine the size of the smallest rectangular box whose faces are parallel to the coordinate planes. . (54). This region contains the most probable parameter vectors on the basis of a given measurement.1) thatP{|jc| <c)=p/100. the last form being the eigenvalue decomposition. so that the dependence of g on x 2 yxk is linear. . it is straightforward to show that the posterior density of Eq. (58) g ( x ) = Afx 1 )x 2 . . it is. . if x can be decomposed into two parts. The equivalent current dipole 1.. if g(x) is a linear function. (50) can be integrated over all but one of the parameters x j . for example. 8 m ) r . the computation of marginal densities is a tedious task. Assuming a constant px(x) and the linearization given by Eq. .x 0 ) + g(x 0 ) . Using marginal densities. the Bayesian confidence region is tedious to compute. one can also consider the confidence limits for a single parameter which describe the variance of one parameter when the others are allowed to attain any values. . The resulting marginal density if>m is a function of Rev. If Bz is measured on the plane z = z 0 above a . |x-x0 It can then be shown (Sarvas. it is sometimes useful to study tf>x(x) itself. the integration over x 2 can be performed analytically (Hamalainen et al. possible to integrate numerically over all but one of their locations to reduce the number of variables to three. Phys.xk.zk) is the marginal density of all dipole locations. If the model consists of only one dipole. x 1 = (xA: . Alternatively.. April 1993 only. 5. since it has many variables. . It is easy to see that the ppercentage confidence interval for xk is \xk x 0k\ < c a k * where c is chosen from the x~N(0. the linear part contains the components of the dipoles and the nonlinear part their locations. to determine the confidence limits for xi9 we take a = ( 8 h . If there are several dipoles. Finding dipole parameters The dipolar appearance of the magnetic-field pattern due to a localized cortical current source was first revealed in a study of the somatosensory evoked field (Brenner et aL.y 1987). . However. No. The linear approximation may also be used if g(x) in the vicinity of x 0 can be reasonably well approximated by g ( x ) ^ J ( x . . viz. 65. . i. . where 8 f/ is the Kronecker delta function. It can be shown (Sarvas. However. and thus any value outside it is less plausible than any value inside. . it can be integrated analytically.. ftf^(x)dxki '"dxk . xk . the determination of the Bayesian confidence region becomes much easier. since it involves numerical integration over multiple variables. in principle. times a constant that depends on m.e.xk )T and )T. . 2.

Kaukoranta. 1983). Still another possibility is to compute alternating subaverages of the evoked responses by reversing the sign of every second response. 4. if the noise levels are over. . . (61) 1=1 is used. Therefore. Vol. et al. . IV. bn are the experimental data and bx. V.. 1986). xlbs *s distributed as x2n -r> where r is the number of model parameters. A similar approach can be used for a dipole embedded in a spherical conductor (Williamson and Kaufman. r = 5 . the model represents the data well and adding more sources is not reasonable.. which is a nonlinear operation and. The data are fitted separately for each time instant. both in t h e leastsquares fitting and in the confidence-region calculations. Confidence limits The problem of finding suitable and adequate confidence limits for the dipole model has been discussed by several authors (Okada et al. If g = 1. indeed. the model agrees completely with the measurement. Overestimation of at leads to missing some of the details in the actual source configuration.E)..or underestimated.an). agrees with the experimental data. and strength are allowed to change with time. at may be obtained from the standard errors of the mean. 1981). 1986. the source is assumed to be dynamic so that its location. The dipole model is useful even for identifying multiple. we can simply use the values sampled within the base line and calculate their standard deviation. . the electric potential distributions are smeared and largely overlapping. . In the so-called moving-dipole model. Stok. Pohs changes quite rapidly from 0 to 1 when the at9s increase. When both SI and S2 are simultaneously active. 3. t h e responses are likely to be very similar and one obtains an estimate for the noise. Hamalainen et ah. T h e outcome of t h e model validation tests and the sizes of the confidence regions depend crucially on the reliability of the cr/s. 1984). g = l-i. individual responses must be filtered before computing the standard errors. 2 = d i a g ( a 1 . digital filters can be applied to t h e subaverages to account correctly for their effects. we lose information about any time dependence of a {. . simultaneously active sources lying far away from each other. . Correspondingly.: Magnetoencephalography horizontally layered conductor. This choice of g is analogous to the measure widely used in linear regression analysis (Kaukoranta. with temporal correlations ignored.2 and Figs. ages is kept small. et al.A) and subject noise sources (see Sec. . where the or/s contain the effects of both the environmental (see Sec. Phys. does not commute with filtering. If P o b s « l . 1993). to establish the proper covariance matrix 2 .bn are the values given by the equivalent dipole. In this case the orientations of the sources are particularly favorable for M E G . Mod. If g =0. The standard method of estimating the location of a simple source is to determine the equivalent current dipole by a nonlinear least-squares search (Tuomisto et al. . 54 and 55). which is determined by means of a leastsquares search. For a dipole in a spherically symmetric conductor. However. Estimation of noise Xobs= 2 We saw in Sec. If our data have a base line containing only background activity. orientation. a low probability indicates that the dipole model is unsatisfactory. the goodness-of-fit value. with our relatively small n. which makes it difficult to see that there are. I n contrast.C. If the length of the subaverRev. VI. Hamalainen..F that estimates of noise levels as well as their correlations are needed. The dipole lies below the midpoint of the field extrema and is perpendicular to t h e plane passing through t h e minimum and the maximum and parallel to the z axis. . However.(bi-$i)2/^b? i=i ' .. 1984. I. Therefore. while their underestimation may fool the experimenter to a more complex model than is actually allowed by the noisy data. therefore. we suggest that several methods should be applied and their results compared. Below we shall assume no correlations between the channels. the model is irrelevant and does not describe the measurements any better than a zero field would. three spatially separated sources (see Sec. Since one is now dealing with a linear operation. the magnetic-field pattern for each dipole is distinct and the existence of three separate sources is evident. Therefore. one easily obtains misleading results (Supek and Aine. Hamalainen. the relation between the field pattern and the dipolar source is particularly simple. If we are averaging several responses. . . Validity of the model To describe how well the field pattern of an equivalent current dipole. No..436 Hamalainen et a/. It can be shown easily that the depth d of the dipole is related t o the distance A between the extrema by d = A / v / 2 . Deviations of g from 1 are caused by measurement noise and by the inadequacy of the source model. April 1993 i=l -2 • (62) °~i With Gaussian errors in the measured data. if we apply digital filtering to the data. t h e latter in both hemispheres. With reliable estimates for ut it is also useful to consider the x2 value: 2. such as possible variations of incoherent background activity during the response. The probability Pobs=P{xl-r <Xobs) t h e n gives a quantitative measure for the goodness of the model. 2. . We can use several approaches to find the noise level estimates for individual channels. 65. here b j . However. 1987. as demonstrated by measurements of activity in the first and second somatosensory cortices SI and S2 (Hari et al.

the dipoles are allowed to change their amplitudes in order to produce a field distribution that matches the experimental values. x VQ% 1 1 20 1 40 1 1 60 Depth of the equivalent current dipole (mm) FIG. and slow waves ( X ) observed during sleep]. 1989). This difference can be qualitatively understood from the effect that dipole displacements have on the field pattern (Cohen and Cuffin.C. The same is also evident from a comparison of the linear and nonlinear confidence sets (Hamalainen et al.. 30. An example of this approach is the separation of activities from the first and second somatosensory cortices (Sec. respectively. and Sarvas. since the modeling error is only weakly reflected in the (a) 437 goodness-of-fit measure g (Hari. and Sarvas. Phys. at least qualitatively. being about half of the deviations in the longitudinal direction and in depth. 1985) that the application of the single-dipole model to the interpretation of the field produced by a side-by-side distributed source results in an equivalent dipole that is deeper and stronger than the actual one (Okada. The source determination was based on nonaveraged spontaneous signals. 1983). 1988). and a dipole distribution along a line. 21. Joutsiniemi. but the same approach can be used in M E G studies as well (Scherg et al.. His data are. Modified from Hari (1991a). where (b) Measured Simulated 1 1 1 ! % 400 — - • X • % °x • XxX o x • * 60 80 „• - . the field patterns may show only minor overlap and they can be fitted individually using the singledipole model. If the distance between the individual dipoles is sufficiently large ( > 4 cm) and their orientations are favorable.Hamalainen et a/. (a) Simulated 95% confidence regions in the dipole-depth-dipolestrength plane.. Another point of concern is the resemblance between the field distributions produced by a single dipole. 21). 2. we must resort to a multidipole calculation to obtain correct results. A 10-nAm dipole was placed either 20. one finds that the deviation in the transverse direction is smallest. for superficial current sources it is sufficient to consider the linear approximation given by Eqs. also their orientation throughout the time interval of interest. 1990). it is often possible to recognize each source separately. H. in the transverse direction (perpendicular to the dipole and its location vector). Mod. Formulation of the problem The obvious generalization of the single-dipole model is an assumption of multiple. or 50 mm beneath the surface in a spherical conductor with 100-mm radius. care must be exercised when the accuracy of estimates for the current dipole depth is determined. Comparing the limits given by Eq. However. when the sources overlap both temporally and spatially. Simulations indicate (Cuffin.. VI. and in depth (along the location vector). However. who showed that the true Bayesian confidence set will be highly nonsymmetrical (banana-shaped) in the dipole-depth-dipole-strength plane (see Fig. vertex waves (•). (b) Depth-strength dependence of ECDs for different magnetic phenomena [magnetic counterparts of K-complexes (O ). in agreement with Hari et al. Correlation between strength and depth of a dipole. 1988). However. It is useful to calculate the confidence limits in the longitudinal direction (in the direction of the ECD). 65. The noise level was 30 fT. We shall denote the measured and predicted data by the matrices Mjk and Bjh. Joutsiniemi. and the strongest depth-strength dependence is illustrated. April 1993 . (55) and (56). Rev. a sideby-side dipole pair. 1987). 1985). Therefore.2). see Scherg. Similarly. spatially separated dipolar sources. Vol. (60) with those given by a Monte Carlo experiment. Multidipole models 1. Stok compared the single-parameter linear confidence limits of Eq. No. The basic assumption of the model is that there are several dipolar sources that maintain their position and. This method was first applied to E E G analysis (for a review. if the temporal behaviors of the dipoles differ. Hari. optionally.: Magnetoencephalography 1987. H*£ »*• -0 £• 5" 2 200 - 40 • ~~\ x — 0 1 . This is a serious problem. 40. (56) in these three directions. An effective approach to this modeling problem is to take into account the spatiotemporal course of the signals as a whole instead of considering each time sample separately.

whereas ||-||^ denotes the square of the Frobenius norm: 2 2 4 =Tr( ArA) (64) We assume p dipoles located at rd9 d — 1. . . even if the dipole model is correct and we have chosen r such that the remaining 5 m i n can be explained by noise. = 1 . r J R ( ^ X r ) Q ( r X m) (65) where the dimensions of each matrix appear as superscripts. The matrix R contains the differentiation between fixed. . . however. respectively. n indexes the measurement points and k = 1. Vol.and variableorientation dipoles. are known. 2. 5 m i n . . . .ee) and bj(rd. Yin and Krishnaiah. Phys.. 1990) that among all B's of rank r the minimum of Eq. m corresponds to the time instants tk under consideration. the data predicted by the model can be written as B (nXm) = G ( w X 2 " ) (r 1 . 0 3. (65)] contain the amplitude time series of the fixed dipoles at tk9 Rev.: Magnetoencephalography 438 j. 1987) or just by studying the general behavior of Smin when r is increased./?. . while the remaining 2p2 rows are the time series of the two components of each of the variableorientation dipoles. It can be shown (de Munck. the remaining minimization problem Smin=min||GQ-M||! (71) .| . pointing in the directions of the unit vectors e# and e^ of the spherical coordinate system. Here G is a gain matrix composed of the unit dipole signals k = 1. . we want to retrieve r —p\ + 2p2 dipole wave forms. m.xa (63) |2 \F > where xl9 . . . .. There will be p1 fixed-orientation dipoles and p2 dipoles with a variable orientation (p =px +p2)• Therefore. . . R = I and the data are given by (2p ) where Ir 22 is the 2p2X2p2 identity matrix.. . 65. 22. However. 2. j = (66) !. as was pointed out by Mosher et ah (1990). This does not mean. (63). If all the orientations are varying. Finally.« where bj(rd. Assuming that the dipole locations rd. . No. The conventional method for doing this (de Munck.and variable-orientation dipoles: COS/3] 0 0 sin/?! 0 0 0 COS/? 2 0 0 sin/?2 0 0 •• • 0 0 R = 0 0 cos/3Pi 0 0 sin/?^ 0 0 *k (69) > where kk are the singular values of M . ...) . . . 1985. With these assumptions. . arranged in decreasing order. . (70) Equations (69) and (70) tell us to what extent the measured data can be explained by r sources that have linearly independent wave forms and spatial distributions detected by our set of sensors. respectively. April 1993 B = Q(TU. A reasonable value for r can be obtained either by comparing kk/Vm ("effective noise level") with the estimated noise level of the measurements (Wax and Kailath.. . Mod. the first px rows of Q [see Eq. we still have an ambiguity in dividing the available r degrees of freedom between variable-orientation and fixed dipoles under the constraint p { + 2p2 = r. . respectively.9TP)Q . . We seek the minimum for the conventional least-squares error function IM — B(xl9 . . . the number of fixed. satisfies rank(M) > 2 jt2d==bji<td^(p)y d = l . one for each fixed-orientation and two for each variable-orientation dipole. . . p . p . . Linear optimization j(^2) (67) If the model consists only of variable-orientation dipoles. . (63).Hamalainen et al. that the sources are necessarily dipolar. . .xg are the unknown model parameters. . . correct values must be chosen for px and p2. d = 1. 9pl9 with respect toe0. . Examples of g m a x and kk /Vm as functions of r are depicted in Fig. ' k = l A. . and further restrict our attention to the spherical conductor model so that any measurement is insensitive to radially oriented dipoles. The fixed dipoles form angles pk9 k = 1 . We obtain for the goodness of fit /rank(M) ranmiv r A: = l G (68) / y. Selecting the correct model Before minimizing Eq. 1990) is based on an analysis of the singular-value decomposition (SVD) of the measured data M = UAVr: where A is a diagonal matrix containing the singular values of M while U and V are unitary matrices formed by its left and right singular vectors.e(p) are the signals that would be produced by unit dipoles at rd. r=2p2 = 2p and R = I (2/. ..

However. the results may strongly depend on the person who is in charge of the analysis. . while ulk and vlk are the corresponding left and right singular vectors. d = 1. Characteristics of multiple-source modeling. are formed. (73) can be used to find 5 m i n at each iteration step of the nonlinear optimization algorithm. be even more time consuming than finding the orientations by nonlinear minimization. which cannot be explained by the given number of independent sources. Q*«*i*[*i**f*] (74) where klk is the larger of the two singular values of Q^. 2. (72) where the dagger denotes the pseudoinverse.\Piml.1fc=(cosj5fcSin/Sfc)r. recently proposed approach (Mosher et al.. The data were bandpass filtered to 0. April 1993 readily identify i. P ± [Eq.p. This method is identical with the MUSIC algorithm (multiple signal classification).y 1992) for searching the optimal locations of several simultaneously active current dipoles. Then. This stems mainly from the fact that reasonable initial guesses are difficult to make. (73)]. Subsequently. From Eq. it was proposed (de Munck. 22. (74) to fixed dipoles only. In the traditional approach of Scherg (1990). one may .y 1990) that one could start with variable-orientation dipoles only and then find an optimal orientation for the fixed dipoles. . 1990) that the orientation angles could be found by a linear approach. viz. When one minimizes over the nonlinear parameters Td. we are still at a stage where the expertise and physiological intuition of the experimenter are crucial for providing reasonable solutions. each containing two successive rows of Q. (72) is first applied to find Q for r variable-orientation dipoles. whereas s u = ||(i-QQ^MWy. The basic idea is to find the parameters giving rise to 5 m i n in four steps: (1) Decide the number of variableorientation and fixed dipoles.Hamalainen et a/. (a) Behavior of gmax as a function of the number of independent sources. Equivalently. The term in brackets is the time series for the fixed dipole. So far the selection of the initial dipole positions has been based mainly on heuristic methods. (b) Level of noise corresponding to the residual field. . . .: Magnetoencephalography 0 2 4 6 8 10 Number of independent sources 439 0 5 10 15 20 Number of independent sources FIG. The curves were calculated from somatosensory measurements performed with a 24-channel gradiometer.. No. has been one of the central problems in multidipole modeling. therefore. G Q is the projection of M to the column space of G: G Q = P. . k = 1 . In the next two sections we shall concentrate on an elegant. Four different fingers were stimulated. d — 1. . (74) we can Rev. The variableorientation and fixed-dipole cases can be combined simply by applying Eq. respectively.. Mod. In this procedure. . Phys. 2 X r a submatrices Qfc. Eq.. Recently it was shown (Mosher et al. For this purpose one just replaces G in Eqs. (73) where P ± is a projector to the left nullspace of G. One then finds the best rank-one approximation for Qk through the SVD. Eq. the time range considered in the analysis was 0-100 ms. this variation requires an iteration between two linear optimizations and may. (2) Find the best projector. the calculation may get trapped into a local minimum and 5 m i n is never reached. . In fact.M.y 1989). Consequently. introduced earlier in another context (Schmidt.05-100 Hz. Although an expert-system approach has been proposed (Palfreyman et al. is linear with the solution Q = GfM . 65. regardless of G and R. 4. and Q is found by the above linear approach./?. 1986). all dipoles have fixed orientations. (3) Find G and R for which I — (GRKGR) 1 " is as close as possible to the optimal projector. In addition. He includes the orientation angles in the set of nonlinear parameters. . Vol. Especially when M E G is applied in clinical practice it is of vital importance to provide the means to remove these ambiguities. (71)-(73) by G = G R . Finding dipole positions Bm The multiplesignal-classification approach The remaining nonlinear optimization for rd./?. . assuming a certain combination of variable-orientation and fixed dipoles.

we can proceed as in the fixed-dipole case. Fixed-orientation dipoles fixed-orientation . Normalization is necessary to distinguish orthogonality from small values due to small gain. Phys. If. . therefore. The r independent sources are no longer dipolar and. . To detect the variable-orientation dipoles we may. The combined case When there are both fixed. (80) where G r f = G ^ / | | G ^ | | is a normalized gain matrix. both eigenvalues of the bracketed term in Eq. The algorithm for finding the dipole positions now readily emerges. (78) The best orthogonal projector minimizing SL is thus Pi=UxUf. produces reasonable results. It can be shown (Mosher et aL.(r)mfe = min . m^G. Applications and limitations The appeal of the scanning method is that one can sort out the feasible source positions relatively quickly.Gp ). For — \Gk(r)mk\22 ^/(r)=^mi„[UG4(r)rPiUG/i(r)] . (1992) we shall consider first variable-orientation dipoles only (r =2p2). If our assumptions are correct. . c. and ML is the corresponding orthogonal complement projection. . For convenience. u r ) and U 1 = (u r + 1. I.: Magnetoencephalography 440 require that G R be orthonormal to P x . . d. and ||-|| 2 denotes the Euclidean norm of a vector. Sf{r)=mm f mk b. To find the dipole positions we can again construct a contour plot of SJ-(T) or 1/Sy(r) and search for extrema. each Gd must be orthogonal to P 1 . Of course. we cannot find a set of r single-dipole gain submatrices G^ which are perpendicular to P ± . (82). . if there are strongly correlating sources whose field patterns overlap. However. Starting from the definition of X. the situation may be complicated by noise and. (75) k=l k^r+\ (76) where My is the best rank-r approximation of M. indeed. Therefore. Mosher et ah (1992) have also shown with both simulations and applications to actual measurements that the method.(r)rP1G. it is instructive to provide a contour plot of SV(T) or \/SV(T) on plane intersections to get a comprehensive picture of the situation. (82) are close to zero. 1992) that (82) n EEMH + M J . the source is a collection of p2 variable-orientation dipoles. an alternative description exists for the nonuniqueness of the inverse problem. compute SV(T) and check whether it is at a minimum as well. Therefore. M H = P „ M = U„U B r M and M 1 = P i M = U 1 U f M . . We can again make use of the SVD of M and write r placed by G R and that in the final step our scanning function is dipoles we need to find the op- timal orientations as well. we should find exactly p2 minima. (43). Current-distribution models One can also search for more general solutions of the neuromagnetic inverse problem instead of working with source models in the strict sense described above.Hamalainen et a/. in addition. r=zpl. the predictions will. . . in general. We write G = (Gl9 . We next proceed to find G such that SG = 110^11^ (79) is minimized. . however. Only current distributions that yield b^O for at least one / can be detected. the search can be accomplished quite quickly. April 1993 where ^ m i n [*] denotes the minimum eigenvalue of the bracketed 2 X 2 matrix and U j is the matrix formed by the left singular vectors of Gk. . mk m£Gfc(r) 7 Gfc(r)mfc (81) where m^ = ( cos/3ksm(3k ) r . it will be assumed that n < m and that rank(M) = n. Vol. Since we are just moving one dipole.un ) . where the Gfc's are nX2 submatrices containing the unit dipole fields. G^m^ is orthogonal to P x for all m^ and. Now.-(r) in Eq. If we hit a location of a variable-orientation dipole while evaluating Eq. . it is meaningful to investigate the "scanning function" (Mosher et ah. except that G is reRev. indeed. 65. . (4) Find the dipole time series with the methods described in the previous section. therefore. 1992) 5u(r) = ||P1G^(r)||^.and variable-orientation dipoles. be misleading. We proceed as in the previous subsection. The extreme simplicity of assuming one or a few pointlike sources can be replaced by an unambiguous solution with minimal assumptions. we can always use the fixed-orientation dipole scanning method.. The rank of P ± is n — r —n — 2p2 and the optimal PL minimizes S ± = ||P X M|[|. (77) where U (| = (u 1 . Therefore. e. . r =P2p2. No. One computes Sv(r) in a grid of viable dipole positions and searches for minima. therefore. Mod. 2. Variable-orientation dipoles Following Mosher et dl.

1989.Xn). No. This means that directions in 5P with poor coupling to the sensors are suppressed. Minimum-norm estimates In the following. the approach was further refined (Clarke et al. Eq.. Minimum-norm estimate / * visualized as a projection of the current vector Jp into a subspace spanned by the lead fields JC\ and JL2. 1987). with V r V = I and A = diag(A.. \JP\2=<Jp.. 23. n—2 measurement channels were assumed. In general. (1990) indicates that the point-source and distributed-source methods can be naturally linked by introducing restrictions to the set of feasible current source candidates. With this notation. the inner-product matrix II is nonsingular and w=n~ 1 b (86) However. Generalization to a typical multidimensional case with n ^ 100 is straightforward. Figure 23 illustrates some of the concepts discussed above. Hamalainen and Ilmoniemi.: Magnetoencephalography The first attempt of this kind was the minimum-norm estimate (MNE.j). depending on the nature of the problem. . (85) can be compactly written as J*=wTX. . When referring to current distributions as elements of the current space.For simplicity.n.X. but contains a component in the complement subspace £±. (Xi. . J* will be a linear combination of the lead fields. 1985. we use capital letters. a surface. The term minimum-norm estimate derives its name from the fact that / * is the current distribution with the smallest overall amplitude that is capable of explaining the measured signals in the sense of the norm defined by Eq.Jp) = f \Jp(T)dG (84) .0) to obtain a regularized inverse n _ 1 = V A _ 1 V r . . . . (43) it is evident that measurements of bi = {jLi. Requiring that / * reproduce the measured signals. From Eq. .. I I ~ 1 = V A ~ 1 V T . recent work by Ioannides et al. . 1987. viz.Jp). A related procedure.1 by A ~ 1 = diag(Xr 1 . Crowley et al. Then. In other words. April 1993 FIG.JO = f Jpi(r)'Jp2(r)dG (83) The overall amplitude of a current distribution is described by its norm. . Mod. (84). 65. we obtain a set of linear equations b = IIw.wn)7 and II is an n X n matrix containing the inner products of the lead fields. The inner product of J\ G 5^ and J\ G ^ is defined by (Jpi.Hamalainen et al. 5^ will be called the current space. The set G into which Jp is confined may be a curve. Ioannides et al.JLn ). was introduced by Kullmann and Dallas (1987). 1 . 1984. The idea of a M N E is that we search for an estimate /•* for Jp that is confined to 5P. . .„>0 are the eigenvalues of II. . The cutoff value k <n is selected so that the regularized M N E does not contain excessive contribution from noise.. . using Fourier-space methods. .. A measurement will give only the projection of Jp on o£|j. or a combination of discrete points. Wang et al. .. 1989. the third axis is perpendicular to all lead fields of the magnetometer array under consideration.Jp). .i = \. The 7=1 where Wj are scalars to be determined from the measurements. 2.. 1.J*)=bi = (Xi. Subsequently. 0 .bn)1 w= --(wl9. which is the minimum-norm estimate / * .. . the primary current distribution Jp is not confined to the subspace JL^ spanned by the lead fields. only yield information about primary currents lying in the subspace 5P that is spanned by the lead fields: 3 r ' = span(X 1 .n. 1993). 1989. The nonuniqueness of the inverse problem is Rev. a volume. . Let n = V A V r . . . ^ J . where x=(x 1 . . Vol. any primary current distribution to which the measuring instrument is not sensitive may be added to the solution. 2. . Thus II can possess some very small eigenvalues.. . Regularization If the lead fields are linearly independent. To avoid this numerical instability. Sarvas. (85) 441 manifested by the fact that the actual current distribution producing b may be any current of the form J =J* -\-Ji9 where JL satisfies (Ji. Two of the axes are in the plane defined by the lead fields X j and X 2 .x„) . Phys. Then. where b= (bl. . i = l.Xi)=0. Hamalainen and Ilmoniemi. IImoniemi et al. which causes large errors in the computation of w. primary current distributions will be considered as elements of a function space $ that contains all square-integrable current distributions confined to a known set of points G inside a conductor. the solution must be regularized (Sarvas. 1992). Regularization may be carried out by replacing A .. which is generally the case when the measurements are made at different locations. Three dimensions of the current space are depicted.. the X / s in 5^ may be almost linearly dependent. . . where kx>k2> ••'•A. Iiij = {jCi. . Furthermore.

FIG. since lead fields given by Eq.. regularization means that those "eigenleads" that correspond to small eigenvalues and are hard to measure with sufficient signal-to-noise ratio are ignored. Vol. However. Different octants of the visual field were stimulated with foveal (F. 24). 24. Mod. 1987). 1992. Analogous simulations for a spherically symmetric conductor were computed by Hamalainen and Ilmoniemi (1993). April 1993 . second and third row) and parafoveal (P. 25. No. The source to be estimated was an assembly of current dipoles lying in the plane. In terms of current distributions. 2.442 Hamalainen et a/. The effect of regularization on these data is shown in Fig. The M N E ' s approximated reasonably well the locations of the original sources. Modified from Ahlfors et ah (1992). The magnetic field was recorded with a 24-channel gradiometer over the occipital cortex (see Fig. 37 and inset at the upper left corner).X ^ O ) . the minimum-norm estimates also will be solenoidal and thus nonsolenoidal currents are not properly reconstructed. Minimum-norm estimates (small arrows) and equivalent current dipoles (shadowed arrows) calculated from evoked visual MEG data. 65. see also Fig. where b = n w = n n _ 1 b .: Magnetoencephalography resulting estimate does not reproduce exactly the measured signals.. but the misfit b—b. is in accordance with experimental errors (Sarvas. Phys. first and last row) checkerboard patterns. (87) Examples of M N E ' s for primary currents in a horizontally layered conductor were discussed by Hamalainen and Ilmoniemi (1984). (43) are solenoidal ( V . Rev. The regularized minimum-norm solution is J* = (f[~1b)7X . M N E ' s were also applied to the analysis of visual evoked responses (Ahlfors et al.

therefore. Since the inner products must be recalculated during each iteration. see Sec. cavities filled with highly conducting cerebrospinal fluid. The results clearly call for additional experiments for verification and other modeling approaches to confirm the source locations and their temporal behavior. this is often an oversimplification. 60 responses were averaged. solutions that are intermediate between the M N E and the dipole model can be obtained. Only the first steps in this direction have been taken. Although the preliminary results are promising. while (b) and (c) represent two trials of 20 stimulus repetitions from the same measurement location.I. (1990) have utilized a transputer system to speed up the calculations. In (a). Phys. (88) The solution is obtained as the limiting value of a sequence of current estimates J{k\ k = 0 . In such a case.9 1991). 24.or multidipole models represent two extreme approaches to source modeling. The M N E is an optimal estimate when minimal prior information is assumed or used. The resulting source estimates are well localized and they seem to be able to reproduce the sites of simulated point sources remarkably well. 3. An interesting iterative approach has been introduced by Ioannides et al. (1990).: Magnetoencephalography (Jl>J2>a>=f FIG. it is not yet clear whether the two source regions that were observed to have a constant phase shift can be reconstructed consistently or if the observations are due to biasing the analysis with the desired enhancement of deep sources. The two particular functions tried by Ioannides et al. 25. Ioannides et al. The value of d at the lower left corner of each map indicates how many eigenleads out of the possible 24 were used. For k = 0 . P E T and the functional M R I provide metabolic blood flow data. Mod. No. . Each of the estimates is computed as in the minimum-norm solution of Eq. . 1 . Minimum-norm estimates with a priori assumptions J. The dipole model is preferable when one knows that only a small patch of the cortex produced the measured field. thus J(0)=J* is the minimum-norm solution. Rev. white matter with its considerable anisotropy as well as anisotropy of the cortex itself.. a weighting function that increases monotonically with source depth has been applied to differentiate cortical and thalamic activity (Ribary et al. it is necessary to combine M E G data with supplementary information. the weight function employed to calculate the dot-product matrix II changes at each iteration and. The important feature of this method is that it can reveal the number and locations of activated areas. (87). one actually has a sequence H{k) of inner-product matrices as well.. they had to alleviate the computational burden by choosing a suitable set of lead fields to be used as basis functions. Modified from Ahlfors et al. We need a system in which M R I gives structural information. holes in the cranium.2. co{0)(r)= 1. unless one can prove that some of these complications are not important in practice. 65. 2. if supplementary information is available. with the only prior assumption being that there is a small number of localized sources. However. IV. . Quite recently. In successive iterations the weight function depends on the previous estimate. 443 CO(T)Jl(T)'J2(r)dG . The variable thickness and conductivity of the skull. and E E G supplements M E G in obtaining data about signal processing in the brain. at least in simulations. (1992). co{k)=f(J{k~1)). this procedure is computationally demanding. without any prior assumption about their number or sites. The method is able to resolve localized sources. However.Hamalainen et al. all have to be taken into account. . Vol. viz. For more details. April 1993 . The improved accuracy of anatomic information and the combination of M E G with E E G will necessitate considerable improvements in modeling the conductivity of the head. Effect of regularization on the data shown in Fig. were the square and the square root of J{k~l). Because of the nonuniqueness of the neuromagnetic inverse problem. Even so. The iterative distributed-source approach assumes a weighting function in the inner product of two current distributions. Future trends Minimum-norm estimates versus single.

problems may still occur. laboratory instruments. In this case.. 1987) produce fields that are larger than the brain signals by 14-15 orders of magnitude. various laboratory and hospital instruments generate strong noise. the monitor must be outside. V. In urban areas. sounds are typically presented to the subject via plastic tubes and earpieces. may ruin the experiments by decreasing the SQUID gain and increasing the noise level. geomagnetic applications. For amplitudes and spectra of various noise sources. Decreased instrumental noise has revealed. which may be difficult to arrange in some hospitals. measured over the chest. 2. Fortunately. To a large extent. new sources of disturbances that were not seen previously. In addition. This is much more troublesome. A safe separation between an M E G system. behind a hole in the wall. trains and trolleys can be a nuisance. With modern thin-film dc SQUIDs. the same principles and methods can be applied to studies of cardiac signals (MCG). removal of the power-line frequency and its harmonics with comb filters does not help. much emphasis must be put on the development of techniques for handling signals that arise from a multitude of simultaneous sources. large disturbances due to vibrations will be seen in the magnetometer output. However. Visual stimuli can also be led into the room via a mirror system or along optic fibers. The gradients used in the image formation.B. and Cantor et al.. in particular if they are dc operated. is two orders of magnitude larger than that of typical evoked brain responses. In addition. and monopole detectors or to any measurement employing multiple. or computers. must be removed. INSTRUMENTATION FOR MAGNETOENCEPHALOGRAPHY In this section we shall discuss instrumentation for neuromagnetic recordings in detail. Electric motors. Below 1 Hz. Phys. however. the frequency band of the disturbances overlaps that of the signals. are rapidly switched on and off. artefacts may be caused by mechanical movements of the body in the rhythm of the heartbeat or breathing. Ahlfors. 1984). due to broadcasting. especially if they are strong or infrequent. (1985). If a magnetically shielded room is used. The fundamental noise barrier for measurements is the Nyquist current noise of the conducting human body. watches. Static fields can cause flux trapping in the superconducting structures of the SQUID. the magnetic noise level is several orders of magnitude higher than the biomagnetic signals to be measured. Many stimulus generators also produce artefacts.444 Hamalainen et a/. and electric stimuli in somatosensory experiments are delivered through tightly Rev. For example. Environmental and instrumental noise sources In a laboratory environment. power lines. the magnetometer noise can be well below the level of the brain signals. However. such as spectacles. 8. In laboratories that are near public roads. and the contribution of geomagnetic fluctuations is negligible. If the measuring device is not well supported. The limiting factor for the intrinsic noise of the magnetometers is the thermal noise in the SQUID itself. April 1993 twisted pairs of wires.. simultaneously active SQUID magnetometers (Ilmoniemi et al. A. Mod. radio-frequency fields. the heart is farther away. all cause disturbances. Electric currents in other muscles evoke magnetic fields.1 fT/V'Hz (Varpula and Poutanen. moving automobiles and buses may also cause significant interference. both may be time-locked to the stimuli. contributions from the state-of-the-art electronics are negligible. which also results in a degraded performance. and hooks. being on the order of 0. naturally occurring geomagnetic fluctuations constitute a significant source of noise. but such artefacts have not produced any significant problems during M E G recordings. Cardiac disturbances can usually be dealt with by signal averaging as well (see Sec. In addition. inside a shielded room and measured with a gradiometer (see Sec. Thermal noise generated in the aluminum foils used as radiation shields in Dewars is now a main factor limiting sensitivity if other external environmental disturbances have been eliminated.B. V. elevators. etc. operated inside a magnetically shielded room. Even if the interference is well outside the measurement band. the spectral density of such field variations can exceed l p T / V H z (Fraser-Smith and Buxton.I and V. (1991). In particular. However. Field sensitivities of a few fT/VHz have been reported. the main field of an M R I magnet is highly stable under normal operation. by Maniewski et al. 1989). The video monitor employed for visual stimuli is also a source of noise.: Magnetoencephalography We also need more refined filtering methods for the estimation of signal wave forms from noisy data. Sometimes the magnetic field due to cardiac activity can contaminate cerebral measurements. Vol. Knuutila. especially at low frequencies. all magnetic material on the subject. et al. see Fig. Noise reduction 1. and an M R I installation is a few tens of meters. Therefore. Besides man-made disturbances. it is negligible compared with instrumental noise. posing no severe problem.4).B. Therefore. Eye movements and blinks are important biological sources of artefacts. 65. The peak field due to the heart's contraction. No. (1987). typically on the order of 10 m T / m . too. the total noise reduction can be 100 dB or more. 1975). V. Shielded rooms The most straightforward and reliable way of reducing the effect of external magnetic disturbances is to perform . superconducting magnets used in magnetic-resonance imaging (MRI) systems (Hutchinson and Foster. as well as of some biomagnetic signals. B. for example.2).

field sensors such as flux-gate magnetometers are employed to control the current fed to pairs of orthogonal Helmholtz coils around the room to eliminate the external field. is shown in Figs.-metal (Varpula and Poutanen. it can be reduced if the innermost layer is made of /i. Scheer. 1977.. Shaking is performed by applying a strong continuous 60-Hz field on the ferromagnetic layers so as to increase the effective permeability at other frequencies. 26. in practice. (d) symmetric series axial gradiometer. the shielding factor is comparable to that of the Helsinki room (Erne. based solely on eddy-current shielding. This room (see Fig. Nenonen and Katila.01 Hz the shielding factor is 45 dB. et al.2 0 dB increase in the shielding factor. The wall thickness in these rooms is about 5 cm. The shielding factor varies from about 50 dB at 1 H z to 80 dB at 100 Hz. (e) asymmetric series axial gradiometer. These enclosures are made of two ferromagnetic /x-metal layers with inner dimensions of 3 X 4 X 2 . With active compensation and "shaking. (c) parallel planar gradiometer. In our laboratory. 1970b). No. To make such an enclosure." a shielding factor of 59 dB at 0. The Vacuumschmelze room is a compromise between performance and price. The newest shielding method is to employ high-T c superconductors to make whole-body shields (Matsuba et aL. disturb the measurements. The expected shielding factor. Malmivuo et a/. (b) series planar gradiometer..1 1 0 dB above 1 Hz. 1985. The effect of this noise source is most significant in eddy-current shields. Nicolas et aL. based on calculations and on a 1/10-scale model experiment. At the Physikalisch-Technische Bundesanstalt in Berlin a heavy shield has been built of six layers of jti-metal and one layer of copper (Mager.r c superconducting shielding. have been built (Zimmerman. This arrangement is insensitive to a homogeneous magnetic field which im- -1(b) CD (d) (e) -1(c) C^> (f) (9) FIG. Phys. With suitable gradiometers it offers a sufficiently quiet space for practically all types of biomagnetic measurements. 1981. . (f) symmetric parallel axial gradiometer. Maniewski et aL. 1983.: Magnetoencephalography the measurements in a magnetically shielded room. more than enough for all practical purposes. Various types of flux transformers: (a) magnetometer. Active shielding. 65. below 0. A first-order axial gradiometer. Stroink et a/. The better performance at higher frequencies is due to eddy currents induced in the high-conductivity aluminum layers.. 1981). Two other commercial manufacturers have also entered the market. 9) consists of three shells made of highpermeability /i-metal. a remanent field of 5 nT has been measured inside. 1981). 2. resulting in a 1 0 . It attenuates external fields by 9 0 . 1970a). a cubic shield with inner dimensions of 2.9 1985). Several such enclosures. sandwiched between aluminum plates. although the Berlin room is superior at lower frequencies. The first shield of this type will consist of a horizontal tube of 1-m diameter and 3-m length. Several shielded rooms for basic biomagnetic research and for clinical use have been constructed commercially by Vacuumschmelze G m b H (see the Appendix for address). the shielding factor is proportional to frequency. eddy-current shielding. in which an oppositely wound coaxial coil is connected in series with the pickup coil. 1981. and Trontelj. Many rooms have been built for biomagnetic measurements utilizing combinations of these techniques. Vvedensky. there is an option for active compensation.1 H z was obtained. Mod. such as the heart. is presently not in use. external magnetic disturbances can be reduced with gradiometric coil configurations. The open end will have a set of ferromagnetic cylinders to increase the effective length of the tube. Vol. the shielding was 100 dB (Cohen. and (g) second-order series axial gradiometer. 2. mechanical vibrations of the Dewar in the remanent magnetic field and nearby noise sources.Hamalainen et a/. and the recently introduced h i g h . the improved shielding is. active compensation. The plates of each shell were welded together at their edges to allow for the flow of eddy currents which oppose changes in the external magnetic field. closed at one end. 26(a)] are rendered practically useless. being about 50 dB at 50 Hz. 1982). Naurzakov. 4 m 3 . Gradiometers In addition to or instead of shielded rooms. 7(b) and 26(d). At low frequencies. 1992). The ultimate performance of magnetically shielded rooms is determined by Nyquist current noise in the conducting walls. is 160 dB. Even inside shielded enclosures. Amuneal Corporation and Takenaka CorporaRev. superfluous when first-order gradiometers are employed. so that simple magnetometers [see Fig.. April 1993 445 tion (see the Appendix for addresses). For active compensation. Above 10 Hz. In the frequency range 1-100 Hz. This room has its walls made of three ferromagnetic moly-permalloy layers and two sheets of aluminum in the form of a 26-faced polyhedron. The first shielded enclosure for biomagnetic measurements was constructed at the Massachusetts Institute of Technology (Cohen. however. Less expensive shielded rooms can be made from thick aluminum plates. After demagnetization. Hahlbohm. 1984. four different methods exist: ferromagnetic shielding. 1988).4 m was built in 1980 (Kelha et a/.

Bnet. Mod. These sensors thus collect their signals from a more restricted area near the sources of interest and there is less overlap between lead fields.and second-order axial gradiometers are 8 and 1 5 % smaller. an adequate base line for an axial gradiometer is 1-2 times the typical distance to the source. 27. one has to make the effective turns-area products of the pickup and compensation coils equal. a better sensitivity is obtained than with a symmetric gradiometer. The dipole moment is 10 nA m. and a planar gradiometer are illustrated in Fig. 1985. No. is calculated perpendicular to the dipole along a circle passing through it. (34)]. this reduces the total inductance by a factor of 4 from the configuration in Fig. To obtain cancellation of homogeneous fields. 26(d). 1987). 27). 26(e) and 26(f). The base line of Rev. and a planar first-order gradiometer (see Fig. with modern SQUIDs. an initial imbalance of a few percent is typical. 30 mm below the surface. The locating accuracies of planar and axial gradiometer arrays are essentially the sam^ for typical cortical sources (Erne and Romani.. Figure 26(g) shows a second-order axial gradiometer in which two first-order gradiometers are connected together in opposition so that the detection coil is insensitive to both homogeneous fields and uniform field gradients. the pickup and compensation coils are connected in parallel instead of in series. 1986). Amplitudes of magnetic signals due to a tangential current dipole in a spherically symmetric volume conductor. a planar. 1983). 27. the area where measurements must be made to obtain the dipolar field pattern can be smaller. 26(e). This provides sufficient far-field rejection without severe attenuation of the signal (Vrba et ah. an asymmetric first-order gradiometer is illustrated. the magnetic field produced by the brain is sensed essentially by the lower coil only. Duret and Karp. respectively. as with conventional axial gradiometers (see Fig.446 Hamalainen et a/. April 1993 300 200 1 100 i/ / / 0 CD v x --^\ \ / \ v ^ y ^ c -100 — -200 \ • s -300 -40 -20 0 20 6 (degrees) L^ 40 FIG. cannot be avoided with gradiometric techniques. a first-order and a second-order axial gradiometer. Successful experiments with second-order gradiometers of this type have been performed in unshielded environments (Kaufman and Williamson. that noise from nearby sources. 1985. 65. 26(c). as the field would be recorded by a magnetometer.5 cm. It can be reduced by changing the effective areas of the coils . an axial first-order gradiometer. The first-order gradiometer is thus effective in measuring the inhomogeneous magnetic fields produced by nearby signal sources. an axial second-order gradiometer. Note that for the planar gradiometer the maximum signal is just above the source. respectively. planar sensors is typically shorter than that of conventional axial gradiometers. Note also that the signals of the first. 2. this may be advantageous if measurements have to be performed in very noisy environments. Traditionally. Knuutila et ah. In wire-wound gradiometers. Because the inductance of its compensation coil is smaller than that of the pickup loop. 26(b) and 7(c) have advantages over axial coils: the double-D construction is compact in size and it can be fabricated easily with thin-film techniques. the difference signal between the gradiometer loops. The parallel connection is easier to match to the signal coil of a SQUID. than that of the magnetometer. Vol. with the coils connected in series. most neuromagnetic measurements have been performed with axial gradiometers. Carelli and Leoni. Other possibilities for arranging first-order axial gradiometer coils are shown in Figs. In addition. Phys. The radius of the spherical surface where the measurements are performed is 100 mm and the dipole is assumed to be at 0 = 0°. In general. with the direction of the source current being perpendicular to that of the field gradient. offdiagonal planar configurations of Figs. This arrangement further improves the cancellation of magnetic fields generated by distant sources. In Fig. 26(d). 26(f). so-called double-D gradiometer (Cohen.: Magnetoencephalography poses an opposite net flux through the lower (pickup) and the upper (compensation) coils. If the pickup coil is close to the subject's head and the distance between the two coils (the base line) is at least 4 . the spatial sensitivity pattern (see below) of off-diagonal gradiometers is narrower and shallower. 1982. In Fig. such as the Dewar. 26). In Fig. 1979) is shown. A disadvantage of higherorder gradiometers is that to avoid reduction of the signal energy coupled to the SQUID the height of the coil may become unpractically long (see also Fig. However. and the base lengths of the axial and planar gradiometers are 60 mm and 15 mm. Therefore. which has a small inductance. off-diagonal. however. Since the maximum signal is obtained directly above the source instead of at two locations on both sides. which is probably mostly due to thermal currents in the thin metallic layers of superinsulation. calculated for a current-dipole source in a spherically symmetric volume conductor [see Eq. 27). The "signal profiles" of a magnetometer. A parallel version of this gradiometer is illustrated in Fig. whereas the more homogeneous fields of distant noise generators are effectively canceled. One should note. The double-D gradiometer gives its maximum response just over the source. further improvement requires reduction of the Dewar noise. A disadvantage is that homogeneous magnetic fields give rise to shielding currents in the detection coil that couple to neighboring gradiometer channels.

In tests. Second-order gradiometers made electronically from first-order gradiometers have also been tested successfully (Vrba et ah. a very large dynamic range and a high slew rate (the system's ability to track rapid changes) are needed for the magnetometers. Gradiometers may also be made electronically from magnetometers.Hamalainen et a/. but they detect distant noise sources. Obviously. 1989). B™{(t). a final imbalance of about 1-10 ppm can be achieved. the compensated outputs (i=x. (90) where Wtj(t) are subtraction coefficients that depend on the character of the disturbing field. Fortunately. and thus an adaptive algorithm can be established. manufactured by Dornier Systems G m b H (see the Appendix for address). the best results have been achieved with conventional heavy magnetic shields and high-sensitivity first-order gradiometers. Electronic noise cancellation As we mentioned already. The weights Wtj (t) may be estimated statistically on the basis of measured signals. A drawback is that noise in the reference channels sums up vectorially. in the analysis of most components of evoked responses and of spontaneous activity. 65. nevertheless. Duret and Karp. These compensation channels are insensitive to brain signals. Electronic balancing by applying the compensation digitally with a software algorithm offers much more versatility (Robinson. including gradiometers. and z components separately. even after careful balancing at low frequencies. For example. the reference signals subtracted from the outputs can be band-limited. Furthermore. a balance at rf frequencies is not guaranteed. 1988). 1985). 4. in the instrument at the University of Ulm. it is always safer to prevent disturbances than to compensate for them.. and z refer to a local coordinate system of the sensor. The reference channels can be employed to remove common-mode disturbances from the gradiometer output caused by imbalance. 1987). Phys. for a planar firstorder gradiometer. For example. where x> y. The system may include additional "reference" sensors measuring the three magnetic-field components and possibly some of the field gradients as well. a recording passband of a few hundred H z is typical. 1992b). Three tabs are needed to adjust the x. (89) where Cx. The main drawbacks of active noise cancellation have so far been the complicated control of channel weights. and Cz are experimentally determined imbalance coefficients and Bxef(t). 28 magnetometers are used to form 22 effective gradiometers. in order to make electronic gradiometer systems work. even a low-pass . In addition to five second-order gradiometers near the subject's head. Vol. In multi-SQUID systems. In magnetoencephalographic studies. y. So far. 1989. They are. which measures the derivative dBz/dx =GX. Cy. 1993). Still. their use in multichannel devices is too complicated. extra channels measured the three magnetic-field components and the axial derivative of the radial component. April 1993 447 ments. however. Filtering and averaging of neuromagnetic data Bandpass filtering is a simple way to reduce the effect of wideband thermal noise. Usually. Mod. Vrba et ah. A similar arrangement has also been tested at the PhysikalischTechnische Bundesanstalt (PTB) in Berlin using their 37-channel magnetometer (Koch et ah. using analog hardware (Williamson et ah. highly desirable for this application. digital bandpass filtering may be employed for adaptive frequency-dependent noise cancellation. A magnetometer with a low intrinsic noise is. Furthermore. insufficient dynamics. A magnetically shielded room has now been installed at N Y U as well. Electronic noise-cancellation schemes offer an elegant way for reducing environmental disturbances. Although superconducting tabs are suitable for balancing single-channel instruments. which reduces the total random noise. Another possibility is to place small pieces made of magnetically soft material outside of the Dewar (Thomas and Duret. the reference channels can be employed to subtract terms common to all channels and caused by the homogeneous and gradient components of disturbing fields. in general.. 2. the effective noise level of the electronically formed gradiometer is increased by a factor of Vl. the noise level could not be reduced sufficiently for some measureRev. three orthogonal field channels and one gradient channel improved the external noise reduction by 40 dB. and a too high noise level of the SQUIDs. time dependent. 3. since they are positioned far away from the subject's head. Assuming that ^ are the outputs of the reference channels. Indeed.: Magnetoencephalography with movable superconducting pieces (Aittoniemi et al. rejection of external noise in an unshielded environment improved by about 20 dB. Another possibility is to apply electronic cancellation directly to the measured signals. although many successful experiments were performed (Kaufman and Williamson. as well as in another study employing a single-sensor magnetometer and a reference magnetometer located 20 cm away (Drung. a noise reduction of 40 dB was achieved in an unshielded environment. 1991). In the latter experiment.z) of the first-order gradiometer signal channels are GcorrU) = GmeasU)_ ^ Wij( t )Vj(t) . the corrected output is G™TT(t)= G?»*(t)-CxBf(t)-CyBf(t)-CzBf(t) . For example. and B*ct(t) are the reference signals. 1991. 1978. With this system. No.y. 1983). In New York University a system was adopted where balancing was improved electronically. active compensation of external fields is a practical technique in conjunction with shielded rooms.

Although sophisticated filtering methods exist for extracting the signal from noisy data. 1981. but possibly in other leads as well. obtained with proper magnetic shielding. Phys. as will be discussed in Sec. This method can. where N is the number of averaged responses. as an example. Real signals measured from humans are not perfectly replicable. These methods are simple special solutions to the general problem of extracting a signal from noisy data. for example. r . based on continuous estimation of the signal-to-noise ratio in several frequency bands. If the spectra of the signal Ts(f) and the noise r n ( / ) are known a priori. April 1993 . low-noise sensors. is a simple and effective method against disturbances. they can be regarded as independent. for example.y 1991). 1981) The power of a signal cannot be estimated with an arbitrarily small resolution simultaneously in the frequency and time domains. The lower the ratio. 2. Therefore. Since noise and external disturbances. for obtaining reliable and stable low-noise operation. for example. using a bank of filters. It allows concave surfaces to be made in order to accommodate the subject's head.. Fiber-glass-reinforced epoxy has proven a reliable low-noise material in Dewars for biomagnetic research. in practice. u{t)=s{t) + n(t) (91) . viz.1 H z is usually employed. a generalized a posteriori "Wiener" filter should use time-dependent estimates of the signal and noise spectra. essential that all feedthroughs are rf filtered and grounding arrangements are properly made. rf filters are necessary in the flux transformer (IImoniemi et al. V. for example. a high-pass filter below 0. More sophisticated filtering and estimation procedures have been developed and. H(f. It is.. be used only in conjunction with repeating phenomena such as evoked responses. weak electric somatosensory stimuli. including the background activity of the subject. we discuss here briefly the time-varying filter (de Weerd. ( / ) and r „ ( / ) are not known. However.t)=-z f s (/. and welldesigned gradiometers are still the most important ingredients of good M E G results. Besides bandpass filtering. Not only must the sensors be brought as close to the source as possible.3 0 0 responses. M o d . by E E G leads. e. The measured output u (t) is assumed to be the sum of a noiseless signal s(t) and independent random Gaussian noise n(t). The generation of the stimuli themselves. strong rf fields may occupy its interior because the room acts as a resonance cavity. Bertrand et aLy 1990). in association with spreading depression. therefore. In the lowfrequency end. rf filters and grounding arrangements A magnetically shielded room can easily be made tight against electrical disturbances. epilepsy. 5. even helmet forms are possible. (93) Construction of the cryogenic system for a neuromagnetometer is a demanding task. the main effort should be directed towards noise reduction in the flux sensed by the SQUID. We then obtain the a posteriori "Wiener" filter simply by replacing r s ( / ) ' a n d T „ ( / ) in Eq. the filter is realized by dividing the frequency range of interest into several bands. viz. etc. the optimal linear filter to estimate the signal from an ensemble average is the "Wiener" filter with a transfer function (de Weerd.5 0 H z is adequate. To minimize the distance from the sensor coils in . Vol.3.f) -± Rev.. and by other wires acting as antennae. this is strictly true only for "ideal" signals which can be described accurately by Eq. The bands are then weighted and recombined. the more the corresponding band is attenuated. of course. Since the evoked responses are transient. averaging is a simple and powerful way of improving the signal-to-noise ratio..: Magnetoencephalography 448 filter with a cutoff at 4 0 .Hamalainen et a/. but must be estimated from the ensemble. Seppa. Dewars and cryogenics However. can also give rise to artefacts if the leads are not properly shielded and grounded. C. A practical limit for N is thus reached when the contribution of random noise is smaller than that of the extra non-Gaussian noise generated by the averaging process. However.g. (92) by their estimates f s(f) and A „ ( / ) . due to drifts. 1984. (91). In practice. This nonideality may be thought of as non-Gaussian extra noise that cannot be reduced by averaging. No. The signal-to-noise ratio of each band is estimated separately for short periods of time using. dc measurements would be useful because very slow shifts may occur. Thus their influence may be reduced by averaging typically 2 0 . 65. 1987). it is questionable whether such a time-invariant filter will produce the best results. Clean raw data. normal and alternating averages. but also the materials used in the Dewar must be nonmagnetic and the noise generated by thermally induced currents in the electrically conducting parts must be very low. because of fatigue in the subject during an experiment._with an improvement in the signal-to-noise ratio of VN. are normally not time-locked to the stimulus.D. The weights of the filter bank are adjusted on the basis of the timedependent estimates of the signal-to-noise ratio.. This is not only to suppress external disturbances but also to damp intrinsic resonances excited by high-frequency oscillations in the SQUID itself. especially for reducing power-line interference. External rf fields typically are carried inside by cables from the stimulus-producing equipment. and noise may vary with time.. Because of practical limitations. and migraine (Barkley et al. Careful elimination of ground loops.

The Josephson junction is characterized by a critical current Ic. 65. Multiple.g. In this paper we emphasize aspects that are particularly important in the design of successful instrumentation for neuromagnetic research. Furthermore. For simplicity. the readout electronics for dc SQUIDs is simpler. effective ventilation is important. 1965). The Dewar must be equipped with proper relief valves to prevent an explosion in case of sudden evaporation of the liquid following an accidental loss of the insulating vacuum. Montonen et al. I the current. Phys. The constant 3>o=/z /2e = 2 . 28(a)]. and V the voltage across the junction (see Feynman. 1989) and in textbooks (e. In practice. 1970). 1966. shown schematically in Fig. we shall consider these devices only. 28. 1965. SQUID basics. a shield made of vapor-cooled coil-foil should be used (Anderson et al. A few layers of it wrapped tightly into the vacuum space of the Dewar form "floating" thermal radiation shields. (94) becomes V R c dV dt (96) . Zimmerman and Silver.. Eq. 1975. No. Lounasmaa.2e (94) 277 K (95) where 9 is the quantum-mechanical phase difference across the junction. brought in the immediate vicinity of a human subject. Clarke. is a safety risk. glued together to form a sheet. Superconducting quantum interference devices The only device with sufficient sensitivity for highquality biomagnetic measurements is the SQUID magnetometer (Zimmerman et al. there is an additional compensation coil in series with Lp. Therefore. T. 1976) consists of a superconducting loop interrupted by two Josephson junctions [Fig. Coil-foil consists of tightly spaced thin insulated wires. there is a capacitance C and a resistance R across the junction.: Magnetoencephalography liquid helium to the signal source in the brain requires a compromise between thermal noise and the helium boiloff rate.. Previously all sensor systems for biomagnetic studies were based on rf SQUIDs because this device was simpler and cheaper to manufacture than its dc counterpart. .Ls) to the SQUID. The boiloff rate of liquid helium may be a very important practical factor in the design. (a) Schematic diagram of a dc SQUID magnetometer. Sudden evaporation of helium may rapidly generate cold gas in amounts comparable to or even exceeding the volume of the shielded enclosure.. which prevents the formation of large loops for noise currents. In a gradiometer. 1961. 1992).. the foil has electrical and thermal conductivity primarily in one dimension only. and the junction dynamics is governed by the well-known relations (Josephson. advances in thin-film fabrication technology have changed the situation: reliable low-noise dc SQUIDs can be produced now in large quantities and at lower prices than before. 1964. where the temperatures of the layers are determined by radiation equilibrium. In the resistively shunted junction (RSJ) model. Ryhanen et al.. However. Mod. for example. consider first only the uncoupled noiseless SQUID. Superinsulation is thin plastic foil with an aluminum film evaporated on one side. 1 3 liters of liquid helium at 4.. a device without the flux transformer coils Lp (pickup coil) and Ls (signal coil). In the following. 1962) I =Icsin9 D.. 2. since the Dewar is often operated in the closed space inside a shielded room. multiple layers of superinsulation and additional thermally anchored radiation shields can be employed. The external field Bext is connected via a superconducting flux transformer (Lp.. Vol.. vapor-cooled radiation shields and superinsulation cannot be used in the tail section of the Dewar below the sensors: these would lead to wide gaps between the inner and outer surfaces and to large thermal noise currents (Nenonen et al. 0 7 fWb is the magnetic-flux quantum. April 1993 449 39 _ . Its basic principles and theory of operation have been described in detail. defined by the ratio of the junction voltage to the tunneling current above Ic. a white noise level can be achieved that is more than an order of magnitude smaller than in rf SQUIDs. a period of 3 . Typically. i. A vessel containing perhaps several tens of liters of liquid helium. perhaps with a few layers of striped superinsulation foil. 1.. SQUID basics The dc SQUID (Jaklevic et al. most contemporary instruments are based on dc SQUIDs.4 days between helium transfers can be attained without an unduly massive and clumsy reservoir for the cryogenic liquid. in review articles (Koch.e.Hamalainen et al. Lounasmaa. 1974). Instead. Feynman. Furthermore. 28(a). 1989. 1966. In other parts of the Dewar. Typically.2 K is equivalent to 1 m 3 of gas at 20 °C! FIG. especially with Dewars intended for hospital use. Rev. Clarke et al. Usually R is determined by an external shunt resistor. (b) The average voltage V across the SQUID depends on the dc bias current IB and is a periodic function of the flux O a coupled to the SQUID ring. much smaller than the equivalent "intrinsic" resistance of the junction. 1974). 1989.

By changing variables to and 8(g) is Dirac's delta function. FIG. dv -h——hsinvcos<p = z . we obtain the equations of motion. The signal coil Ls introduces a parasitic capacitance Cp across the SQUID (see Fig. (99) and (ioO). SQUIDs are manufactured entirely of refractory metals (Kroger et al. This pair of nonlinear equations must be solved numerically. At the smallest bias.. 1968. For a magnetic flux 4> threading the ring.9 1984b) 2TTR2ICC (97) In the dc SQUID. dr1 dr TrpL (100) where /3 L =2ZJ C /<I> 0 . Thin-film fabrication techniques . with the tunnel barrier formed by oxidizing the N b surface. v=(6l + e2)/2 . The voltage across the SQUID is given by V=RIc(dv/dr) and the total flux by <£> = O0<p/7r. The voltage modulation is maximal at IB2 > 21c and fades away for IB3»2IC. The flux transformer couples the external field 2?ext to the SQUID. dr dr1 (99) '2. but if I»IC. April 1993 . 1928) in the shunt resistors. Stewart. this happens if the Stewart-McCumber parameter (McCumber. Since the magnetic flux threading a closed superconducting loop is conserved. the expectation values of the noise terms are E{inJr)}=E{inJr)} The majority of SQUIDs have been made from Pb alloys and pure Nb. which substantially complicates the SQUID dynamics but which can be taken into account in numerical analysis by modifying Eqs. The transition between these two regimes is hysteretic if the damping due to the resistance R is small. 1968. No. IBl<2Ic. and they are separated by "artificial" tunnel barriers such as A1 2 0 3 . a shielding current Ish=zBextAp/(Lp-\-Ls) is induced [see Fig.+!-</. 1987): both electrodes are made from N b or N b N .^-+cosvsin<p+—— (q> — ^ f l ) = 0 . Gurvitch et aL. cp=(dl-e2)/2. This can be taken into account in the circuit analysis by adding noise terms inyV(r) and int<p(r) to the right-hand sides of Eqs. Mod. 1985. Often the base electrode was fabricated from Nb. the SQUID stays for part of the cycle in the zero-voltage state: the sum of the current circulating around the SQUID ring plus half of the bias current exceeds the junction critical current only at certain values of the flux.y 1989). Ketoja et al.• A ) (98) where L is the inductance of the ring and <f>c is the externally applied magnetic flux. 28(a)]. the junction is in zero-voltage regime. respectively.. Braginski. and (pa=7r<&a/<$>0. with Ap being the effective area of the pickup coil. i =I/2Ie 2.: Magnetoencephalography 450 If the current / fed through the junction is smaller than J c . 29. 65. The flux <E>a =MIsh is coupled to the SQUID through the mutual inductance M. The A1 2 0 3 and MgO barriers are superior to conventional materials: they are more stable. the quantization relation obtained by integrating the quantummechanical phase around the SQUID ring in the presence of a magnetic field reads 3> = f>-*2>=*. In particular. except for a window defining the junction area. Nowadays.. 2. Phys. The electrodes were usually separated by a layer of Si0 2 . In Fig. 1983. 28(b). or amorphous Si. 29). the bias current IB is the sum of currents through the two junctions. Simplified structure of a planar thin-film de SQUID with an input coil. Vol. the critical current is Signal coil spiral dc SQUID washer Current nodes Distributed parasitic capacitance " Josephson junctions =E[inJr)inJr)}=0 Counter electrode and E{injT+^injT)}=ElinJr+i)inJr)} = 2irkBT&(g)/(Ic4>0) Rev. (99) and (100) (Ryhanen et al. Here. It is common to assume that there is a white-noise source generated solely by thermal fluctuations (Nyquist. 13 —-¥+ -. . Raider. and T=2wRIct/<&0 . The distributed parasitic capacitances across the SQUID washer and the locations of the current nodes for the microstrip resonance are indicated. The quantization relation (98) ensures that the response of the SQUID to magnetic flux is periodic in the applied flux <f?a with the period <f>0. and the top electrode made of a Pb alloy. IB=Il+I2The RSJ equation (96) is applied to both branches separately. V^RI (running state).Hamalainen et a/. the voltage has been sketched schematically as a function of the applied flux for three different flux bias currents. typical structures containing layers of both metals. ft d2v . 1981. MgO.

typically at a few M H z . A microwave transmission line. 1987). At the same time. Muhlfelder et aL. Phys. and lift-off. Computer simulations and experiments have shown (Ryhanen. which is typically above the working point of the SQUID. 1981) is deposited over the relatively wide conductor loop. As a result. April 1993 451 only a few picofarads. and the junction capacitance is small. and Seppa. These transitions are enhanced by the presence of resonances. Advances in semiconductor processing technology have. feasible critical current densities. the resonances should be properly damped. No. 1988.g. 1992). This is because thermally activated transitions occur between different voltage states. 1991. 1987. and specific junction capacitances. for Cp < C. Knuutila et aL. with the junction barrier first formed on the entire wafer and patterned afterwards. This imposes a limit on the number of turns nt in the signal coil.. (101) cannot be reached in practical devices: many problems appear because of the parasitic effects introduced by the signal-coupling circuits. and self-aligning "fullwafer sandwich" methods. Since optimization usually leads to very small dimensions (see Sec. C the junction capacitance. the length of the signal coil is first selected by adjusting the number of turns to be high enough to lower the transmission-line and the LC resonance frequencies well below fj. Ben-Jacob and Imry. however. forming the SQUID inductance (see Fig. 1985.Hamalainen et aL: Magnetoencephalography uniform and reproducible. increases the noise significantly if it is not carefully reduced (Knuutila. However. (102) In modern SQUIDs with small junctions. has been thoroughly analyzed (Tesche and Clarke. it is clear that improvements in fabrication technologies result in better SQUID performance. Ketoja et aL. 65. its Q value is enhanced by the signal coil. 1985. the leakage current at voltages below the gap is negligible. 1988. Seppa. The ultimate performance given by Eq. be reasonably well controlled by properly adjusting the geometry. 29). 1984a. The signal coil also gives rise to a variety of resonances. The SQUID washer itself acts as a A / 2 resonating antenna. of course. All these resonances can be hazardous to the operation of the SQUID. Knuutila et aL. having its resonance frequency lower than the k/2 resonance of the loop. Here L s l i t is the parasitic inductance of the SQUID ring. For example. is formed between the SQUID washer and the stripline. Vol. The various resonance frequencies are thus determined by the dimensions of the SQUID washer and by its number of turns. 3.. which reduces radiation losses.e. 1988.. the proceedings of the latest SQUID conference (Koch and Liibbig. since a small inductance is needed for low-noise performance.3). Clearly. including minimum linewidths and conductor separations. and the numerical constant y ^ l 2 according to simulations. the output inductance L. 29). called SQUID washer. However. Proper optimization of SQUIDs is. Ryhanen et aL. For a typical washer with 1-mm circumference. the spiral-like signal coil (Jaycox and Ketchen. the X/2 resonance frequency is on the order of 100 G H z . Sensitivity of the dc SQUID In the literature. Even the relatively low-frequency resonance of the flux transformer circuit. well-defined and reproducible patterns with smaller dimensions have become possible. Ahonen. 1989). between zero-voltage and running states. the problem of too low JL. Sputtering. 1991. only possible if the critical parameters can be controlled accurately. (103) to which the magnetometer or gradiometer coils are connected. 1977. 1989). Ideally.=^(L-LsHt)+Lstrip . i. The well-known result for the optimal energy sensitivity is z = ykBTV~LC . where the line leaves the SQUID washer and the wave impedance changes abruptly (Hilbert and Clarke. Ryhanen et aL. (102) can only be reached if all the above-mentioned conditions are fulfilled and the resonances are properly damped.. 2. 1989). (101) where L is the inductance of the SQUID. 1986). C can be Rev. remains small. and Tesche. wet etching.D. required film thicknesses. 1981. 1992). 1984. the limit given by Eq. can be circumvented with the use of an additional inductancematching transformer (Muhlfelder et aL. plasma etching. Drung. The distributed parasitic capacitance across the washer is seen as an additional capacitance Cp across the SQUID. Fabrication technology poses a number of constraints. In addition. a compromise is necessary to prevent Cp from becoming too large. The most probable standing waves in the stripline have current nodes at the ends of the signal coil (see Fig. mostly associated with the slits in the washer near the junction area (see Fig. The resonance frequencies can. if they occur near the Josephson oscillation frequency f3 = F / O 0 at the working point. All these technology-dependent constraints limit the range of possible parameter values of the SQUID. for example. Cantor. If necessary. 1987. 29). V. Ketchen. 1992) that the minimum-energy resolution. Its inductance and the lumped capacitance form an LC resonance circuit. including their stray in- . the sensitivity of the autonomous SQUID. The minimum linewidths that can be achieved with the new methods are typically a few micrometers. however. e. is then E = ykBT\/L(C+2Cp) . Ryhanen et aL. and L strip i s the inductance of the conductor strip in the signal coil. a small size is advantageous. are replacing older techniques like evaporation. so that Cp easily dominates. a SQUID without any signal-coupling elements. which may lead to substantially increased noise (Seppa. Koch. Ryhanen and Seppa. made submicron dimensions feasible in SQUID structures as well (Imamura and Hasuo. Mod. For recent developments see. Optimization of SQUID structures should take into account all signal-coupling circuits.

5X 10"-" 6 O 0 /v'Hz at 0. however. can be reached (Cantor et al.1 Hz (Ketchen et al. Unfortunately. Using relatively large-area junctions. For a coupled SQUID with a highinductance signal coil and with flux-modulation electronics (see Sec. even the pickup coils should be brought into the analysis. 30. 4. the corresponding field noise is (E {•} denotes the expectation value) Frequency (Hz) FIG. In a gradiometer (see Fig. In principle. 1983. the flux noise of this uncoupled SQUID was 1 X 1 0 ~ 6 < 1 > 0 / V H Z . Rev. Example of the flux-noise spectrum of a SQUID as a function of frequency. At 0. 1987). All SQUID elements are potential sources of 1 / / noise.1 . Lcomp. be eliminated by proper electronics (Simmonds. 1993). Mod. However. 1980. 1991). are a substantial source of \/f noise. No.E. 1993). 1986. The difference of the fluxes threading the pickup and compensation coils. At 1 Hz. 3>net. 1987). The low-frequency flux noise entering the SQUID through the motion of flux lines trapped in the loop materials or in the junctions cannot be reduced by electronic means. 1986. Vol.Hamalainen et al.: Magnetoencephalography 452 ductances and parasitic capacitances. noise values of 2. Optimal energy resolution is achieved when PL=2LIc/^0^2. V. however. Therefore.. the SQUID can be optimized as a separate unit. If O rt is the flux noise in the SQUID. including the junctions. and ks is the coupling coefficient between the signal coil and the SQUID. The 1 / / noise is dominant below 20 Hz. where \/f noise begins to dominate over white noise. 1986). Drung et ah. The straight lines show a fit to a simple model consisting of \/f noise plus white noise only. the flux coupled to the SQUID by the transformer is VLLS *. The ultimate limit of the low-frequency noise cannot be estimated on the basis of existing models. In addition. Phys. 26). 30.2X 10_7<I>o/v/Hz. caused by electron trapping and release in the tunnel barrier (Wakai and van Harlingen. the critical current fluctuations have been found to scale inversely proportionally to the junction area (Seppa et ah. In many SQUIDs the crossover point. Typical experimental values. 1989). in existing neuromagnetometers the transformer usually has been realized as a wirewound coil coupled to an available SQUID. The fact that critical current fluctuations scale with the junction area presents the possibility of optimizing the performance with respect to 1 / / noise as well. 6X I O ^ O Q / V H Z in dc SQUIDs with high-quality tunnel junctions and equipped with special electronics for the elimination of critical current fluctuations have been measured (Foglietti et ah. Rogers et ah. Cantor et al. to a large extent. Lp • ^comp ' Ls ~T~Ll Here Lp. is the number of turns in the pickup coil and .1 Hz. and Ls are the inductances of the pickup. 1988. Foglietti et ah. An example of such a noise spectrum is shown in Fig. Optimization of flux transformer coils The flux transformer should be optimized together with the SQUID. The sensitivity of a magnetometer is defined as the local homogeneous field in the pickup coil that would produce at the output of the system a signal equal in magnitude to the system noise over unit bandwidth. and signal coils. 1991). depends on the field distribution. The SQUID inductance is L. 65. the SQUID loop itself. shunt resistors. Lw denotes the inductance of the connecting wires. 1988).I).6 and (3C=2TTR2ICC/Q>^0. and the signal circuits. From Gronberg et al.. energy sensitivities in the range of 10~ 3 2 Js —h> corresponding to a flux noise well below 1 0 ~ 6 O 0 / ^ / H z . 1991) have been obtained. and the white-noise level above 1 kHz is 2. none of the present theories can e CD CO O 10" 1 10 1 02 103 104 explain the invariance of the 1 / / spectral form. thus the only free design parameters have been the dimensions and numbers of turns of the pickup and compensation coils. appearing in a wide variety of physical systems (Weissman. provided that the resonances are under control (Knuutila et ah. compensation. Fortunately. (104) by setting Lcomp=0. Koch et ah. 2. Their contribution can. (1992). Low-frequency fluctuations of junction critical currents. In practice. shifts towards higher frequencies when the white noise is being reduced (Wellstood et ah. the origin of the \/f noise in SQUIDs is still a mystery. are near 10~ 5 4> 0 /1/Hz. The corresponding expression for a magnetometer is obtained from Eq. April 1993 (104) ^r n E\B2Axn = V2(Lp+Lcomv+Ls VLsti +Lm) PAP E[*l 1/2 2L (105) where n. a coupled SQUID with a flux noise of 5X 1 0 _ 7 O 0 / v / H z at 1 Hz without electronic noise reduction has been realized (Seppa et al. respectively. Low-frequency applications of SQUIDs are limited by the 1 / / noise which is associated with the quality of the device itself..

Hamalainen et a/.: Magnetoencephalography
Ap is their average effective area. If we assume that the
SQUID and its signal coil are of fixed design, only the
pickup and compensation coils can be adjusted. The
inductance-matching condition,
Ls=Lp+Lcomp+Lw

,

(106)

maximizes the transfer of energy from the pickup loop to
the SQUID. In practice, however, the minimum external
field sensitivity, obtained by varying the coil parameters,
does not generally fulfill this condition (Ilmoniemi et aL,
1984). In particular, this is true if the SQUID and the
pickup coil are considered together (Ryhanen et aL,
1989). Careful optimization reveals interrelationships between parameters not taken into account by the simple
analysis, e.g., screening effects. Furthermore, the diameter and the length of the pickup coil are often determined
by other considerations, like the spatial selectivity and
the distribution of the field to be measured, and the dimensions of the compensation coil are fixed by the gradiometric balance condition and by space limitations.
For example, assuming that the diameters of the pickup and compensation coils have been selected, the optimum number of turns np in the pickup coil is found by
differentiating Eq. (105), where the total flux coupled to
the pickup and compensation coils is assumed to be proportional to np (Ilmoniemi et aL, 1984), giving,

Lp+Lamp+Ls+Ll-np^-(Lp+Lc+Ll)

=0 .

(107)

Therefore, the inductance-matching condition, Eq. (106),
maximizes the sensitivity for tightly wound coils only. In
general, no simple analytical expression exists for the
dependence of the inductance on np. As a result, the optimum number of turns and other coil parameters must
be determined numerically. The feasible maximum dimensions must be specified as constraints, often dictated
by the space available in the Dewar.
Enlarging the pickup coil increases the field sensitivity
but, at the same time, the loop integrates flux from a
larger area. Thus the coils can no longer be approximated by point magnetometers. In practice, this effect becomes significant only when the coil diameter exceeds the
distance to the source (Romani et aL, 1982a; Duret and
K a r p , 1984). An increased length of the coil, i.e., less
tight winding, allows more turns without an excessive increase of inductance. The sensitivity improves, but the
signal becomes weaker, because the distance from the
source is increased. For cortical current dipoles these
two effects tend to cancel, leading to a broad maximum
in the signal-to-noise versus coil-length curve (Ilmoniemi
et aL, 1984; Knuutila, Ahlfors, et aL, 1987).
A simple formula has been proposed for optimizing
flux transformers; it takes into account the abovementioned effects (Duret and Karp, 1984; Thomas and
Duret, 1988). The figure of merit is defined as

Rev. Mod. Phys., Vol. 65, No. 2, April 1993

453

where <J>SQ is the net flux coupled to the SQUID, calculated by assuming a certain source geometry, e.g., a
current dipole in a conducting sphere. Here, B0 is the
field at the center of the loop nearest to the source, and
ASQ is the coupling area of the SQUID. This figure of
merit also takes into account the base length via the net
flux <DSQ.
5. Examples of dc SQUIDs for
neuromagnetic applications
dc SQUIDs suitable for neuromagnetic measurements
have been reported by several groups. They were all fabricated using thin-film techniques and incorporate a
densely coupled signal coil (Jaycox and Ketchen, 1981)
with an output inductance high enough for use with
wire-wound coils; the flux noise is less than
1 0 _ 5 4 > 0 / v / H z . We mention here a few illustrative examples, and the list is by no means complete. Integrated
structures, incorporating a thin-film pickup coil as well,
will be discussed in the next section.
One of the first all-thin-film dc SQUIDs used successfully in neuromagnetic research, at the Helsinki University of Technology, was the Nb/NbO^/Pb-alloy-based device fabricated by IBM (Tesche et aL, 1985; Knuutila,
Ahlfors, et aL, 1987). The SQUID features a 0.7-/iH signal coil. When it is operated in a practical setup with
flux modulation and a flux-locked loop (see Sec. V.E.I)
and connected to a gradiometer coil, a flux sensitivity of
4 - 5 X I O ^ O Q / ^ H Z is reached, provided that resonances
are properly damped (Knuutila, Ahonen, and Tesche,
1987). The onset frequency for the I / / n o i s e is below 0.1
Hz.
A coupled dc SQUID with even smaller white noise
and better low-frequency properties was later made by
IBM (Foglietti et aL, 1989). It employs a modified version of the fabrication process. These SQUIDs are used
by the neuromagnetism group at the Istituto di Elettronica dello Stato Solido in Rome in a 28-channel system
(Foglietti et aL, 1991).
The use of new artificial tunnel barrier materials,
A1 2 0 3 and MgO, is a general trend at present (Daalmans
et aL, 1991; Dossel et aL, 1991; Houwman et aL, 1991;
Ketchen et aL, 1991; Koch and Liibbig, 1992). The quality and properties of SQUIDs processed with these techniques have proven superior over conventional oxide or
other deposited-barrier structures. For example, an allrefractory N b / A l 2 0 3 / N b fabrication process was introduced by IBM, resulting in SQUIDs with a flux noise less
than 2.5X10~ 6 <S> 0 /VHz at 0.1 H z and a white noise of
0.5 X 1 0 ~ 6 < V V H z (Ketchen et aL, 1991).
The Technical Research Center of Finland has made
SQUIDs with low \/f noise (see Sec. V.D.3). Their performance was optimized in the frequency range of 0.1 H z
to 1 kHz. The sensor, fabricated using a 10-mask-level
N b / A l - A l O ^ / N b process, has 25-/xm2 junctions with
J c = 50 /xA. The 7-pH SQUID inductance is matched to
an output inductance of 320 n H by means of an inter-

454

Hamalainen et air. Magnetoencephalography

mediate transformer. Resonances have been carefully
eliminated, resulting at best in a white flux noise of about
1 X 1 0 ~ 7 O 0 / v / H z , corresponding to the intrinsic energy
resolution of 4.7A (Seppa et al., 1993).
dc SQUID sensors are also available commercially.
The first manufacturer was BTi (Biomagnetic Technologies, Inc.; see the Appendix for address), which offered a
hybrid SQUID with thin-film junctions, a toroidal cavity
machined of N b , and a wire-wound signal coil. The BTi
SQUID has a white-noise level of 1X 1 0 ~ 5 O 0 / V H z
above 1 Hz, measured using a special bias-current and
flux-modulation technique to reduce the contribution of
critical-current fluctuations to the low-frequency noise.
Presently, all-thin-film structures are also available,
for example, from Friedrich-Schiller-University, from
2 G / H y p r e s , and from Quantum Design (see the Appendix for addresses of these suppliers). Typical white noise
in these SQUIDs is 5 - 1 0 X 10~~6<l>0/VHz, with the \/f
corner frequency at 0.1 - 1 Hz.

6. Integrated thin-film magnetometers
and gradiometers
To avoid the problems of wire-wound gradiometers,
which are expensive to manufacture in large quantities
and have only a modest balance, planar integrated sensors have been investigated (Erne and Romani, 1985;
Knuutila et al., 1985; Carelli and Leoni, 1986; Bruno and
Costa Ribeiro, 1989). The main advantages of thin-film
gradiometers are their compact structure and excellent
dimensional precision, which provides a good intrinsic
balance.
Two main methods have been used for coupling the
signal to the SQUID in integrated sensors. In one type,
the pickup loop is a part of the SQUID structure. In the
first reported integrated gradiometer (Ketchen et al.,
1978) the SQUID ring shared a common conductor with
a larger field-collecting loop, thus forming an inductively
shunted dc SQUID device. The coupling of the flux to
the SQUID was poor in this construction because of a
mismatch between the inductances of the pickup loop
and the SQUID. For a pair of 24 X 16 mm 2 coils, connected in_ parallel, a gradient sensitivity of 20
fT/(cmV / Hz) was achieved.
Later, several other experimental first- and secondorder gradiometers, in which the pickup loop formed an
integral part of the SQUID itself, were reported (de Waal
and Klapwijk, 1982; Carelli and Foglietti, 1983; van
Nieuwenhuyzen and de Waal, 1985). The early integrated magnetometers have been discussed in several review
articles (Donaldson et al., 1985; Ketchen 1985, 1987). A
successful magnetometer has been reported in which the
SQUID coil is divided into eight loops coupled in parallel
to reduce the total inductance (Drung et ah, 1990;
Drung, 1992a). The field sensitivity of the 7 . 2 X 7 . 2 mm 2
device is 2.3 f T / V H z above 500 H z and 5 f T / V H z at
1 Hz.
Rev. Mod. Phys., Vol. 65, No. 2, April 1993

Another design possibility is to connect the pickup
loop to the SQUID via a signal coil (Jaycox and Ketchen,
1981), which allows good coupling and the necessary inductance matching. The first such fully integrated magnetometer, built for geomagnetic applications, was reported by Wellstood et al. (1984). Their device had a
pickup-coil area of 47 mm 2 , and the sensitivity was
5 f T / V H z above 10 H z and 20 f T / V H z at 1 Hz. Detectors based on the same signal-coupling principle have
been reported by groups working at Karlsruhe University
(Drung et al., 1987), at the Electrotechnical Laboratory
in Tsukuba (Nakanishi et al., 1987), and at the Helsinki
University of Technology (Knuutila et al., 1988).
In the Helsinki device, optimization of the whole
structure as a single unit was done for the first time. The
predicted energy sensitivity o f l . 2 X 1 0 ~ 3 1 J s was reached
within a factor of 2. Although the integrated prototype
sensor showed excellent sensitivity and tolerance to
external disturbances, technical difficulties in the fabrication process kept the yield low. Thus the 24-channel system (see Sec. V.F.5.a), which was completed in 1989, had
to be realized with wire-wound coils.
Integrated magnetometers and gradiometers based on
the same optimization principles have been reported by
Cantor et al. (1991). Their 4 X 4 mm 2 magnetometer has
a white flux noise of 5 X J X ) _ 7 O 0 / v / H z , corresponding to
a field noise of 3 f T / V H z and a coupled energy resolution of 1 . 5 X 1 0 - 3 2 Js measured with a SQUID
preamplifier. The sensitivity of the 11X17 mm 2 gradiometer is 1.1 f T / ( c m V H z ) .
The Technical Research Center of Finland has also
produced integrated SQUID magnetometers and gradiometers designed according to the principles discussed
above. The flux noise of a SQUID magnetometer with
3-/im 2 junctions, a 12-turn signal coil, and a SQUID of
11-pH inductance is 2.8X 1 0 ~ 7 O 0 / V H z at 1 kHz and
1.5X 10~ 6 ^> 0 /VHz at 1 H z (Gronberg et al., 1992). The
SQUID is connected to a pickup loop with an inductance
of a few hundred n H by means of an intermediate transformer; the measured white-noise performance corresponds to a gradient sensitivity of 1 f T / ( c m V H z ) . The
device is fabricated using an eight-level process which
employs a N b / A l - A l O ^ / N b trilayer.
7. High-Tc SQUIDs
The discovery of superconducting ceramic compounds
with critical temperatures above 90 K has raised hopes
for practical high-temperature SQUIDs. However, the
unfavorable mechanical, chemical, and crystalline properties of high- Tc materials complicate the development of
good-quality films and wires and well-controlled Josephson junctions. For instance, the extremely short coherence length results in low pinning forces, allowing flux
creep. The rough surface structure complicates construction of tunnel junctions; granularity in bulk samples
and in poly crystalline thin films leads to phase-locked
Josephson junctions between the grains, resulting in flue-

Hamalainen et a/.: Magnetoencephalography
tuations of current paths and lower critical current densities than in conventional superconductors. In addition,
many h i g h - ^ materials are very sensitive to moisture
and air. For a recent review, see the article by Braginski
(1992).
Nevertheless, high-7^ SQUIDs are being investigated
enthusiastically in several laboratories. As an example,
an uncoupled energy sensitivity of 1.5X10~ 3 0 Js, or
6 X l 0 " 6 O 0 / v / H z above 10 k H z and 1 . 2 X 1 0 - 2 8 Js, or
6 X l 0 ~ 5 O 0 / v / H z at 10 Hz, both at 77 K, has been
achieved with a 60-pH grain-boundary Y B a 2 C u 3 0 7 _ s
(YBCO) dc SQUID (Gross et al, 1990, 1991). The
lowest uncoupled energy sensitivity, so far reached, is
3 X 10~ 3 1 Js at 77 K and 71 k H z (Kawasaki et a/., 1992).
Flux transformers made of high- Tc materials have
been investigated as well (Dilorio et al., 1991; Wellstood
et al., 1991), but they seem to produce excessive lowfrequency noise, especially in coils made of films with a
low critical-current density. Vortex motion in films is
sensed by the SQUID in two ways: directly and indirectly via screening currents induced in the flux transformer.
This was clearly seen in the experiments of Wellstood
et al. (1991), where an YBCO transformer on a separate
substrate was sandwiched with a conventional l o w - r c
SQUID chip. The flux noise at 1 Hz, with the transformer coil at 60 K, was at least two orders of magnitude
higher than that of the bare SQUID, resulting in a field
sensitivity of 0.9 p T / ^ H z . When the loop was cut and
the indirect contribution thus eliminated, the noise
dropped by a factor of 10.
At frequencies that are important for neuromagnetism,
all high-!r c SQUIDs studied so far have much_larger
noise levels than expected according to the VT law,
which is based on thermal fluctuations alone. Well above
several kHz, however, a good agreement has been found
with the theory for white noise caused by Nyquist fluctuations in resistive shunts (Gross et al., 1991). The
enhanced flux creep, resulting in excess noise, is a consequence of the short coherence length, an inherent property of high-T c materials (Ryhanen et al., 1989). However,
other mechanisms contribute as well and, on the whole,
the origin of the excess 1 / / noise still remains unknown.
Future development of reliable h i g h - r c SQUIDs depends especially on the solution of two problems: (1)
How to fabricate reliable, reproducible, and wellcontrollable junctions, and (2) how to avoid excess lowfrequency noise. For construction of practical devices,
however, high-!Tc SQUIDs alone are not enough. Other
components, such as superconducting wires and contacts,
as well as thin-film flux transformers, must be made.
E. dc SQUID electronics
Electronics for SQUID magnetometers must meet two
fundamental requirements. First, the periodic response
of the SQUID to applied flux, i.e., the <!>— V characteristics (Fig. 31), must be linearized. This is achieved by
operating the SQUID in a "flux-locked-loop" mode,
Rev. Mod. Phys., Vol. 65, No. 2, April 1993

455

where negative feedback is used to keep the working
point of the SQUID constant. Second, the electronic circuits must be matched to the SQUID output impedance
of only a few ohms. The optimal signal impedance
(Netzer, 1974, 1981)
Zopt=en/in

(109)

of typical preamplifiers is several kilo-ohms at least.
Here, en and in are, respectively, the voltage and current
white-noise spectral densities of the amplifier. Therefore,
impedance matching is essential.
Several readout schemes have been proposed to meet
the requirements, and they have been demonstrated
experimentally as well. So far, however, only fluxmodulation electronics and direct readout have been used
in practical devices. In addition to them, we shall discuss
digital readout techniques.
1. Electronics based on flux modulation
In applications where a SQUID monitors a lowfrequency magnetic field, the most commonly used technique is flux modulation (Forgacs and Warnick, 1967).
This method is illustrated in Fig. 31. Square-wave modulation of O 0 / 2 peak-to-peak amplitude is applied to the
SQUID, and the signal is detected by a demodulator circuit. The output at the modulation frequency is proportional to the deviation of the low-frequency flux from a
half-multiple of O 0 . The phase detection scheme helps to
eliminate some sources of low-frequency noise, such as
electromotive forces, drifts in the junction critical
current and in the SQUID parameters, and 1 / / noise
from the preamplifiers. The detected signal is usually fed
back to the SQUID ring through a resistor and a coupling coil. Since the high-gain feedback loop tends to
O 0 /2
V

;* i 1
FIG. 31. Flux-to-voltage (<t> — V) characteristics of a currentbiased dc SQUID. A modulation of the flux through the
SQUID appears as an output voltage at the same frequency.
When the working point, indicated by dashed lines, is at a peak
or a valley, only the harmonics are seen. In the flux-modulation
scheme, the voltage detection is phase sensitive and locked to
the fundamental modulation frequency.

for example. No. Under quiet conditions. as well. with effective reduction of low-frequency noise. With an additional room-temperature transformer. comod the angular frequency of the flux modulation. The synchronous detector contains a switching circuit. Here co is the angular frequency. The modulation frequency must be sufficiently high to avoid \/f noise from the preamplifier. 1987). A transformer is preferable. Bias-current modulation. a conventional PI controller is sufficient to ensure proper operation. current sources prevent conductance fluctuations in wires and contacts from adding noise when the system is operated with large feedback currents. 1988). Using a tuned transformer with a transfer ratio m and a quality factor Rev. yielding a feedback gain G(co)^jcomod/co. By increasing the frequency dependence of the feedback gain. The feedback electronics can lock the intrinsic flux at any multiple of 3> 0 /2. but low enough to exclude input current noise which increases drastically with frequency. can be recorded with a single SQUID. A dc SQUID readout circuit is illustrated in Fig. The electronics is operated at 100 kHz. Unfortunately.. The low output impedance can be increased by feeding the signal through a cooled inductor into a capacitor in parallel with the preamplifier (Clarke et aL. 1988). but in SQUIDs with complex 4>— V characteristics. Thus the periodic SQUID response is converted into a linear one. In general. an integrator. 1975). Rillo et aL. A loop filter with 8 = 1 . modulated at different frequencies. the impedance seen by the preamplifier is Zs^(RD+R)m2Q2 . Many switching transistors with a large gate area provide excellent noise characteristics for reasonable source impedances and are ideal for dc SQUIDs. The slew rate could be improved. If the wideband noise extends to frequencies well above the . or by using an ordinary tuned transformer (with or without a ferrite core) immersed in liquid helium. 1980). one may increase the bandwidth even further (Wellstood et aL. 2. The realization of such a filter requires. (111) where co0 is the resonant frequency and Ct the effective capacitance across the secondary of the transformer. the feedback circuit must be designed with care. 1983). The cooled input circuits are designed to transform the 10-H SQUID impedance to 4 kft. since an increased output capacitance reduces the bandwidth of the reactive circuit. some loss of sensitivity may result. However. and j = V — 1. Giffard. the voltage across several series-connected SQUIDs. and a sample-and-hold circuit which can effectively attenuate the modulation frequency and its harmonics without introducing a phase shift at signal frequencies. 1972). 65. Only the most important features will be mentioned here. The feedback loop may become unstable if the gain is very high and if the output of the demodulator circuit is not filtered before feeding the signal back to the SQUID ring. 32 (Knuutila et aL. and a demodulation circuit at room temperature can separate the different channels (Shirae et aL. Since the dynamic range of the dc SQUID may be extremely high. circuits presented in the literature increase the noise level because of the aliasing effect. Koch et aL. Phys. The output of the synchronous demodulator is fed via the PI controller and a transconductance amplifier (buffer) back to the modulation coil. a higher slew rate can be achieved. Evidently. Shirae et aL. Many schemes. The maximum rate of flux change (slew rate) is limited to the order of ^o/como^ but loop stability is easily attained (Giffard et aL. 1988. In multichannel applications the electronics should contain a special unit to rebalance any of the SQUIDs after loss of lock. 1972. however. In dc SQUIDs the increase of amplifier noise with frequency limits the range of modulation frequencies. G{co)^(jcomod/co)8. In Fig. Kuriki et aL. a strong external pulse may kick the system from one stable point of operation to another. the feedback voltage is proportional to the external flux.456 Hamalainen et aL: Magnetoencephalography maintain a constant flux in the SQUID. however. but stability against variations in loop parameters is then impaired (Giffard. inside a magnetically shielded room. complicates the electronics and therefore is not a very tempting solution in devices where the simultaneous operation of a large number of SQUIDs is required. by increasing the modulation frequency up to 500 kHz. The Q value is Q^Zsco0Cto. The signal from the SQUID is fed through a cooled resonant transformer into a preamplifier consisting of two parallel FETs (Toshiba SKI46). Usually the feedback loop contains an integrator (proportional plus integrating = PI controller). 1984). 1990) produces a phase shift of 135°. the better the system can screen the SQUID loop against an external flux jump. April 1993 Q. 1986. Vol. The rapid progress in instruments with many channels has given much impetus to the development of integrated and cooled readout circuits with multiplexing (Furukawa et aL. In this scheme. 1988). 1986. A transformer is needed to match the SQUID to the optimal input impedance of a few kilo-ohms for a lownoise J F E T amplifier. 1989). 1980. whose noise temperature is as low as 2 K at 100 kHz. Mod. many components. 32. Drung et aL. (110) where R is the resistance of the transformer primary and RD is the dynamic output impedance of the SQUID. the higher the feedback gain.Zopto0Ct . which is optimal for this amplifier. The increase of the open-loop gain is ultimately limited by the modulation frequency. which is independent of the amplifier gain. Equations (110) and (111) show that it is possible to select independently both the impedance scaling and the bandwidth. 5 (Seppa and Sipola. SQUID electronics operated in the lock-in mode have been extensively discussed in the literature (Giffard et aL. 1980. resulting in a sharp transition in the output voltage. well-designed lock-in electronics does not increase the flux noise of the system. have been described (Foglietti et aL. The 1 / / noise in SQUIDs can be reduced by reversing the bias-current synchronously with flux modulation (Simmonds.

Hamalainen et a/. 1991). For more details. modern operational amplifiers with a voltage noise of only 1 n V / V H z down to a few Hz. and between the feedback coil and the SQUID. Furthermore. Mod. Since there is no need for flux modulation or impedance transformation with reactive components. noise performance may be improved even more. No.2 K 457 Outside the shielded room Phaseshift compensation Flux modulation Biascurrent control Reset and autozero Manual reset * Main electronics SQUID FIG. are shown as well. 65. Vol. Mutual inductances M and Mfy between the signal coil and the SQUID. Block diagram of dc SQUID electronics based on flux modulation and phase-sensitive detection (Knuutila et aL.. a flux noise of 5X10~ 7 <I> 0 /v/Hz above 1 k H z and 2 X ###BOT_TEXT###~6<S>0/VWz at 1 H z (Seppa. have changed the situation. This effect can be avoided by placing extra filters between the SQUIDs and the readout electronics. IB and VB are the bias current and the bias voltage. high-frequency noise is mixed down to lower frequencies. With additional optimized feedback. . see text. but the cooled multichannel unit then becomes rather complex. a. and Lp and Ls the inductances of the pickup and signal coils. the bandwidth of the electronics can be increased substantially. April 1993 U SQUID R I out FIG. However. Direct readout The fact that the preamplifier noise at low frequencies has exceeded that of the SQUID itself has effectively prevented the use of direct readout of the SQUID voltage. sampling frequency of the multiplexing. noise in the flux-locked loop was decreased to 6. and Koch.: Magnetoencephalography 4. 1988). G its gain. 2. Phys. The compensation coils of gradiometers were omitted for simplicity. with amplifier noise cancellation. Additional positive feedback A schematic diagram of a direct readout circuit with positive feedback and current bias is shown in Fig. 33. Drung. Successful operation of a high-gain SQUID with a low-noise operational amplifier (Linear Technology LT 1028) has been demonstrated. Iout the feedback current of the amplifier. either with additional positive feedback (Drung et aL9 1990) or with amplifier noise cancellation (Seppa et aL.4X 1 0 ~ 7 O 0 / V H z above 1 k H z (Ryhanen. respectively. Rev. 33(a). respectively. and SQUIDs with a large dV/d<£>. 1992) have been reached. Cantor. 1991). dc SQUID electronics based on direct readout: (a) current bias with additional positive feedback and (b) voltage bias with amplifier noise cancellation. out 2. 32. Reference measurements with a second dc SQUID as a preamplifier have confirmed that the intrinsic SQUID noise can be achieved with both noise-reduction methods. respectively. Vout is the output voltage. Similarly.

we obtain dV 3<D * « . exactly in the way as the amplifier noise. since the flux-to-voltage characteristics are skewed. Rev. As previously. by replacing the resistor with an adjustable element. With voltage bias. the linear region of the SQUID is reduced. Although the circuitry looks similar to that used in additional positive feedback.V characteristics are transformed as illustrated in Fig.Hamalainen et a/. we see that the <P — I characteristics are not skewed like the corresponding <I> — V curve in the current-bias scheme with positive feedback. 65. on the falling slope the effective gain 8 F / 8 0 is increased and on the rising slope it is decreased. If ee is the noise voltage and /„ the current noise caused by the operational amplifier. / (113) where RD(3—dV/d&. provided that | / ? M y | > l . the system becomes unstable. If Rf <(\/3Mf\ — l)RD. Fortunately. and Mf is the mutual inductance of the feedback branch. curve (1) represents the largest value of Rf. To illustrate the noise cancellation. RD the dynamic resistance of the SQUID. the normal <l> . As a result of the r feedback 1 i {a) ^_^ < > a> o> (0 o /~> >. * / . 1991). ef = \/4kB TR f is the noise of the feedback resistor. a fixed resistor was used to set the additional gain. April 1993 cancellation 1+Rf[\ tBMft + G(a>) + R /RD]/R (114) Here G(co) is the gain of the operational amplifier and R the resistance in its feedback branch [see Fig. Vol. the flux seen by the SQUID is <*> = <*>ext/ -OQ/2 -<V4 0 3> OQ/4 <V2 FIG. the SQUID sees. 1992. 34(a). Seppa et al. 33(b)]. decreasing the maximum slew rate. In the experiment of Drung et al. (b) Measured flux-to-current characteristics (Seppa. However. The 4>— V characteristics become skewed with increased gain. 33(b). a modulation signal (artificial noise) has been added to the input of the amplifier. The curves have been separated vertically for clarity. SQUID characteristics with direct readout. However. No. For the small-signal voltage v.. an adaptive noisecancellation system can be constructed when the resistor Rf is replaced by an adjustable component and a controller is set to monitor the amount of cancellation.: Magnetoencephalography 458 If the external flux to be measured is <E>ext. (1990). this has been verified experimentally. the SQUID can be voltage biased and the total noise reduced by cancellation of the amplifier noise (Seppa. the contribution of the preamplifier noise can be reduced. in the presence of feedback. Mod. Amplifier noise Instead of the commonly used current bias. *V l+PMf (115) .. This technique can effectively prevent instabilities and higher noise. 2. f* / • t i \ •1 \ > • i \ 1 -OQ/2 • -OQ/4 * \ * // • * / *» \ y \ \ i i f i i 0 \ \ ' 1* ^Q/4 branch. Such a system is depicted in Fig. OQ/2 b.' X \ /// ' \» s^**\. The amplifier noise that is coupled magnetically to the SQUID via the feedback coil is canceled by a signal that is proportional to the amplifier noise but of opposite sign. Note that the modulation is canceled on the rising slope when Rf is decreased. adaptive gain control can be established (Seppa et al. 34. Still. 1992) of a voltage-biased SQUID with amplifier noise cancellation for four different decreasing values of the feedback resistance (see text). then the total noise at the output is (considering only small signals) R •+ 1 RD 1+PMf R. Since G{co)»R /RD »1 at low frequencies. Phys. no positive feedback can actually take place because the voltage over the SQUID stays constant. the total noise decreases until the limit set by the flux noise of the SQUID is reached. the flux Mf Rf+RD 4> = <*>ext/ dV a<i> (112) where Rf is the feedback resistance. By increasing the gain of the SQUID. eR =\//4kBTRD is the SQUID voltage noise. (a) Fluxto-voltage characteristics of a current-biased SQUID without positive feedback (solid line) and with positive feedback (dashed line). this effect is partly compensated by the large bandwidth. 1991). and O n is the flux noise of the SQUID. Curve (3) corresponds to the optimum Rf — {\pMf\ — \)RD.

: Magnetoencephalography The contribution of ee can be canceled exactly if f3Mf = —(Rf-\-RD )/RD. Seppa and Sipola. the dynamic range of the voltage-biased SQUID with adaptive noise cancellation is not reduced. 35. where the feedback loop gain is proportional to o>~3/2 (see also Sec. is constant at the normal working point. The total gain is proportional to 1 /(1 — y )• When y=0. no dramatic effects occur contrary to the situation with the current-biased SQUID (Seppa et al.E. the effective control current changes as well and the switching probability is altered very steeply. and at y — 1 the system becomes unstable. 36. . The output voltage can then be changed into digital form at low temperatures. 34(b). The output Vk switches to the voltage state with a probability of 50% when the sum of the control current. 36(b)]. especially for multichannel magnetometers. no meaningful data can be obtained for the current-biased SQUID because the working point is unstable. The bandwidth of this device. Examples of measured flux noise as a function of the amount of feedback or cancellation on both voltage. Mod. see text.9 1991). April 1993 -cz> Tl. with an open-loop gain of 130 dB at 10 Hz. No. used as a comparator (Drung. Hysteretic SQUID -L \ Write gate • Storage ring Magnetic coupons FIG. 1990). and with the enormous 170-dB dynamic range at 1 kHz. The use of a direct readout system results in very simple electronics. including a magnetically coupled storage loop. Flux noise as a function of the parameter y = RD(\PMf\~\)/Rf (see text). 65.. However. For details. J ck . Rev. Another possibility (Fujimaki et al. When y>\. connected to a high-noise preamplifier (17 nV/v'Hz). Vol. plus a term directly proportional to the voltage over the SQUID.9 1992). 1986. The system becomes unstable if J^y < (| jSMy \ — 1 )RD. Since the <E>— V curve is not skewed. As before. The distribution of positive and negative pulses F a c then depends on the external flux. Open circles refer to additional positive feedback and solid ones to amplifier noise cancellation. as has been demonstrated already by Koch et ah (1991). These pulses can be used to add or subtract flux (a) 'M 459 ickT ic Ti•gk 100 UDC H D/A © CD VMYJ I o Comparator (hysteretic SQUID) o SQUID (b) FIG. V. we must have \l3Mf\ > 1. 1992) can be realized by employing adaptive noise cancellation and a PI 3 / 2 controller. is 300 kHz in the flux-locked-loop mode. The solid circles refer to a voltage-biased IBM SQUID (Tesche et al:j_ 1985).and current-biased SQUIDs are given in Fig.Hamalainen et a/. Germany) and connected to a low-noise preamplifier (1 nV/VHz). This is illustrated in Fig. 36(a). By connecting an up-down counter (UDC) and a D / A converter to the comparator output. for example. The straight lines are only for guiding the eyes. with a hysteretic dc SQUID of nonunique flux-to-voltage characteristics. biased with a constant current J g . Some of the signal wave forms have been indicated schematically in the insets. no feedback is present. An example of the amplifier noise cancellation for different values of Rf is given in Fig. The comparator is probed with a current pulse Igk presented at the clock frequency / c l . From Seppa et al. Even if the feedback exceeds the critical value by a small amount.y 1988) is that the SQUID itself is strongly hysteretic and ac biased with biphasic pulses 7 ac [Fig. when connected to an ultra-low-noise SQUID (Gronberg et al. Digital readout An interesting alternative to conventional readout techniques is to employ an integrated A / D converter on the same cryogenic chip as the SQUID. The use of voltage bias is convenient because the circuit is simple. the operating point is locked and the feedback current / ^ replicates the external flux to be measured. dc SQUID electronics based on digital readout: (a) with a separate comparator and external feedback and (b) with a hysteretic SQUID and internal feedback. A highquality dc SQUID electronics system (Seppa. Drung et ah. The open circles were measured for a current-biased dc SQUID fabricated at the Friedrich-Schiller-University (Jena. 2. The promising experimental results obtained so far make this scheme attractive. Phys. 1989). and then by feeding back the converted analog signal V0 to the SQUID as flux. 35. (1991). if the SQUID output changes.I. 3.

e. Thomas and Duret. one has to find a sensor configuration that gives the best locating accuracy. Knuutila and Hamalainen. The limits depend not only on the signal-to-noise ratio. Vrba et al. .4 and IV. almost all biomagnetic measurements were carried out using single-channel magnetometers with pickup and compensation coils wound of superconducting wire on three-dimensional formers. increasing the pickup-coil diameter improves field sensitivity but reduces spatial resolution.E. In this way./= 1. We shall discuss both methods in detail below. without the need of additional D / A conversion. 1982.4. Figures of merit for different configurations can be obtained only by making assumptions about the signal sources and their Rev... Montgomery. III. Optimal coil configuration Because of the large variety of possible source current distributions in the human brain.F.2.^ / V H z at 0. The optimal configurations have also been discussed by several authors (Romani et al. .4. as discussed in Sec. The Fourier transform of a function Bz(r) sampled at discrete points is equal to a set of spectra obtained by shifting the region S in KN periodically by kn = 2f=i*it>. The locating accuracy of a magnetometer may be simulated numerically by adding noise to the calculated field values and by fitting an equivalent dipolar source to the data by means of a least-squares search. general criteria for optimal magnetometer design do not exist. the freedom of the designer is restricted by several constraints: the size of the Dewar. Therefore. have been reported (Drung et ah. as was discussed in Sees. 1.. related to (a z } by aI--b7-=8/_/-.-. a . 65. . Duret and Karp. Unfortunately. is the Kronecker delta.. 2. Digital readout has been demonstrated to work in a rather promising way. . V. Knuutila. when an additional low-frequency noisecompensating modulation scheme was used. Correspondingly. In practice. Ilmoniemi et al. To obtain a criterion for aliasing.460 Hamalainen et al:. A positive pulse causes the junctions at the gate to switch sequentially and momentarily to the voltage state to allow the entry of one flux quantum. No. The output can be deduced from the pulses at room temperature and counted. 1983. the properties of the SQUID sensors.H e r e > {*>. 1984. 1989).N. the use of a multichannel device for mapping the spatial pattern of the magnetic field caused by these sources is very desirable. 1964. and the base line of the gradiometer followed the principles described in Sec. 1964. .. Flux-noise values of 7 X 1 0 ~ 7 O 0 / V H z above 1 kHz and 5X l O . . 1993) can be used to determine the largest spacing allowed for the measurement grid if aliasing is to be avoided. A useful figure of merit for multichannel magnetometers is the precision in locating cortical current sources. . A more effective way is to determine directly the confidence regions of the least-squares fit. Locating current sources in the brain and the followup of their dynamic behavior are the main objectives of the neuromagnetic method. spatial sampling theory and information theory can be employed for determining an optimal coil configuration.> where ni are integers and [a. a suitable source current model and a volume conductor model must be chosen. 1984. such as studies of spontaneous brain activity. / = 1. and that Bz is spatially bandlimited. IV. and the characteristics of the brain signals. Sampling of neuromagnetic signals If the spatial frequency content of the signal to be measured is known.D. A current dipole in a spherically symmetric volume conductor is usually appropriate. Farrell et al. the distribution and strength of external noise sources. Vol. F.. 1985. 1983. multidimensional sampling theory (Petersen and Middleton. such as their diameter and length. The slew rates of digital SQUIDs are below those of conventional devices. the feasible number of channels. Phys.N] is a base of KN. where 6 / ..-. Repeated simulations then reveal the average locating error. Ahlfors. • • . This problem occurs when the grid is not dense enough to capture the spatial variations of the field. these two goals are sometimes contradictory. et al. In addition to calculation of the confidence limits. The pulses are coupled magnetically and injected into the write gate. the spatial distribution of the biomagnetic signal and of the disturbances must be taken into account. 1988). For example. Ahonen et ah. 1989). 1962. To estimate the error limits. but also on the depth of the dipole and on the measurement grid. Multi-SQUID systems Until the middle of the 1980s.N] form a basis of the space R N. the calculated confidence limits can be applied for comparing various magnetometer arrays (Knuutila et al. increasing the clock frequency should improve the situation. However. In addition to maximum sensitivity in every channel. .G.. a negative feedback circuit is established on the SQUID chip..1 Hz. Such instruments not only speed up the measurements but also give more reliable data and make possible experiments that require simultaneous recordings over a large area.. In maximizing the field sensitivity and discrimination against external noise. April 1993 characteristic field distributions. Magnetoencephalography quanta trapped in a storage ring which is magnetically coupled to the actual SQUID sensor. 1982a. The optimization of the coil parameters. . Assume that the vertical component of the magnetic field Bz is sampled at points r„ = 2/^=1 w . 1987. The storage consists of a write-gate interferometer which is connected to a large flux-collecting coil. a negative pulse subtracts one flux quantum. restricted to a bounded area S of the reciprocal (momentum) space KN. 2. first into the write loop and then into the storage ring. Therefore. Mod. the number of turns. ij = 1 . i.

and K is the volume of the unit cell in the reciprocal space. Interpolation can then. k „ ^ 0 . except at some singular points. only be carried out up to an additive constant f0. multiple. Information conveyed by the magnetometers The angles between the lead fields (see Sec. 1962).k „ ) . we require again that Eq. the amount of information gathered increases much more slowly than the number of channels. Equation (117) represents a sum over the periodically extended images Sp of the support S: Sp = [S +kn. As a result. 1987). instead of the amplitude of BZ(T). Obviously. and independent measurements at one point allow a certain degree of overlap. when adding more channels to a given area.. A is the volume of the unit cell spanned by the base vectors a. If the shifted images Sp do not overlap. 1964). all components of the gradient vector VBz(r). But what happens if there are two dipoles. (117) n where 3&z and § represent the Fourier transforms of Bz and g. (119) where j is the imaginary unit. if each point k belongs at most to N different periodically extended images S +k w (Ahonen et aL. Knuutila.F.(Sp—S). (122) p where J denotes the source ( = primary) current density and the integration extends over the source area. is indeed sufficient to avoid aliasing. as will be discussed in Sec. even when there is at most an (iV — 1 )-fold overlap in the imRev. Eq. The lead field JLt of the magnetometer i measuring the component Bt is defined by [see Eq. depending on what types of gradiometers are used. Thus the sampling grid {a. Ahlfors..e. No. After this critical density has been reached. In particular. Thus we arrive at the conclusion that in practice a grid spacing of 30-40 mm. respectively. et aL. (116) n The exact reconstruction of BZ(T) from the samples Bz(rn) is thus reduced to deriving g(r). 1964.X1)<X2. (117) is. As above. Suppose now that we measure. we can assume that J50 = 0.B) of two channels provide a measure of the amount of different information conveyed by each channel (IImoniemi and Knuutila. the result is obvious: since twice as much information is gathered at each point. typical cortical dipoles are more than 30 mm from the sensors.2 i r / k . (123) V<Xi. (120) ( k . j should be dense enough to provide a grid {bf-j which is sufficiently sparse to prevent overlap. Then. this limit is thus set by practical constraints..6. To avoid aliasing. permitting the use of a less dense sampling grid (Montgomery. This is a generalization of the classical one-dimensional Nyquist sampling criterion. i.k<E(S+k n ). this can be transformed to %z(k)= -2irjK 2. (43)] Bt^fjCiiThyWdvzziXttJr) n 461 . of course. However.£ ( k ) = 0. April 1993 ages of S. in an iV-dimensional space. 3. (121) Here. the overlap could be iV-fold (Montgomery. V. kn = 0 . (117) must be an identity for a properly sampled signal. Eq. the sampling grid of these dual sensors can be sparser by a factor of Vl compared with what is needed for sampling just Bz. Eq. IV. 65. one can find g(r) such that the exact reconstruction of BZ(T) from the samples Bz(tn) is possible by choosing S(k)=l/K for fcES and S(k)=0 when kE. Mod. Intuitively. 2. Petersen and Middleton. the "unit cell" around a point in the sampling grid can be twice as large.. such that everywhere in R N.S ( k ) = A . These conditions can be met.Hamalainen et a/. 1964). if we measure the planar orthogonal derivatives dBz /dx and dBz /dy at the same point instead of just Bz. 1984. an identity. (118) n Without loss of generality. 1993). The angle <f> between lead fields X 2 and JL2 is then defined as UUJC2) 0= arccos—=========.%z(k-kn)(k-kn)-S(k) . stronger dipole is reduced below the noise level. k G 5 \/kn = 2 f = i rt/bf}. 1964.9 1993). This result has practical consequences. Petersen and Middleton. If the amplitude is measured together with the gradient vector. Thus an appropriate grid spacing for locating a single source would be 30 mm for magnetometers and axial gradiometers and 40 mm for planar gradiometers. and no aliasing occurs (Petersen and Middleton.Bz(Tn)g{T-Tn) . Equation (116) can be transformed into £2(k)=K J/3Bz(k~kn)S(k) . at a separation / which is less than the distance to the detector array: can the two dipoles be resolved from a single dipole? It turns out that the discrimination capability is reduced with decreasing / only because the difference between the amplitudes produced by the pair of dipoles and a single. This can be accomplished if . In practice. indeed. k G 5 .X2) The concept of marginal utility.: Magnetoencephalography we try to construct an interpolation function g(r). stating that the inverse of the sampling interval should be at least twice the largest frequency present in the signal to be sampled. (119) be an identity. (116) is changed to ^(r)=£0 + 2V£z(rJ-g(r-rJ . Bz(T) = -2. Vol. Phys. simultaneous. may be studied best by computing the information-theoretical channel capacity of a given magnetometer configuration (Kemppainen and II- . especially in designing devices covering the entire head. but not because of the appearance of short-wavelength spatial variations in the field that would cause aliasing in the sampling grid (Ahonen et al.

1991). In the second phase of development.. In addition. T o find an adequate number of channels N. the equation for the total channel capacity is Aot = | i l o g 2 ( P / + l ) . having typically more than 20 SQUID channels distributed over an area exceeding 10 cm in diameter. To obtain an expression for the channel capacity of a multi-SQUID magnetometer.X. et aL. it is necessary to orthogonalize the lead fields by the transformation X. We shall now briefly review these devices. compact modular construction is advantageous. Williamson et aL.F. spaced 30 mm from each other and located on a spherical cap with a diameter of 125 mm and a 125-mm radius of curvature. careless cabling between the liquid helium and room-temperature parts of the system may increase the boiloff rate excessively. a large measuring area requires thick Dewar walls for strength and thus a wide gap between the room-temperature surface and t h e 4. T h e first-order axial asymmetric gradiometer coils were laid in a hexagonal array. (129) This figure of merit has the advantage that it properly takes into account the sensing-coil geometry. materials used inside t h e Dewar must be carefully chosen. V. 1984. Every unit measures the off-diagonal derivatives dBz /dx and dBz /dy. A spherical bottom is a compromise approximating the average curvature of the head.>S2AT'2. Vol. T h e device is shown schematically in Fig.g. In these calculations.9 channels have been reported since (Carelli et aL. V. a. 1989). Several new instruments with 7 . we compute the matrix II containing the inner products of the leadfields. No. For example. A t first. nearly 10 cm in diameter. 37 and in a photograph in Fig. One can choose M = U r because then IT = M I I M r = A .2-K interior. April 1993 used simultaneously. 1989. The magnetometer outputs ut are Ui=bi(t) + n(t)^{j:i9JP)+n(t) . By comparing the increase of the channel capacity as a function of N with the additional cost and increased complexity of the construction. and z refer to a coordi- .C).viz. Such devices have been discussed in detail in a review on multi-SQUID systems (Ilmoniemi et aL. 1991) was in use from 1989 to 1992 when it was replaced by a whole-cortex device (see Sec. 1985). where x.jCi)s2/o2 . J t o t increases steadily with N. the possibility for computer control of the electronics is important.Xj) . The 7-SQUID gradiometer at t h e Helsinki University of Technology (Knuutila. to keep crosstalk between channels sufficiently low.D6sselef aL. a robust grounding arrangement is essential. we calculate 7 t o t as a function of N. The noise level of this instrument was 5 f T / V H z . 1989.a). 5. The noise in these devices was about 20 f T / v / H z . Practical aspects In a multichannel magnetometer. When many channels are Rev. Furthermore. J t o t saturates. and uniformly distributed in the conductor volume. Examples of multichannel systems The first multi-SQUID magnetometers had 4 . However. e.. (125) Here. no Dewar cap can be made optimal for all subjects. Making Dewars for wide-area multichannel magnetometers is problematic.5 channels on an area of only a few cm in diameter (Ilmoniemi et aL. the distance to the source area. Ahonen et aL.Hamalainen et al. T o find the transformation matrix M with components M(j. as well as the sensor distribution. Guy et aL. t h e source currents Jp are assumed to be completely random with zero mean and a variance s2. Assume now that we want to cover a certain area with magnetometer units. cables must be well shielded and symmetrized.. the inner product of the lead fields (X^Xj) describes the strength of the coupling of channel i to the sources.-M/y. After this limit.\ Magnetoencephalography 462 moniemi. UiJ = Ui. The 24-channel gradiometer at the Helsinki University of Technology Our 24-channel system (Kajola et aL. 65. This is because of overlap by the lead fields of additional channels when they are brought closer to each other. after a certain limit.Zi . In addition. Because of considerable variations in head shapes. a new generation of instruments emerged. 1985.=2. Ahlfors. the orthogonalized signal-to-noise ratios P / give only a negligible contribution to J t o t . For high-sensitivity magnetometers. Mod.6. (124) where n(t) is Gaussian noise with zero mean and variance (j 2 . 1989). d27) where crJ2 = 2y(Ml-/-cr-)2. we make the spectral decomposition I l ^ U A U ^ where A is a diagonal matrix. Finally. 7-channel dc SQUID magnetometers were introduced. 10. below 10~ 2 . In 1989. Phys. The fact that some of the channels are rather far from the scalp imposes high demands on the sensitivity of the SQUIDs. The signal-to-noise ratio of one channel is given by Pi = {jCi. It has 12 two-gradiometer units.. (126) Then. Romani et aL. too (see Sec. (128) Next. 4. 2. 1989. the signal-to-noise ratio is P/=U. Ohta et aL. This is also necessary to provide immunity against ambient external noise. 1989. we thus obtain a practical optimum for the number of channels. 1987) covered a relatively large area. It has been used for comparison of various multichannel magnetometer configurations (Kemppainen and Ilmoniemi. the noise level. y. 1989).

2. The low-pass-filtered results are digitized with a dataacquisition unit and transferred to a computer with a Rev. V. The sensor array is housed in a custom-designed Dewar (CTF Systems. some . The SQUIDs are above the coils.D. Phys. The pickup area of the coils is 6 cm 2 . 37.. each containing one chip with ten dc SQUIDs (Daalmans et al. As discussed earlier (see Fig. which are connected to the measurement computer by a local area network. each on a Euro-1-sized printed circuit board. Vol. this design produces its maximal response just over the dipole. April 1993 real-time operating system. The results may then be analyzed later off line in U N I X workstations. are indicated by solid and dashed lines. 1985. The noise of the system is typically less than 10 f T / v ' H z at frequencies over 10 Hz. 27). The system includes three additional magnetometers to measure the x. and the preamplifiers. 1991). fabricated by IBM (Tesche et al..7 cm 2 . is typically less than 2 % .5 fT/(cmA/Hz). The SQUIDs are inside of four separate superconducting shields. Because of the flat-bottomed Dewar. 9). The detector-controller units. with resonant impedance-matching transformers. with a modulation at 90 k H z applied in phase to all channels. The equivalent gradient noise of the sensors is 3 . are mounted on a conical former above the wirewound flux-transformer coils. The system uses conventional flux-locked-loop electronics. The bias. and flux-modulation cables from the SQUIDs to the preamplifiers also run in groups of eight inside Cu-Ni tubes. the direction of the dipole is given by the expression (dB2/dy)ex — (dBz/dx)ey. Inc. where ex and e^.y 1989). and z components of the external magnetic field for noise cancellation. The resonant impedance-matching transformers are located in the helium storage area of the Dewar. measuring dBz/dx and dBz/dy. are outside the shielded room in an rf-shielded cabinet. the pickup and compensation coils are within a 19-cm-diameter cylinder in two planes separated by 70 mm.: Magnetoencephalography (b) 463 SQUIDs 30 mm 125 mm FIG. This gives a three-day period between refills. The crosstalk between channels is below 1%. voltage-measurement. 7(c)]. The distance from the sensors to the outer surface of the Dewar is 15 mm. see also Sec.Hamalainen et a/. see the Appendix for addresses) with a curved and tilted end cap and a 10-liter capacity for liquid helium. y. 65.. The hexagonal arrays of hexagonal first-order gradiometer pickup and compensation coils were fabricated on flexible printed-circuit boards and glued onto a support structure so that axial gradiometers were formed. located on a spherical cap. are unit vectors.5). in groups of eight. respectively. nate system local to each sensor separately [see Fig. The Siemens 37-channel "Kren/kon" systems Figure 38(a) shows the flat layout of the pickup coils in the 37-channel axial gradiometer manufactured by Siemens A G in Erlangen (Hoenig et al. The SQUIDs. Mod. Spontaneous data are saved on a digital tape. The imbalance of the gradiometers. are on top of the cryostat in detachable aluminum boxes. The coils. " K r e n i k o n " is most suitable for cardiomagnetic measurements. Schematic illustration of the flux-transformer array in the 24-channel gradiometer at the Helsinki University of Technology: (a) arrangement of the pickup coils. operated inside our heavily shielded room (see Fig. No. b.. and (b) one of the 12 two-SQUID units. The base length of the gradiometers is 13 m m and the area of the pickup loops is 3. This computer also calculates averages of evoked responses online and stores the results on disk. This system is intended for use inside a moderately shielded room. The electronics is based on 100-kHz flux modulation.

would be quite far from the brain. Department of Experimental Neuropsychiatry and Otorhinolaryngologic Clinic. 38. Inc.B. The traces are averages of 557 single responses. Germany). Vacuumschmelze shielded room (see Sec. The latest BTi model is a 37-channel system ("Magnes"). especially the outer ones. separated by 7 cm. The gap between the liquid-helium space and the outer surface of the Dewar is 20 mm. 65. fV. 1991). 300 ms) were presented once every 2. Tone bursts (1 kHz. The interstimulus interval was 4s and 384 responses were averaged.I) data-acquisition and computer system with for data analysis. BTi "Magnes" system. The hexagonal first-order axial gradiometer coils are on two flat printed circuit boards. (a) 37-channel gradiometer of the BTi "Magnes" system in a measurement situation. 39. Unpublished data courtesy of Dr.y< i-r-iL^ WWm^— ^ hfr— w*v^v fc H 7 3 100 -100 0 100 ms FIG. Only one of the boards is shown. E. Mod. Herrmann (1992. (b) Auditory evoked magnetic responses obtained with the Krenikon system over the left hemisphere.). Vol.2 s to the right ear. The BTigradiometers The first commercial multichannel neuromagnetometer was introduced in 1985 by BTi.: Magnetoencephalography 464 of the channels. From Pantev et ah (1991).. and they start 100 ms before the stimulus onset. 2. Phys. It had seven channels in a curved-bottom Dewar on a spherical cap of 55-mm diameter and 100-mm radius of curvature. April 1993 n r ^ 30 28 rhr^ 100 i r l -100-100 500 ms FIG. It has a hexagonal sen- 190 mm (b) 12H 1CH lb) s^Arf^ '-^^fr^-J^T^ I Jl. Some of these installations include two such Dewars that make it possible to record from both hemispheres simultaneously. No. (a) Sensor arrangement of the Siemens 37-SQUID "Krenikon" system (Hoenig et ah. University of Erlangen. .8. discussed in Sec. are attached to the subject's head and to the Dewar (courtesy of Biomagnetic Technologies. Krenikon systems have been in Erlangen and in Stockholm. Grummich and Dr. (b) Auditory evoked responses to tone bursts recorded with the BTi 37-channel instrument over the left hemisphere. The coils of the probe-position indicator. The capacity of the helium reservoir is 25 liters. The complete installation also in- eludes a plus a software installed lustrates ment. V.F. requiring refills every four days. 39(a). Figure 38(b) ilan example of data obtained with this instru- c. Siemens "Krenikon" system. P.Hamalainen et a/. The approximate positions of the sensors are shown on the schematic head. V. Rev. shown in Fig.

BTi delivers complete systems with shielded rooms. Japan). Situations in which M E G recordings of focal slow activity are believed to be useful include cerebral ischemia. The radius of curvature of the Dewar bottom is 118 mm. with the SQUID gain increased by means of positive feedback (Drung et aL. with only 20 dB of shielding at low frequencies (ter Brake. Phys. in New York University Medical Center. and in the University of Minister (Germany). and Rogalla. 1991). one at the University of Twente in the Netherlands (ter Brake. 1992). The magnetometer at Philips La- . with roughly Z>shaped curved bottoms. 1991). _ T h e equivalent noise of the magnetometers is 15 f T / V H z and that of the gradiometers 2 0 .Hamalainen et a/. This system was designed mainly for magnetocardiographic measurements inside a heavily shielded room. in Scripps Clinic (San Diego). They are spaced 22 mm apart and are located on a spherical cap with a 120-mm radius of curvature. is 27 mm Rev.5 mm apart. but the coils are wire-wound around 20-mm formers. The second application is based on the observation that the spectrum of the spontaneous electrical activity from functionally injured brain tissue often exhibits a shift towards lower frequency. The field sensitivities at 1 H z range from 6 to 10 fT/VHz. They can be reduced by an order of magnitude by subtracting the signals of two magnetometers. 9 cm above the array. The readout electronics of the SQUIDs is based on low-noise operational amplifiers. 2. In E E G analysis. nine installations of the 37-channel device were in use. a stainless-steel tube with cabling. In August 1992. The remaining three SQUIDs are connected to three orthogonal reference coils. 1992b). as well as output presentations. with the noise increasing by almost two orders of magnitude at 0. laid out in a hexagonal grid and covering a spherical cap of 140 mm in diameter and with a 125-mm radius of curvature.: Magnetoencephalography sor arrangement similar to the Siemens device. 65. 56 mm apart. Wieringa. in collaboration with C T F . 140 mm. No. trauma. Every unit can be mounted into the system individually. The low-frequency disturbances were found to be mostly environmental. spaced 20. was made inhouse. and degenerative diseases. 1991) and one at the Philips Research Laboratories in Hamburg (Dossel et aL. The diameter of the coil array is 144 mm. The noise of the system is 1 0 . The device uses integrated multiloop dc SQUID magnetometers. The system has two Dewars. d. fabricated on individual 9 X 9 mm 2 chips (Drung et aL. The Dornier two-Dewar system A 2 X 14 channel rf SQUID system was fabricated by Dornier. Figure 39(b) shows typical auditory evoked responses measured with the BTi instrument (Pantev et aL. e. April 1993 465 and the noise due to the radiation shields is 2 f T / V H z . The first is localization of the somatosensory cortex by evoked-response techniques to minimize paralysis caused by surgical injury to the motor cortex. positioned as close to each other as possible with the straight sides adjacent (see Fig. This arrangement is intended for recordings of brain signals over both field extrema of a neuronal source. Each magnetometer is a separate entity consisting of the SQUID package. The average of the outputs of all SQUIDs or of those on the outer ring can be used as the reference signal (Drung. tumor. 1991). Vol. Eleven coils are connected to rf SQUIDs via a switching matrix which allows choosing any of the 48 coils to be used in the measurements. have been introduced. in National Institute for Physiological Sciences (Okazaki. The flat-bottom Dewar. for instance. The SQUIDs have eight loops connected in parallel. Together with their clinical installations. The diagnostic value of focal slow activity is enhanced when its generators are located with respect to brain anatomy.2 0 f T / v / H z . or for simultaneous measurements from the right and left hemispheres. BTi is developing application protocols intended for routine hospital use. the vacuum gap at the tip.. Mod. computers. thus forming a gradiometer electronically. f. The channels cannot. Both devices have wire-wound gradiometer coils. et aL. 19-channel gradiometers Two 19-channel systems. there are eight additional SQUIDs for cancellation of external disturbances. connected to orthogonal magnetometer loops for active compensation. with an inner diameter of 250 mm. from each other. however. Two applications are currently being tested by BTi for clinical use. The capacity of the liquid-helium reservoir is 25 liters and the boiloff rate 6 liters/day with all sensor units mounted. each having pickup loops in three orthogonal directions. Both Dewars contain a magnetometer array with the coils wound on 16 cubic formers. 1990) and then laid out on a planar surface along three concentric circles of 70 mm. The Twente system includes three additional SQUIDs. and 210 mm in diameter. rhythmic activity are commonly associated with localized pathology. since the device is intended to be operated in a single-layer ^t-metal room. The PTB 37-channel magnetometer A 37-channel instrument is in operation at the Physikalisch-Technische Bundesanstalt in Berlin (Koch et aL. when warm. particularly below 8 Hz. are being standardized so that a system operator/technician should be able to complete an examination procedure in one hour. A gradiometric response of the device is formed by calculating differences electronically. be reconnected when the device is operational at liquid-helium temperature. 1990). enabling easy replacement of sensors.3 0 f T / ^ H z . for the University of Ulm. including preparation of routine information needed by the referring physician. Flokstra.1 Hz. 40). and software. Patient handling and data acquisition and analysis. focal regions of low frequency or slow. The sensing coils are connected to SQUIDs made inhouse. and an electronics box. In addition to the 37 first-order gradiometers.

. covering an area of 160 mm in diameter. No. including the noise from construction materials and radiation shields used in the Dewar. The reason for placing the planar sensors on the outer rim is to save space. provided that the magnetometer is correctly positioned initially. If the device is placed symmetrically above a current dipole.8 fT/A/Hz and 5 . respectively. April 1993 dipole the field is fairly uniform. it is important to measure even over a larger area than has been possible earlier. Determining a proper sensor spacing is to a large extent an economic question: in order to have sufficient coverage of the whole head. there are 12 planar gradiometers of 10-mm diameter and 15-mm base length. because of the wide volume required by the axial gradiometers. a 122-SQUID system (Ahonen et al. The helmetlike Dewar bottom must be large enough to accommodate most heads. Whole-head coverage Multichannel systems described in Sec. Obviously. radially oriented axial gradiometers cannot be used. 37 and 38 have been drawn on the same scale. 40.466 Hamalainen et a/. 1993) has been operational . see text. Instead. 1992). the Dewar would otherwise be too large. The planar gradiometers are also less sensitive because they are smaller than the axial ones. outermost circle.. Other important questions are the proper sensor type and their spacing. There are several practical problems in building such a device.. the utility of the additional planar channels is marginal in this device. The "Neuromag-122" system In our laboratory. if necessary. allowing each channel to be changed individually. In the center and along two surrounding concentric circles there are 16 axial gradiometers with a coil diameter of 18 m m and a base length of 85 mm. a. Mod. This means that a high sensitivity and low environmental noise level must be achieved.. Another possibility is the use of magnetometers with the gradiometric signals formed electronically. The coils are wound on cubic formers. over a hundred SQUIDs are needed.5 possess enough channels spread out on a sufficiently large area for locating cortical sources from measurements without moving the instrument. measuring the derivative along the tangent of the circle. Torrioli et al. 65. since their maximal response is just above the source. V. the coils are wound obliquely on formers parallel to the axis of the cylindrically symmetric Dewar to approximate a radial orientation. whereas far from the Rev. This coil configuration and those in Figs. because they measure only one gradient component at each point and because they are on the outer rim. Sensor arrangement in the Dornier 2X 14-SQUID magnetometer. Two whole-cortex instruments meeting this goal are operational at present. The device has both axial and planar gradiometers. The 28-channel "hybrid" system in Rome A 28-channel system has been developed in Rome at the Istituto di Elettronica dello Stato Solido of C N R (Foglietti et al. On the third. However. especially in clinical studies. However. the planar gradiometers detect only small signals. 1992. 1991.F. g. boratories is modularly constructed. evidently. Knuutila et al. Phys. at least some of the channels are then bound to be quite far away from the current sources in the brain. 6. and during various complex investigations.8 fTAcmv'Hz). The noise of the axial and planar gradiometers is 6 . The detectors are located on a spherical surface of 135-mm radius of curvature. 2. Since the device is in a curved-bottom Dewar. planar gradiometer coils are attractive because of their small size. Vol. For details.: Magnetoencephalography 100 mm FIG.

65. 42(a) as prominent deflections near the right and left temporal lobes.and right-hemisphere measurements separately. Multipair shielded cables are laminated into the surface layer of the neck plug and lead from the top plate to the connector boards where the components for the SQUID gain control are located.: Magnetoencephalography since June 1992. 7(c)]. Figure 42 shows results from an auditory evoked measurement. Planar gradiometers couple strongest to current dipoles located directly underneath the pickup coil. the four connector boards. The sensors have an equivalent gradient noise of 3 . a two-dipole fit was applied to the magnetic-field data at Rev.I.. The separation between two double-sensor units is about 43 mm. Tone pips at 1 kHz. a full two-dipole fit was used to find the globally best solution." leading to a very reliable construction.b). which is easy to test and assemble.5 fT/(cmV / Hz). To locate the sources for the auditory evoked response. covering the entire scalp (see Figs. On the right. Phys. which is a significant advantage of an array of planar gradiometer sensors over more conventional axial gradiometers.a and Fig. Vol. The distance from the outer surface of the Dewar to the gradiometer coils is 16 mm. because measuring two planar gradients simultaneously corresponds to a conventional axial-gradiometer sampling grid that is denser by a factor of Vl. 41. Data handling is distributed among workstations. IV. 42(c) the equivalent current dipoles resulting from a fit to the field patterns are shown for both hemispheres.Hamalainen et a/.F. The Dewar bottom is helmet-shaped. they are connected to dc SQUIDs fabricated by IBM (Tesche et al.3 and V. Modular construction is employed throughout the probe unit of "Neuromag-122.. A special elastic construction guarantees a welldefined position and orientation of the gradiometer coils.2. V.5. The electronics is based on direct readout with amplifier noise cancellation (see Sec. The main modules inside the Dewar are the top plate made of fiber glass and a printed circuit board. Figure 42(b) shows a rough estimate for the source currents in both hemispheres. the entire electronics is placed outside the magnetically shielded room. with radii of curvature between 83 and 91 mm. No. This gives the approximate locations of the sources directly from the measured data. Neuromag Ltd. V. which is now being used for basic research in our laboratory. are encapsulated into identical plug-in units. this is sufficient.6 cm 2 . The effective area of each oppositely wound loop is 2.5 mm. The dimensions. The 122-channel gradiometer ("Neuromag-122") built in Helsinki. as in our 24-channel system [see Sec. and the base length of the gradiometer is 16. is remotely controlled by a dedicated processor. Thereafter. 1985). In Fig. The thin-film pickup coils are deposited on 2 8 X 2 8 mm 2 silicon chips.5). V. The interior is broader in the parietal than in the frontal regions.F. V. 108 ms after the stimulus. The initial guesses for dipole positions were obtained by fitting single dipoles to the left. thus fitting better to the typical shape of human skulls. The lower maps show the calculated field patterns. measuring dBz/dx and dBz/dy. have been chosen to accommodate about 95% of adult heads.E.2. four connector boards. The operation of the electronics. 9 and 41).8) that is employed to determine the position of the subject's head with respect to the sensors. and a wiring unit onto which the 61 gradiometer components are mounted.D. the neck plug of fiber-glass-coated foam plastic. The acquisition computer can be used to generate stimulus sequences. with 50-ms duration. and the wiring unit with 61 two-channel plug-in gradiometer components. This eliminates the need for separate preamplifiers. The field distributions were obtained from the measured data using the minimum-norm estimate (see Sec. The acquisition computer also controls the electronics of the probe-position indicator (see Sec. including the tuning-of. 41) consists of a top plate made of fiber glass and a printed circuit board. together with the dc SQUIDs. connected via a local area network: one for data acquisition and the rest for analysis. Mod. The NIOOm response could be recorded simultaneously from both hemispheres.F. a neck plug of fiberglass-coated foam plastic. The device employs 61 planar firstorder two-gradiometer units. but it is also possible to employ external triggering. 2. were presented to the left ear of the subject. the SQUIDs. The result was obtained by rotating the two-dimensional field gradients by 90°. . The instrument (see Fig. these systems employ Finnish dc SQUIDs (see Sees. they are within 1 mm of the Dewar bottom. the distribution of the sensors (below) and the gradiometer chip (above) with the two orthogonal figure-of-eight coils are illustrated schematicaliy. April 1993 467 Top plate FIG. The two-gradiometer chips.l). As was discussed in Sec.D. Similar instruments are available commercially. V. This can be seen clearly in Fig. (see the Appendix) has produced the prototype system.

nonadaptive algorithm (Vrba et al. 211 responses were averaged and digitally filtered with a passband of 0. 1993). Note that the field is stronger over the right hemisphere. The all-niobium SQUIDs have a whitenoise level below 10~ 5 4> 0 /V / Hz. The onset of the \/f noise is at 1-5 Hz. 42. Vol. The upper field pattern was obtained from the measured data using the minimum-norm estimate (Hamalainen and Ilmoniemi. A noise level of 10 f T / V H z in the frequency range from a few Hz to 60 Hz was reached in an unshielded environment (Vrba et al.. 41).and second-order gradient components to characterize the background fields. 41].03-45 Hz. 1) over the right and left hemispheres. Mod.or third-order gradiometers can be created electronically to make operation possible even without a magnetically shielded room.. to be made in collaboration with . Auditory data recorded and analyzed using the "Neuromag-122" system (see Fig. The 2cm-diameter wire-wound coils have a base length of 5 cm.: 468 Magnetoencephalography t. b. (c) Measured and calculated field patterns of the source dipoles at 108 ms (the NIOOm auditory response. Phys. L =left. (sec) Two-dipole fit to auditory data (t = 108 ms) ->/•*• >«v~ Right Calculated field Left FIG. see Fig. Another whole-head system planned by C T F will have magnetometers inside a whole-body high-7^ superconductor shield. The electronics is based on digital signal processors. The electronic Rev. No.Hamalainen et a/. the estimate was obtained by rotating the two-dimensional field gradient by 90°. the helium boiloflf rate is 12 liters/day. (a) Responses to 50-ms 1-kHz sinewave tone pips presented to the left ear with an interstimulus interval of 1 s. The helmet-shaped lower portion of the Dewar has a distance of 15-16 mm between the 300-K and 4-K surfaces in the area under the signal channels. has built an instrument covering the whole head with 64 axial symmetric first-order gradiometers on a quasiregular grid (Vrba et crt. 65.7 days between refills. R — right) show the two independent tangential derivatives of Bz. and the average center-to-center separation is 4 cm. The magnetometer architecture allows expansion up to 80 sensor channels. the field component normal to the helmet-shaped bottom of the Dewar at each of the 61 measurement sites. The lower part of the figure shows the calculated field. April 1993 cancellation is carried out digitally in real time with a time-independent. The CTF instrument C T F Systems Inc. Along the longitudes and the latitudes [see inset heads and also Fig. The two sets of data (nose up. 1993).y 1991). 2. giving a period of 5 . The system also contains 16 reference channels to measure three field components as well as first. (b) Coarse estimate for the source currents at t = 108 ms after the stimulus. and with the help of the reference SQUIDs second. contralateral to the stimulated left ear.> 1993).

. Since in M E G experiments the field distribution generally differs from that used during the calibration. the uncertainty in the position of the field point may be considered as a source of extra noise. Probe-position indicator systems The accuracy with which the position of the magnetometer can be determined in relation to the subject's head affects the calculated location of cortical sources in an essential way. such as magnetic shielding. 1987). When a multichannel magnetometer covers an extended area. the single-channel magnetometer has been positioned with some alignment marks on the Dewar and on the subject's head. In single-channel experiments. be found with respect to a coordinate system defined by at least three fixed points on the subject's head. Instead. . and let O represent the undistorted applied flux values. Feedback is applied to the flux transformer rather than to the SQUID. the situation becomes complicated if the mutual inductances cannot be measured directly. based on feedback. Vol. the Dewar bottom is quite large and its accurate positioning according to markers becomes difficult. crosstalk between the detection coils is eliminated. . a small coil may be consecutively placed in several positions on a grid (Buchanan and Paulson. or three or more small coils fixed on a planar plate by means of printed-circuit techniques may be employed. In practice. This automatically takes into account the interchannel coupling corrections during the measurement of the calibration coefficients. the calibrations Kt cannot be calculated simply from ^t/Vi9 where now the applied flux <f> is obtained from the geometry of the experimental setup. an orthogonal Cartesian frame of reference determined by both ear canals and the nasion. the true calibration 3>. Possible uncertainties then cause systematic errors in the location of the current dipole. with the sources in the phantom generating a field distribution which resembles that elicited by the brain. The project is sponsored by the Japan Key Technology Center and nine private companies. . Ahlfors et aL. The accurate position and orientation of the Dewar with respect to the head can be found by measuring the magnetic field produced by currents in small coils attached to the scalp (Knuutila et aL. Nevertheless. the inion (the protrusion of the occipital bone at the back of the head). Therefore. If the inductances of flux transformers and their mutual inductances are known. 8. (130) where L = d i a g ( L 1 . Best results have been achieved using multiple sources. one measures the output voltages V = (Vl9. Calibration of multichannel magnetometers Shielding currents in superconducting flux transformers distort the magnetic field. Of the N(N — l ) / 2 independent mutual inductances between N channels. SQUID sensor development. 1992). was first proposed by Claassen (1975). Let us further define the transpose of a vector of net flux values * ' = (4>'1.LN) is a diagonal matrix containing the total inductances of the flux transformers and I is the identity matrix. Flokstra. reaching an accuracy of 1% or better with a single calibrator source may be difficult./F. Phys. . cryogenics. Let us assume that there are N channels and that the mutual inductances are described by an NXN matrix M . . The plans include most aspects related to building a magnet oencephalographic system suitable for brain studies. One possibility for defining a coordinate system is. Traditionally. 1991). . 2. In practice. April 1993 469 An alternative method. Fixed points on the head that have been used in defining coordinate frames include the ear canals. however. 7. This procedure keeps the current in the transformer effectively constant. one must note that the effective inductance of a flux transformer coil is affected by feedback. The shape of the skull may also be digitized to provide 3D information. In doing so. Matsuba et aL.<E>^)r measured by the flux transformers. and the front teeth. This approach was first used with the 7-channel magnetometer in Helsinki (Knuutila. . For example. the best solution is to calibrate the device with a phantom head. . In this case. In multichannel devices this problem is simpler to solve because the relative sites of the different channels are known a priori. . Japanese plans The most extensive project to develop whole-head magnetometer systems has been announced by the Superconducting Sensor Laboratory of Japan (see the Appendix for address). for example. No.. Consequently. software. the relative positions of which were accurately known with respect to each other. the effects of coupling also differ. which are related to O ' via the calibration coefficients K{.Hamalainen et a/. however. As a result.KN. an individual channel does not sense the original external field. a correction can be calculated easily. . The method has been used successfully (Hoenig et aL. 1991. Accurate calibration also requires a well-designed experimental setup with no external field distortions and an exact geometry. The position and orientation of the Dewar must. usually only the nearest neighbors need to be taken into account. . c. 1987). can be calculated from Eq. Erne et aL.fVN)T. This introduces a random error in the location of the current dipole determined on the basis of the experimental field values.. . et aL. (130). Three small coils were mounted on a thin fiber- . Rev. 1989). Then. and system design. ter Brake.: Magnetoencephalography Furukawa Electric Corporation (see the Appendix for address. Mod. 1985. 65. the nasion (the deepest point of the nasal bone between the eyes). 0 = (H-ML-1)0' .

After proper anti-aliasing. a maximum flow rate of 2 . the magnetometer electronics. The generation site of NIOOm is specific to the stimulus: its location depends. positioning devices. etc. 9. 1990). The position of the coil set with respect to the ear canals and the nasion is determined with the help of a 3D digitizer (Polhemus Navigational Sciences. The locations of the head coils are obtained by a reference measurement: the 3D coordinates of chosen landmarks on the head are determined by touching them with a pointer that has a similar orthogonal-coil transmitter. and the resulting fields are measured.or a 7-channel axial first-order gradiometer and the more recent studies with the 24. which gives an alarm if the subject has moved his head during the measurement. The early recordings were made with a 4. The position of the Dewar can be determined with an rms error of about 1 mm. and electronics. the solution of the inverse problem based on source-modeling calculations (see Sec.05 Hz. A typical off-line analysis of the results includes digital bandpass filtering of the data. Ahonen et aL. Data acquisition and experiment control 1. but the reader can become acquainted with many other interesting experiments by consulting the references given. Dewars. Responses to sound onsets In a neuromagnetic measurement.. if real-time data must be stored for later analysis. 1991) and with the new 122-channel system. This amount of data can be gathered easily. 2. However. Auditory evoked responses . Since the new devices are intended to be employed outside research laboratories and operated by personnel other than the programmers themselves. Mod. Modern multi-SQUID instruments produce large quantities of measurement data. with the most prominent deflection NIOOm peaking about 100 ms after the sound onset. in evoked-field studies. a large amount of memory is needed. proper design and realization of the software is of equal importance and requires extensive development work.. 9). 1 and 43).4 kbytes/s per channel. displays of the measurement results as time traces and field distributions. must be controlled as well. A. and most importantly. The plate is attached to the subject's head at a known point in the measurement area. The method provides direct information on the locations of the measuring sensors themselves and does not rely on knowledge of the relative position of the gradiometer array with respect to the Dewar. an intuitively clear and uniform user interface is necessary. The receiver coils are attached to the subject's head with a stretch band and the set of transmitter coils are secured to the Dewar (see Fig. often in connection with an additional high-pass filter. the Dewar is positioned. Modified versions of this approach use units containing three orthogonal coils each (Ahlfors and IImoniemi. Any abrupt sound or a change in a continuous auditory stimulus evokes a typical response. the same system is also employed with our 24-channel device (Kajola et aL. 1989. The equivalent source is deep inside the Sylvian fissure. IV). even in a system with over 100 channels. running averages and statistics should be calculated and displayed during the measurement. EXAMPLES OF NEUROMAGNETIC RESULTS In the following. 1989). but constructed without ferrous materials. Vol. Higuchi et aL. No. All the data were high-pass filtered at 0. Another device employed in neuromagnetic measurements (BTi) is based on an inductive magnetometer using three orthogonal coils in the receiver and in the transmitter. 10 and 37). 39). a reasonably fast computer is needed. as in studies of spontaneous brain activity. The system may also include a laser-based position-control device. easy use and effectiveness of the analysis system is vital.470 Hamalainen et aL: Magnetoencephalography glass plate. the signal of interest is usually limited to the frequency band below a few hundred Hz. Phys. 1989. The location and orientation of the head with respect to the 122-SQUID gradiometer array is found by a least-squares fit of magnetic dipoles to the test data. van Hulsteyn et aL. 65. at least in part. be processed and analyzed in real time. It is a modified version of a commercial 3D digitizer (Polhemus). The earliest auditory response to a click peaks 19 ms after the stimulus. we describe recordings of evoked and spontaneous magnetic fields of the human brain. in the primary auditory cortex (Scherg et aL. since the signal must. see the Appendix for address). probe- The experience gained during the last ten years has shown that M E G is well suited for studies of the auditory cortex (Hari. 1982b). therefore. for example. This means that the computer system has to be equipped with real-time interrupt handling capabilities. The examples arise mainly from our own work at Helsinki. offset. In addition. The activity continues for a few Rev. the coils are energized sequentially.and 122-SQUID planar gradiometers (see Figs. thereby indicating tonotopic organization of the human auditory cortex (Pantev et aL. SQUID gradiometers. Although the main emphasis in this section has been on the hardware. 1989. including shielded rooms. sampling frequencies up to 1-2 kHz are needed to take into account the finite rollofF of the filters. For example. Similar tomotopy was observed earlier with recordings of M E G responses to amplitudemodulated continuous tones (Romani et aL. 1988). This means. or a change in the sound (see Figs. on the tone frequency. All our measurements were carried out within a magnetically shielded room (see Fig. for example. Various stimulus generators. The position and the orientation of the magnetometer can be found from the measured mutual inductances. 1989). April 1993 VI.

NIOOm. 2) are recorded best on the head midline. originally based on a comparison of source locations and external landmarks on the skull (Elberling et aL. Williamson. Responses to transients within the sound stimulus VP200m 0 F F 471 Recordings applying short words and noise/squarewave sequences as stimuli have indicated that the auditory cortex is very sensitive to changes in sounds (Kaukoranta et aL. respectively). 2. NIOOm' was seen in all . This response. The fricative consonant h lasted for 100 ms.Hamalainen et a/. and the white arrows illustrate the locations of the equivalent current dipoles. thereby reflecting activation of the auditory cortex in the superior surface of the temporal lobe. about 25 mm beneath the scalp.) from the parietotemporal area over the right hemisphere. (b) Isofield contour maps during three deflections. to the time difference between clicks brought to each ear (McEvoy et aL. Hari et aL 1980). As shown in Fig. In addition to the onset responses (deflections P40m. Rev. i. the main deflections of the AEF to noise bursts peaked at 100 ms and 200 ms (NIOOm and P200m. the amplitude ratios should have been stable (Hari et aL. NIOOm is larger in amplitude in the hemisphere contralateral to the stimulated ear.. 1990). ft N100m — out 0. 1984). Phys. Adjacent isofield contour lines are separated by 20 fT. (a) Responses of one subject to 400-ms noise bursts repeated once every 1. N 10 0 m £^H>>::> /: >?v(*iif j ~~^' : ^ " V N H I i ~i>~"~'V~^-^"< j •\ *«-» • V. 1992). The word was the Finnish greeting hei (pronounced hay and meaning hi). V \ \ /^W~^ ! FIG. and P200m). 1982). and Kaufman. April 1993 techniques are used (Scherg. Mod. 1993). two 320-ms stimuli. 1988. 1987). the amplitude ratio changes between the magnetic NIOOm. Subjects were lying in the magnetically shielded room with their eyes open and counting the stimuli to maintain vigilance. and the electric N100.: Magnetoencephalography hundred milliseconds. The upper recording (ant. Equivalent current dipoles (ECDs. which are from extreme measurement locations. as evidenced by several deflections of the auditory evoked magnetic field (AEF). The word hei elicited a clearly different response: NIOOm was followed by another deflection of the same polarity. Solid lines indicate flux out of.1 s. Gradiometer positions are shown by small dots. were alternately presented to the subject with a constant 1-s interval (Kaukoranta et aL. measured at the head midline. 1993. but an additional generator is operating at long ISIs. 42(a). 110 ms after the vowel onset. 1990). with maximum signals recorded in the frontocentral midline. produced by a speech synthesizer. However.e. Magnetic responses to noise bursts. Sams. No.5 pT — in 1 _J X////A//ZZL 800 ms 400 0 (b) PiOm: [>[ y v* c. From EEG data. Modified from Hari et aL (1987). peaked at 210 ms. The special advantage of MEG in the study of the auditory system is that activity of both hemispheres can be detected separately and that hemispheric differences are often directly evident from the raw data.. the auditory cortex responds even to small changes in directional stimuli. indicating that the equivalent dipole is somewhere in the middle. it is thus difficult to differentiate between responses from the two hemispheres unless proper source-modeling N100m P40m OFF post. The data have been digitally low-pass filtered at 45 Hz. et aL. Pantev et aL.<. N100. The data were obtained using the lowest three channels of a 4-channel axial gradiometer (Ilmoniemi et aL.. 43. see Sec. 44. NIOOm'. If the underlying sources were identical.C) in the auditory cortex can be employed to explain qualitatively the scalp distribution of the electric 100-ms deflection. The brain also reacts strongly to amplitude and frequency modulations (Makela et aL. \ 1 /I / S >:. In one experiment.>-'/. 2. The electric scalp potentials originating at the auditory cortex in the upper surface of the temporal lobe (see Fig. a word and a noise burst. Hamalainen. the source areas change slightly as a function of time (see Fig. the head. 43). has received support from combined MEG and MRI studies (Yamamoto et aL. 1988). III. for example. The dipolar field pattern of the NIOOm deflection was interpreted as activation in the auditory cortex.) is from the frontotemporal and the lower one (post. Moreover. The equivalent dipole of NIOOm is perpendicular to the course of the Sylvian fissure and points towards the neck. the sources of the electric and magnetic responses do not seem identical: when the interstimulus interval (ISI) is increased from 1 to 16 s. 1987). Polarity reversal of the signal is obvious from the two responses. Makela et aL. and dashed lines flux into. As is evident from Fig. 65.. 1980. This result. This behavior can be explained by assuming that N100 and NIOOm reflect activity of approximately the same sources at short ISIs. recorded from one extremum of the field pattern. Recently an extra source of magnetic responses has been found in the auditory association cortex (Lii. the response to sound offset (OFF) is also seen. the measurement area is shown by the head insert. Vol. 1987.

The recordings were made from the right frontotemporal field extremum. April 1993 An effective method for studying the reactivity of the human brain is to use an "oddball paradigm" in which infrequent deviants are randomly interspersed among monotonously repeated standard stimuli. subjects and on both hemispheres. intensity. 44. although NIOOm' was originally elicited by words. The slightly more anterior sites for M M F . Periodicity pitch experiment: (Top) amplitudemodulated (depth 100%) noise bursts were used as stimuli.. (b) Averaged magnetic responses to three words. research is still in progress to discover responses that are specific to speech sounds. Furthermore. suggesting different generators for these two deflections. all generated by a speech synthesizer. in comparison with NIOOm (Sams. This type of comparison process may be associated with the auditory sensory 1 1 1 ! ! 1 1 1 ^n«a"> <8° Hz> (a) i 100 ms 1 l l l l l l l l l l l l l l l l l i l l l l ! deviant (240 Hz) MMF N100m-—Jv/ (b) '•' "'• 40 fT/cm 1 (c) /•. and to changes in the interstimulus interval.. M M F peaked later and was stronger in amplitude than the response to the standard stimuli. the dipole for NIOOm' was statistically significantly anterior to that of N 100m. D = deviant. so-called mismatch fields (MMFs). When the duration of the fricative h was increased from 100 ms to 200 ms and then to 400 ms by replacing hei with hhei or hhhhei. (Bottom) principle of the oddball experiment: S = standard. 65.y 1991. Kaukoranta et al.: Magnetoencephalography 0. but is dependent on the direction of the change. et al. Phys.. Different types of deviations within a sound sequence evoke strong magnetic responses. the NIOOm' deflection increased in latency consistently with the length of the fricative. and each trace is an average of 120 responses.or s. in this case triggered by a change in the periodicity pitch. Mod. Only words beginning with voiceless fricatives. f.h. and very similar pairs of responses were observed: NIOOm after the sound onset and NIOOm' after the noise/square-wave transition (Makela et al. In eight of the 14 hemispheres studied. M M F may reflect a mismatch process between the neuronal representation of the standard stimuli and the input caused by the deviant (Naatanen. 44). suggest the existence of two different subpopulations of neurons.5 pT N100m / . (a) Magnetic responses to noise bursts (dashed lines) and to hei words (solid lines). Several control experiments were performed to ascertain which features of the stimulus determine the occurrence of NIOOm'. Consequently. 1992). the passband was 0. 2. separations between the source locations never exceeded 15 mm. Modified from Kaukoranta et ah (1987). Later. s s I 0 100 ms s s s D S S n n n n ni n n FIG. Auditory responses to noise and word stimuli. it would be interesting to see whether it is possible to detect in the human brain phoneme maps which have been suggested to evolve as a result of self-organization in neural networks (Kohonen. For a description of the electric counterpart of this response. Using magnetic recordings. it was not related to the speech character of the stimuli. . Mismatch fields 600 ms FIG. the NIOOm' response was interpreted to be related to the acoustic rather than phonetic features of the stimuli.472 Hamalainen et al.NIOOm' _L 0 (a) 200 ms 3.05-70 Hz. the finding that only ssseee and not its mirror word eeesss elicited an NIOOm'-type signal indicates that this response is not generated by the change as such. Rev. However. Hari. Vol. determined by the gating frequency of a noise burst. noise/square-wave sounds were used to mimic acoustically the hei word. 1984). In other words. The upper and lower traces are from the two ends of the Sylvian fissure. 1988). MMF-type deflections have been recorded in response to small and rare changes in frequency. 45. 1990). or in the duration of a tone. Thus M M F depends. and followed by vowels resulted in an N 100m'-type deflection of similar magnitude and wave-form as that observed for hei. peaking always about 110 ms after the vowel onset (see Fig. in addition to the physical quality of the sound. (Middle) responses of one subject to 80-Hz standards (solid line) and to 240-Hz deviants (dashed line). For example. Figure 45 illustrates such a M M F . No. M M F s are not present in responses to deviants delivered in the absence of standards. In fact. The type of the vowel was not critical for the occurrence of N 100m'. see the review by Naatanen and Picton (1987). one giving rise to NIOOm and the other to M M F . the brain areas where these effects take place have been localized. also on the sound's position in a sequence.

Sams. 2. on a schematic brain. In single-cell recordings made directly from the cortex. and 580 responses to tones and 80 to omissions were averaged. 1989). koivu. The responses to attended and ignored stimuli start to differ at about 120 ms. (Bottom) Magnetic responses of one subject to regularly occurring tones (continuous lines) and to stimulus omissions (dashed lines). Creutzfeldt. This agrees with regional cerebral blood-flow studies where cortical areas in the frontal lobe were activated while the subject paid attention to the rhythm of auditory stimuli (Roland et ah. 65. The storing accuracy of this hypothetical short-duration memory trace seems to be good. We have studied the effects of selective listening on the auditory M E G responses evoked by simple words (Hari. The response to the omitted stimulus reflects the brain's reaction to the expected. 4. 1981). 45). Therefore. which has been studied extensively by means of electric evoked potential measurements (Naatanen. the approximate source areas for responses to tones and to omissions in five subjects. We can thus conclude that recordings of M M F s may give important objective information on the human sensory memory. all beginning with k and ending with a vowel. in the listening condition. An extreme example of deviance is the omission of a stimulus in an otherwise monotonous sequence of tones. The task of the subject was either to ignore all the words by concentrating on reading a book or. to pay attention to the stimuli by counting the target words.. A 7-channel magnetometer was used. 46. 1990. The omission responses depended strongly on the attentional state of the subjects and were often missing if they ignored the tones. 473 FIG.: Magnetoencephalography memory. et ah (1993) recently showed that M M F data agree with behavioral experiments conducted to determine the duration of the auditory sensory memory. Altogether 58 words presented in random order at an interval of 2. Mod. 1978. Clear deflections could be detected 9 0 . Hamalainen et ah. the result indicates that information on the stimulus (gating frequency of the noise burst) is kept in memory for a few seconds. half of Rev. the frontal lobe seems to be involved in this experiment associated with time perception. for example. Adapted from Joutsiniemi and Hari (1989). input.2 5 0 ms after the omission. The stimuli were five-letter Finnish words. and the rest other meaningful Finnish words. No. recorded at one end of the right Sylvian fissure. 1992). selective listening has changed the neuronal activity in monkey and human auditory cortices (Benson and Hienz. Effects of attention Attention is an important cognitive function.Hamalainen et a/. names of animals or plants. The ECDs of the omission responses are anterior and superior to those of NIOOm. During attention the sustained field (long downwards deflection) was considerably stronger. 1987). kalja. Figure 47 illustrates some results of this word categorization test. based on data from five subjects. (Top) Regions of source locations for responses to tones (NIOOm) and to omissions are shown schematically. Phys. like kukko. with large interindividual variability. both giving estimates of about 10 s. which outlasts the stimulus. Since most responses recorded earlier have arisen from primary projection areas of the cortex. Vol. April 1993 the words were targets. Selective attention is needed. since an M M F is elicited by very small deviations which are near the psychophysiological detection thresholds. Omission experiment. A very similar increase of activity during the stimulus was seen in a duration classification task with 425-ms and 600-ms . The transient response peaks 100 ms after the onset of the word and is followed by a sustained field (SF). the possibility of recording responses outside these areas is promising for the future of M E G . Figure 46 shows magnetic responses of one subject to regularly occurring tones and to tone omissions.3 s. In the above example of an M M F produced by a change in the periodicity pitch (see Fig. Hari. Joutsiniemi and Hari (1989) recorded responses in a situation where 10% of the otherwise regularly presented tones were omitted randomly. but missing. The field maps and the E C D s suggest activation of the auditory cortex at the upper surface of the temporal lobe. Figure 46 also illustrates. to pick up a desired discussion among mixtures of sounds during a party.

suggesting that both temporal lobes participate in the early analysis of speech signals. but never as pa. indicating that the observed effect is not speechspecific. whereas 2 0 % were 100 ms long. The auditory stimulus was the Finnish syllable pa presented once every second.. Furthermore. Modified from Hari et al. The traces are averages of about 150 single responses. 5. The field maps are shown during the NIOOm deflection and the sustained field (SF) for both conditions. which is usually dominant for speech production and perception. One interesting example of the integration of auditory and visual speech perception is the striking illusion called the " M c G u r k effect" (McGurk and MacDonald. 50. Effect of attention on responses evoked by auditorily presented words. The benefits from lip reading become particularly evident in a noisy environment or when hearing is defective. Auditory evoked magnetic responses from one subject in an attention experiment.. but with equal probability. (1989). Only responses to the standard 50ms tones. both attended and ignored. were analyzed. The signal consisted. The task of the subject was to count the number of longer tones among either the highor low-pitch stimuli. 1991). in 16% of the cases. while magnetic responses were recorded over the right hemisphere (see Fig. The stimulus sequence is shown schematically in the upper right corner: 3-kHz and 1-kHz tones were presented randomly to the same ear. Mod. Neuromagnetic measurements can thus be used. In the study of Rif and co-workers. The sounds were delivered to both ears simultaneously and a videotaped female face. Hamalainen et a/. not only for picking up evoked responses from specific sensory areas. about 1 cm anterior to the area that was activated 100 ms after any tone. Audio-visual interaction FIG. no statistically significant hemispheric differences were observed. 1 k H z and 3 kHz. 48). was articulating. 1989. tones. 47. not just the left hemisphere. Aulanko. The origin of the coordinate system. The subject was either ignoring the stimuli by reading (solid trace) or listening to the sounds during a word categorization task (dotted trace). et al. but also for investigating the neurophysiological basis of cognitive activity.. Magnetic responses to both stimuli are shown in Fig. 1976). Rev. Field patterns indicated that the change in the response wave form reflected an altered stimulus processing site at the auditory cortex. (1991). 1991. typically. and the x axis forms a 45° angle with the line connecting the ear canal to the eye corner. 65. 2. Phys. depending on the interstimulus interval (Hari. No. The recording location of the 24-channel instrument is depicted on the schematic head. Vol. The isofield contours are separated by 20 fT and the dots illustrate the measurement locations. in 84% of the cases. M E G is particularly suitable for this type of study in which good spatial and temporal resolution are needed. tones of two different pitches.474 Hamalainen et al. April 1993 Seeing the articulatory movements of a speaker's face provides complementary information for speech comprehension. 48. The subjects reported that they heard the discordant stimulus as tay in a few cases as ka. the visual articulation was the discordant syllable ka. the mean duration of the words is given by the hatched bar on the time axis. . A r t h u r et al. seen through a hole in the wall of the shielded room. the difference peaked at 200 ms. We recently applied this phenomenon to find out at which level the information obtained from seen and heard speech is integrated (Sams.: Magnetoencephalography Voluntary attention seems to gate relevant input entering the cortex and to inhibit processing of the ignored input. The majority (80%) of the tones of both pitches were 50 ms in duration. Rif et aLy 1991). of three consecutive FIG. is 7 cm backwards from the eye corner. shown on the schematic head. The experimental setup is illustrated in Fig. the concordant syllable pa but. 49. digitally low-pass filtered at 45 Hz. Modified from Rif et al. Responses to attended tones differed from those obtained during the ignoring condition. were presented to the left ear in a random sequence. Further studies have indicated that the activities of at least two areas in the supratemporal cortex are modified by attention. The responses are shown to 1-kHz tones when they were ignored (solid lines) or attended (dotted lines).

Articulation: pa pa ka pa Sound: pa pa pa pa Perception: pa pa ta pa main deflections. 1973). 1975. No. Cahill and Scheich (1991) recently showed that visual conditioning changes the activity of the gerbiPs auditory cortex. in which different parts of the visual field are mapped to different areas./. which may even be a puppy with a moving mouth. However. whereas the dotted lines show responses to the same auditory stimuli when the visual input was the incongruous ka (V^A). Rev. showed the same initial behavior.\ Magnetoencephalography 475 FIG. Phys. giving rise to the "wrong" perception and thus producing the M c G u r k effect.. it will be of interest to study interactions of stimuli arising from the other sensory modalities. speech can be seen before it is heard: the lips are in the required position even hundreds of milliseconds before the syllable emerges as a sound. Such experiments will become easy to perform with instruments covering the whole head. all stimuli evoked identical deflections. 49. Their study utilized 2-deoxyglucose labeled with a radioactive marker to indicate brain areas of increased glucose consumption. The results indicate that visual information has an entry into the human auditory cortex.V^A 16% 6 1?^w%/^ 10 '-K^t^i^^ u/^ "McGurk effect" 9 &*p###BOT_TEXT###amp;» ^/)^\y^ 100 fT/cm FIG. Mod. et al. 100. peaking around 50. All these signals could be explained by activity in the auditory cortex. April 1993 the very vivid auditory nature of the M c G u r k illusion. Vol. 2). It is interesting to note that the observed neural activity correlated with the auditory perception rather than with the sound. The auditory pa would then activate more vigorously the ka or ta than the pa detectors.5 cm anterior to sources of NIOOm. Data from the McGurk experiment. and the lower visual field is mapped to the upper visual . information from the left visual field goes to the right hemisphere and vice versa.Hamalainen et al. respectively. Responses to the discordant stimulus. Stimuli employed in the MEG study of the McGurk effect. For details. M E G has been used to study the retinotopic organization of the occipital visual cortex VI (see Fig. B. These units evidently develop during childhood as a result of learning a language. see text. 1975). 2. Auditory perception is biased by vision also during "ventriloquism": speech sounds are perceived as coming from the visually observed speaker (Jack and Thurlow.. The solid lines indicate responses to auditory pa stimuli when visual and auditory inputs were congruent (V = A). 50. Aulanko. The visual ka information might prime in the supratemporal auditory cortex those neurons that are tuned to the presence of the k and t consonants followed by a vowel. Teyler et al. 65. Furthermore. Visual evoked responses 1.3 0 0 ms. _ N100m . Interestingly. for example. but visual stimuli shown alone did not produce any magnetic activity at the auditory cortex. It remains to be seen whether activity of the auditory cortex is modified also during this phenomenon. in agreement with V-vW^^' '**X*HJ^ 2 ^^4^-— ^W *. 5 ^^s/ v = A (84% ___. (1991a). Eighty responses to deviants and 500 to standards were averaged. the later part of the signal was clearly changed in relation to responses elicited by the concordant stimuli (V = A). and 200 ms. visual ¥=" auditory.. the difference peaked at 2 0 0 . In face-to-face communication. When the same auditory sound sequence was presented in the absence of the visual input. The source of the difference wave form (responses in the discordant condition subtracted from those in the concordant condition) was about 0. Modified from Sams. Responses from occipital areas Visually evoked fields (VEFs) were reported for the first time in 1975 (Brenner et al.

One convincing piece of evidence for the existence of facespecific brain regions is prosopagnosia. IV. Responses to stimulation of the lower right visual field with an octant checkerboard pattern 1. The superimposed traces depict two successive measurements at each site.. Rev. The maximum signals were obtained over the left hemisphere near the midline. on knowledge about the distribution of retinal receptors and on psychophysical experiments. In M E G studies. 2. 52). The approximate location of the instrument is shown on the schematic head. 1983). i. but the job is difficult for computers. 52 illustrates the mean source locations for six subjects on the left hemisphere and for 15 subjects on the right. The depths of the sources of face-evoked responses varied between 34 and 38 mm. Phys.Hamalainen et al. 52. 65. in severe cases patients may be unable even to identify spouses. April 1893 • V • \*% / VEF \ 95. while the inset below illustrates the minimum-norm current estimate (small arrows. see Sec..g. the inability to recognize previously known familiar faces. the field patterns suggested activation of areas outside the occipital visual cortex. Lu et al. The response over the sides of the head consisted of two transient deflections at 150 and 270 ms and a late.. 51. The first source area was near the junction of the oc- Recognizing one's fellow citizens is fundamental to human social life. with the distance from the central visual field. . Responses to faces in humans as well by using positron-emission tomography (Lueck et al. 2. several visual areas are known to exist in addition to the primary visual cortex in the occipital lobe. No. Humans are very good at this task. even under adverse conditions. This disorder is usually associated with bilateral brain damage in the occipito-temporal region. M E G data can also be employed to estimate the cortical magnification factor (Ilmoniemi and Ahlfors. e. (1991) presented black-and-white pictures of laboratory personnel for 300 ms. Responses to faces. For reference. (a) Neuromagnetic responses to face stimuli over the right hemisphere for one subject. 1990). once every second in a random sequence. and it is largest for central vision. the E C D moved deeper with increasing stimulus eccentricity. indicating retinotopic organization (Maclin et al. (b) The average xy locations of the ECDs over the left hemisphere (six subjects) and the right hemisphere (15 subjects) for the three main peaks in the responses to face stimuli. Figure 51 shows responses to an octant checkerboard pattern in the right visual field. Modified from Lu et al (1991). the position of the source of the auditory NIOOm response to short tones is also shown. which previously was based on less direct information. VEF indicates the source of the 100-ms response in the occipital visual cortex. During each peak. Vol. Mod..l) and the location of the equivalent current dipole (shadowed arrow).-160 ms 190 . and our brains probably contain areas that are specifically activated when we see people's faces.I. measured with a 24-channel gradiometer. slow deflection around 500 ms (see Fig. The schematic head shows the measurement locations at the back of the head.5° in extension. Modified from Ahlfors et al (1992). 1989) and by M E G in the experiment to be discussed next. In monkeys.: Magnetoencephalography 476 cortex and vice versa.5 6 0 ms N100m • vm%y w^0)C^^ A V \*5 FIG. The lower part of Fig.2 9 0 ms 380 .. The magnification factor tells how much of the cortical surface is devoted to a certain part of the visual field. with the main deflection peaking at 130 ms. Twenty responses were averaged for each set of traces. Some evidence for similar areas has been obtained 3 w^X^s^jr** ^\ 4 j y >^\ wVVcMA* 10 12 w ^ \ ^ ^ ' 100fT/cmL 100 ms FIG.e.

It is assumed that the left ankle was stimulated. in contrast to direct stimulation of the cortex applied during brain surgery. when this information is available from other experiments. in general. 1950. The corresponding E C D locations (see also Suk et aL. Recording of M E G responses is completely noninvasive. 52) in the sequence: 100 ms (occipital lobe) —• 150 ms (occipital-temporal junction)—>270 ms (inferior parietal lobe)—>500 ms (temporal lobe). Adapted from Karhuef ah (1991). with the known somatosensory homunculus (Penfield and Rasmussen. The same is true for stimulation of other parts of the body. Somatosensory evoked responses 1. one at 20 ms (N20m) and the other at 30 ms (P30m). it is not easy to see directly from the E E G patterns that there are three underlying sources. 3) is well reflected in the distribution of somatosensory evoked fields (SEFs). Phys. most probably in the second somatosensory area S2 (Hari et aL. 1954. The magnetic patterns are clearly separate over each source. Responses from S2 M E G recordings have indicated that peripheral stimuli also activate other sources bilaterally.Hamalainen et a/. see Fig. 1984. 1984. Kaukoranta. but the two other sources were not. 53. 1984) already indicated different representations for the thumb and the little finger. Median nerve 477 TV^Z N_20m j\ ^lC\^-« P30m [lOOfT/cm Tongue <—I 1 0 nA-m I 1—H 1 ^-^s^******^ '-W^ TWA*9" tv^" 2. 3) and of the S2 areas in both hemispheres. and orientations of the ECDs to median nerve and tongue stimulation. strengths. Differentiation between temporally overlapping SI and S2 activities can be done by a two-dipole time-varying model: the source locations are first found and their . whereas the main deflection of the response to tongue stimulation peaks at 55 ms (P55m). Responses from S1 -^-2 100 ms P55m FIG. whereas the electric potential is more widely distributed. 3). Somatosensory responses to stimulation of the median nerve at the wrist (above) and to the lingual nerve in the tongue (below). the correct solution can be found from the E E G pattern as well. No. Figure 53 illustrates responses to alternate wrist and tongue stimulation. and the third in the middle of the temporal lobe. the second in the lower part of the parietal lobe. in the former case the sources are shown for the two main deflections. 1978. close to the upper Rev. All three locations were significantly different from each other and clearly separated from the occipital visual areas. Interestingly. and showed that the sources activated by stimulation of the ankle were in the inner wall of the contralateral hemisphere. resulting in activation of SI on the inner surface of the opposite hemisphere within the longitudinal fissure (see Fig. Consequently.: Magnetoencephalography cipital and temporal lobes. Okada et aL. an area which shows considerably stronger responses to faces than to other complex visual stimuli.. This investigation is a good example of how magnetoencephalography can be used for noninvasive studies of signal processing in the brain. Hari et aL. as well as to upper and lower lip stimulation. 65. *V^ AA TX^L NKOm 0 The organization of the primary somatosensory cortex SI (see Fig. Vol. on both sides of the head. Mod. lip of the Sylvian fissure. were thus oberved to become sequentially active by the face stimuli. 1986). The traces are averages of about 350 responses. The inset shows the approximate locations. With M E G it may thus be possible to monitor plastic changes of cortical organization following various lesions in peripheral nerves or in limbs. April 1993 h- **y'*v\*« 6 . M/^X*/ "V C. 1990. The dashed line in the inset illustrates the approximate course of the Rolandic fissure. The source of the late (380-560 ms) activity might correspond to the superior temporal sulcus in monkeys. Penfield and Jasper. The same stimulus activated four brain areas (see Fig. The early studies (Brenner et aL. The parietal source was also activated by pictures of birds and simple drawings. The response to median nerve stimulation has two prominent peaks. Three areas outside the occipital visual cortex. However. the first source location agrees with clinical observations on humans showing that prosopagnosia can result from damage in the occipito-temporal junction. Figure 54 gives a schematic illustration about the differences of M E G and E E G recordings in differentiating between activities of SI and S2. 2. 1991) agree. These sites had not been located earlier in humans. Hari et aL. digitally low-pass filtered at 190 Hz.

The S2 responses. data from several sites measured at different times can be combined fairly safely. and electrical stimulation of the skin were used (Hari et aL. and Kajola. On the other hand. Painful stimulation FIG. were almost 3 times as high in amplitude to deviants than to standards (see Fig. the skull. with the sources in the vicinity of the calcarine fissure (the site of the primary visual cortex). After thumb stimulation. the stimuli applied to one finger did not have an observable effect on neural populations responding to stimulation of the other finger. with simultaneous activity at SI in the inner central wall of the right hemisphere and at both S2 areas. Modified from Kaukoranta. Alpha rhythm Studies of spontaneous brain activity benefit greatly from recordings with multisensor arrays. 55). In an earlier study (Hari et aL. April 1993 Responses have also been recorded to painful stimuli. a long-latency peak is seen at 130-140 ms. the source area of the earliest response was. Huttunen et aL. In other words. In evokedresponse experiments. The distribution agreed with a two-dipole model. in agreement with the somatotopical order of SI. it might be valuable to monitor these pain-related responses whose source areas are now known. 1990). strengths are allowed to change as a function of time (see Sec. corresponding to the brain. discussed so far." In objective evaluation of different pain-killer drugs.6 cm beneath the scalp. Later. peaking at about 100 ms. measured with the 122-channel whole-head magnetometer (see Fig.. and j smaller at the left S2. 3. symmetric with respect to the cortical midline and 4 . Joseph et aL. whereas long-latency signals were still seen over both S2 cortices. 56 which shows responses to stimulation of the left tibial nerve at the ankle. 1988). above the foot area in SI. Consequently. which is damped by opening of the eyes. 1991). for example. The magnetic a signal was selectively suppressed on one side by stimulation of the left or right visual fields separately. No. et aL (1986). and vice versa. Vol. a phase reversal was observed between signals measured from the right and left hemispheres. is visible at the top of the head. but no evidence has been obtained for a localized cortical "pain center. Differentiation between signals arising at the SI and S2 cortices is very clear in Fig. a multitude of a sources were found for signals measured with a 4-channel magnetometer (Vvedensky. The first cortical response. These results indicate that an accurate functional representation of different body parts is maintained in the human S2 as well.478 Hamalainen et aL: Magnetoencephalography interval. When the median nerve was stimulated at the wrist. such a model shows maximum activity in SI at 45 ms and at 145 ms. Schematic illustration of electric and magnetic patterns due to activity in the first and second somatosensory cortices SI and S2. responses from S2 were obtained using an oddball paradigm. A similar amplitude enhancement was obtained when the deviants were presented without the intervening standards but with the same long interstimulus Rev. Hari. carbon dioxide stimulation of the nasal mucosa. 1985) and with two 7-channel magnetometers simultaneously (Ilmoniemi et aL. The standards (90%) were presented to the thumb and the deviant stimuli (10%) to the middle finger. one at each parieto-occipital area. . The dipole strengths were assumed equal at SI and at the right S2. a few centimeters lateral to the midline in the contralateral hand SI area. Mod. Phys. The shadowed areas indicate magnetic flux out of the head and positive electric potential. The appearance of the magnetoencephalogram of an awake subject resembles the well-known E E G : the parieto-occipital areas show 8-13-Hz a activity. 65. many spontaneous brain signals are so strong that they can be recorded without averaging. Spontaneous activity 1. 2. 1968). and the scalp. the cerebrospinal fluid. Chapman et aL (1984) used the relative covariance method with an electric reference at the vertex. 41). 54. IV. 57). but in recordings of spontaneous activity. On both sides. whereas activity in S2 reaches its maximum at 115 ms (see Fig. the activated areas were around S2. peaking at 42 ms. and the oscillating source currents were suggested to be parallel to the longitudinal fissure between the hemispheres. the head was assumed to be a spherical conductor with four layers of different conductivities. 1983. IImoniemi. In the simulations. The map showed two extrema. reflecting activity at the left and right S2 cortices. D. The situation corresponds to that observed after left-sided peroneal nerve stimulation at the ankle.H). it can be concluded that painful stimuli activate certain cortical areas. In a comparison of the a sources and anatomical information obtained from M R images (Williamson and Kaufman. 1986. this is not possible because every event is unique. Electrical stimulation of the tooth pulp. In the first study of the human magnetic a rhythm (Cohen.

but about 1 cm deeper. This signal.5 p T / c m . No. evoked by stimulation of the median nerve at the wrist (see Fig.Hamalainen et al.: Magnetoencephalography S1 -source 479 1 "TV 30 fT/cm I 100 ms iC^V ^ S1 -source 45 ms S1+S2-sources 1 _ ^ >*^ 2 S2-source [10 nA m 115 ms goodness-of-fit (%' S2-source both sources / . which can be recorded over the Rolandic fissure. Mu and tau rhythms Another well-known spontaneous E E G activity. 1989). therefore. April 1993 equivalent dipoles of the signals were clustered at the same vertical height as the sources of the visually evoked responses to checkerboard stimuli. 1989). These M E G results confirm in humans the generation sites of the a rhythm. This effect was found by monitoring the average field variance at the a frequency. 2. Kaufman et al. Adapted from Hari et al. N o dipoles were observed with orientations that would suggest activation of the wall of the longitudinal fissure. sources of strong oscillations were found in the region of the fissure between the parietal and occipital lobes (see Fig. a single dipole at the second somatosensory cortex S2. 55. The continuous traces illustrate the original measured signals and the dotted lines those predicted by the models. consists of 10.and 21-Hz components (Tiihonen et al. the stimuli were presented once every 2 s. has a magnetic counterpart as well. This corresponds to a magneticfield strength of several picoteslas at the maxima. The . The maximum field gradient during the eyes-closed condition was about 0. (1993).. and these two sources operating simultaneously. 53). Rhythmic 10-Hz activity is prominent when the eyes are closed. most probably arise from the calcarine fissure or from its immediate surroundings. (1990) have demonstrated that the occipital magnetic a activity is suppressed for about 1 s during a visual memory task. 2. but it is blocked by the opening of the eyes. Either this part of the visual cortex does not display a oscillations or their fields were canceled out due to bilateral activity on the opposite walls of the same fissure. Mod. Three different source models were applied: a single dipole at the primary somatosensory hand area SI. S1-source 80 h 200 ms FIG. the /LL rhythm. The detected a oscillations. and 55 responses were averaged. 65. Figure 58 shows recordings of spontaneous magnetic signals in one subject over the right occipital lobe. The signals were digitally low-pass filtered at 90 Hz. Responses over the left hemisphere to electric stimulation of the right thumb. they are clearly damped by a clenching of the fist. The figure at the lower right corner depicts the dipole moments of the two sources as a function of time and the goodness-of-fit values for the three models. 2). detected earlier by microelectrode studies in animals. The ECDs of the fi rhythm were within 1 cm of the source area of the 20-ms SEF deflection N20m. Vol.. The Rev. Phys. with most sources within 2 cm from the midline.

1 0 H z (Loveless et al. Figure 61 illustrates 24channel magnetic recordings and some electric derivations on one subject during different stages of sleep. Kajola. 1968). 41 and 42). was detected (Tiihonen et ah. such as bursts of noise. During R E M sleep. This is also an 8-10-Hz oscillation. thereby reflecting the importance of the hand. another spontaneous M E G activity. Figure 62 shows source locations for two subjects during the V-waves and K-complexes.^ 4 2 ms -V^. 59). The recording was made with the whole-head 122-SQUID magnetometer (see Figs. The field gradients are shown separately in the polar direction (left) and in the azimuthal direction (right).y 1989). rhythm thus seems to be most prominent in the somatomotor hand area. The r rhythm may affect the reactivity of the cortex to external stimuli. sharp vertex waves ("V-waves") of short duration were seen in the frontocentral E E G leads (Fz and Cz) and on several M E G channels.Hamalainen et a/.-W. 6 and 12 Hz in frequency. 56.05-180 Hz. the recording passband 0. Responses to stimulation of the left tibial nerve at the ankle. and about 150 single responses were averaged. which were classified on the basis of both E E G and M E G data. with the dipole locations within 2 cm from the source area of the 100-ms auditory evoked response (see Fig. / 139 ms ~vs^ 131 ms ^^v FIG. for human sensorimotor organization." During deep sleep (stages 3 and 4. The spatial distribution of the r rhythm suggested activity near the auditory cortex. but it was not damped by opening of the eyes. In fact. During deep drowsiness and light sleep (stages lb and 2). helping the system to avoid starting from "cold" when stimulus processing becomes necessary. Hari. Recently. thereby resembling the typical electric "Kcomplexes. Spontaneous activity during sleep During sleep.-yN. Spontaneous 10-Hz rhythms may either have an idling function.: Magnetoencephalography 480 Polar derivative Azimuthal derivative 50fT/cm L 100 ms i^tsA . (1990) after sensory stimulation. The traces were digitally low-pass filtered at 140 Hz. The ISI was 2 s. M E G activity was lower in amplitude and faster in frequency than during the other stages of sleep. behaving thus clearly differently from the occipital a rhythm. and especially the thumb. The spindles were occasionally superposed on high-amplitude transients. Occasionally the r rhythm was reduced by sound stimuli. 60). which would correspond to intrinsic frequencies of 5 . 1991. 1993). From Hari et ah (1993). the r rhythm. 0. defined on the basis of rapid eye movements and decreased muscular tone. During light drowsiness (stage la). characteristic changes are seen in the amplitude and frequency of the spontaneous E E G activity (Rechtschaffen and Kales. The ECDs are clustered in an area of 4 X 4 cm 2 in the xy plane. best seen in the planar gradiometric recordings just over the auditory cortex (see Fig.y 1992).5-2-Hz polymorphic slow-wave activity was widely spread over the measurement area. et al. Magnetic spindles of 11-15 Hz in frequency appeared during stage 2 as well. or they may synchronize neuronal populations which are necessary for perception. 3. a few . Other types of rhythmic magnetic signals at the somatomotor area. rhythmic r activity appeared over the temporal lobe. The first multichannel M E G recordings of sleep were reported only recently (Lu. have been recorded by Narici et al. an enhancement of evoked responses to the second tone in a pair of tones occurs when the sound onset asynchrony is 100-200 ms.

The sources underlying the sleep spindles were complex: it was not possible to account for their distribution with a single-dipole model. has also been suggested on the basis of animal experiments (Buser. These "fissures" form an angle of about 45° with the Sylvian fissure (see Fig. 59. Phys. The frequency spectra were calculated from 20-s time sequences. Mod. Alpha rhythm on one gradiometer channel from the occipital area when the subject had his eyes closed or open. Vol. Seven-channel recordings of responses at two locations (shown schematically on the inset brain) to frequent electric pulses presented on the left thumb (solid lines) and to infrequent pulses delivered to the middle finger (dashed lines). Since the M E G signals originate primarily from currents in the fissural cortex (see Sec. The signals were lowpass filtered at 40 Hz. and to compare electric and . The sources of the V-waves and K-complexes are in the lower part of the parietal lobe. April 1993 centimeters toward the parietal lobe from the sources of NIOOm. The responses have been digitally low-pass filtered at 90 Hz.: Magnetoencephalography * * * * * 481 ^ | ^ [lOOfT A WffpW EOG V ^yw~ 1s FIG. 2. with a multitude of sources and frequencies.. No. whose course was derived from the orientation of the E C D for the NIOOm deflection of the auditory evoked response. Eyes closed 100fT/cm 1s Eyes open Eyes open Eyes closed 30 -j OJ 0 10 20 Frequency (Hz) 0 10 20 Frequency (Hz) FIG. 65.Hamalainen et a/. Prominent r oscillations can be seen. 1982). 2). III. which may thus be transiently activated during sleep. In order to be confident that the E C D s really are physiologically meaningful. and its activation seems to agree with the early hypothesis that Kcomplexes are linked to arousal and orientation during sleep (Roth et a/.E. the E C D s were distributed more widely. EOG refers to electro-oculogram. 1956). This region of the brain is involved in the regulation of attention. The small vertical arrows define the time of stimulus presentation.2). 1987) and by depth-electrode and scalp-topography studies on humans (Niedermeyer. the figure also shows the estimated course of the fissure from which the signals might be generated. During slow-wave sleep. but they were still concentrated in the lower parts of the lateral recording sites. Modified from Hari (1993). nor with a two-dipole time-varying model. Rev.. Complexity of spindle generation. 0 100 ms FIG. 57. Spontaneous magnetic activity recorded by eight MEG channels (dBz/dx above and 3Bz/dy below for each pair) over the right temporal area from a subject who had his eyes open. 58. These estimates were obtained by drawing lines orthogonal to the main orientation of the ECDs.

The simultaneously measured E E G signal can be used as a trigger for M E G wave forms. Studies of this type were first carried out at the University of California in Los Angeles (Barth et ah. For the coordinate system. April 1993 E. attempts have been made to obtain from M E G data additional information for selection and noninvasive preoperative evaluation of candidates for epilepsy surgery. Modified from Tiihonen et ah (1991). With this method. (a) Subject 1 H u3 VMJJJ j ii (b) J 3 100 ms (d) Subje Dt1 Subj ect2 Subje c t 3 l ^y/cm i i -2-1 i i 't . multichannel devices have considerably speeded up the recordings and improved their accuracy (Stefan et ahy 1989. The isocontours are separated by 100 fT. spike-and-wave complexes. Mod. which is necessary if the data are obtained sequentially from several locations. Magnetoencephalography in epilepsy 1. including spikes. These signals are typically an order of magnitude stronger than the evoked cerebral responses and can thus be measured in real time. In focal epilepsy.. 65. 1989). simultaneous E E G and M E G recordings covering the whole head are necessary. sharp waves. Tiihonen et ah.1 % of the population suffers from epilepsy. seizures in some patients cannot be controlled adequately and surgical treatment must be considered. the head. The shadowed areas indicate magnetic flux out of. both single sources and multiple irritative brain areas have been found (Barth et ah. 1990). Since the early 1980s. Owing to lack of suitable multichannel instruments in the past. as reviewed by Rose et ah (1987) and by Sato (1990). often including invasive recordings directly from the cortex. During the last few years. 1982. No. and the white areas flux into. Source locations of the r rhythm. especially as the sources in both hemispheres seem to function separately. and irregular slow activity which can often be recorded also during the time between the seizures. Interesting results have been obtained by using spikes from electrodes on the scalp a n d / o r in other structures as triggers for M E G averaging. 47. Paetau et ahy 1990. 1991. 60. A consistent depth difference was observed in the source areas of two groups of magnetic spikes. 1982). The inset (c) depicts the measurement region with respect to the subject's head. This result illustrates propagation of the epileptic discharge between the superficial and the deeper brain areas. a common approach was to apply signal averaging. 2. Rev. Background and data analysis The most important clinical application of M E G so far is the localization of irritative brain areas in patients suffering from epilepsy. The open circles and the white arrows show the dipole locations and orientations for the 100-ms auditory response. 5 . 1991. To find the trigger location. In the preoperative evaluation of patients it is essential . For the successful outcome of an operation.Hamalainen et ah: Magnetoencephalography 482 magnetic wave forms. The epileptogenic focus manifests itself as an abnormal spontaneous activity. In western countries 0 . Maps in (a) show the magnetic-field component perpendicular to the head during the six successive peaks 1-6 in a representative section of the r rhythm shown in (b). 1984).^ FIG. a number of epileptic patients have been studied in several M E G laboratories. 1982) and at the University of Rome (Modena et ah. The white arrows in (a) show the locations and orientations of the equivalent current dipoles. Phys. see Fig. small areas of brain tissue may trigger the epileptic seizure. Vol. spontaneous brain activity should be recorded simultaneously from large areas over the head. We expect that these studies will result in a better understanding of changes in the functional organization of the human brain during different stages of sleep. 1992. Shown in (d) are the locations of the equivalent dipoles (solid circles) for ten peak deflections of the spontaneous 8-10-Hz oscillation for three subjects. triggered by activity recorded on the scalp and by electrodes put directly into the brain (Sutherling and Barth. x/p —1 —2 i m . These requirements necessitate extensive preoperative tests. Therefore. In spite of advances in antiepileptic medication. it is necessary that there be only one trigger area that can be located accurately and that it be situated in a cortical region that may be removed without serious side effects to the patient.

Rev. The x gradients at 12 locations are plotted above and the y gradients below. central. Phys. EOGv and EOGh are recordings of the vertical and horizontal oculograms.. Comparison of spontaneous MEG activity over the right temporal lobe during wakefulness and sleep. Mod. and Oz refer to frontal. 61. April 1993 . Vol. The recording site is indicated on the schematic head. Cz. respectively. In the electric channels. The electric amplitude scale refers to all other electric recordings except the EMG (electromyogram). and occipital midline locations of EEG recordings. 65. No.\ Magnetoencephalography Awake Stage 1 a *j##^#^«^ Stage 3 Stage 4 Stage 1 b iHNMflfi^^ 483 Stage 2 ilMlify^^ REM sleep 3B7/ax 3B2/ay 200 uV |M*<MMV|W^*»Ml»M'W^^t¥^<<>H'f^i» «UH*^^'*H^»4^yffrf^>k FIG. Modified from Lu et ah (1992). 2.Hamalainen et al. Fz. for which only relative amplitudes are of importance. respectively.

. 1990) has now become routine in several M E G laboratories. and the thick lines the "fissures" of the ECDs of V waves (see text for more details). 1991). The L- 100 ms Subject 2 3Bz ay -4 0 4 cm -4 0 4 cm FIG. see Fig. to determine how close the irritative focus is to crucial brain areas that should not be damaged during surgery. For the coordinate system.484 Hamalainen et al. assumed to be orthogonal to the orientation of NIOOm. (1991). (Top) Examples of spontaneous spike-and-wave discharges from the left hemisphere of a child with the Landau-Kleffner syndrome. are seen on the x derivative of units 8 and 11. The boy had 1 pT/cm 1 V 2 4 dB2 ax V-wAv^^a^^A^". ^^\. (1992). nor could he discriminate between telephone ringing and other sounds. The usefulness of this approach is also supported by results showing that recordings of somatosensory evoked magnetic fields agree with electrocorticographic data in determining the course of the central sulcus (Sutherling et al. cortical regions involved in motor control as well as in speech production and perception. Auditory evoked magnetic fields to tones and syllables were absent over the right hemisphere and contaminated by spikes on the left side. 62. 47. these include. 9. The upper traces reflect the vertical and the lower ones the horizontal gradients. 7. about 1 p T / c m . The inset illustrates ten superimposed single spikes from the y derivative of unit 8. not even his own name. In Fig. Only ECDs with goodness-of-fit values exceeding 90% have been included.y 1988). and their locations by open circles. Functional localization with respect to source areas of known evoked responses (Tiihonen et al. Example: boy with disturbed speech understanding Figure 63 illustrates epileptic 3-Hz spike-and-wave activity recorded from a 7-year-old boy who suffers from the syndrome of Landau and KlefFner (Paetau et al. The y channel of unit 3 was not functioning properly and is omitted. Source locations (black dots) for V waves and K complexes in two subjects. . well-established landmarks are needed. He first developed normally but started to lose his ability to understand speech at the age of 3. shown by white squares of ten single spikes superimposed on MRI slices. 63.: Magnetoencephalography K-complexes ~T- cm 4 1 r — . the strongest epileptic discharges. In this task. The white lines indicate transsection levels of the two other slices. Modified from Paetau et al. Modified from Lu et al. while the next largest signals are observed over the neighboring units 4. (Bottom) ECD locations. owing to severe defects in auditory discrimination..K^ #WJ|/W 10 r t ^ ^ W W v a < v ' f t v t ^ 11 *~ 12 ra. and 12. the localization accuracy approached that of direct recordings from the cortex. 63. but he could not identify speech. Subject 1 J cm 4 normal hearing thresholds and intelligence.^-A/VW^-fyH-wW\M From left From below fH fVfijPf From front FIG. The orientation of ECDs for the NIOOm responses are shown by white arrows. 2. for example. with combined M E G and scalp E E G data. Study of an epileptic patient. The thin lines in the V-wave boxes illustrate the course of the Sylvian fissure.

fall in all projections on a narrow strip in the auditory cortex along the upper surface of the temporal lobe. i. The arrow in the upper head inset shows the equivalent dipole for NIOOm and NIOOm'. several other clinical applications of M E G are emerging. In these patients the auditory nerve is directly stimulated by an electrode put into the cochlea. however. 485 bility of using AEFs as objective signs of tinnitus (ringing of the ears). 64. Most probably a local epileptic focus has disturbed the normal functioning of the child's auditory cortical areas.. are displayed in the boxes. However. 1990). can be of value in screening patients for surgery. 1991). Some recordings have been made from deaf patients who use an implanted cochlear prosthesis (Pelizzone et aL. F o r example. A lesion in the supratemporal auditory cortex resulted in the disappearance of the NIOOm deflection of the auditory evoked response only when the damaged area extended deep into the temporal lobe (Leinonen and Joutsiniemi. 63). The single-dipole model explained 9 0 . hemorrhages. Intact hemisphere 3 *wW*W 4 5 N100m' moom ^ N^W/W 7 ~fc^*w\fa?tf -*J\^W$ W W W * ww»»Vcfc# 800 ms h00fT/cm «esAaA^*nx^w«>v*^»*0< 11«*s/\**/W* 12 ~A*°w*^W« F. the single-dipole model is not always satisfactory. the main focus of the clinical evoked-potential research at present. extrapolated to a standard series of anatomical pictures. Sixty single responses to contralateral stimuli were averaged and two independent sets of measurements are superimposed. affect this amplitude ratio. The black and white bars on the time scale indicate noise and square-wave parts of the stimuli. Vol. Auditory responses of a patient over both hemispheres 16 months after a stroke.9 9 % of the field variance during all spikes. with the normal receptor mechanisms bypassed. The CT data in the middle illustrate the extent of the lesion. 65. Clear changes from one irritative phenomenon to another suggest the existence of several sources. 1989. although they do not rule out a deeper common trigger. Figure 64 illustrates the evoked responses and the extent of the lesion of one such patient. 1989). Both AEFs and spontaneous activity have been recorded recently in patients with different types of cerebrovascular accidents. Pelizzone et aL. and the spatial distribution of the field was very stable. 1987. the amplitude ratio of P200m and NIOOm h a d decreased in a patient with tinnitus that was caused by an acute noise trauma. When extended or multiple source areas are simultaneously active. . superimposed on three MRI scans (see the bottom part of Fig. as often happens during the epileptic discharge. useful information can be obtained by studying the field pattern at different time instances. Responses from the channels that recorded the maximum signal 16 days after the stroke. Phys. The sites of the two dipoles differ by less than 1 mm. Adapted from Makela and Hari (1992). which resembles the classical use of E E G recordings. can be assessed by magnetic evoked responses. Spikes were detected only on the left hemisphere. 15y months earlier. 1989). The source locations. Proper functioning of sensory pathways from the periphery to the brain.#>w*"»*»w>*«5CW~"-c ^^Y^^ff^^v^ '^KX^gWW^ fi 11 ^ y k ! V ^^<^s<j<30& 10N A ^ V ^ ^ * ^ ^ ^ ^ 12 800 ms [l00fT/cm FIG. Mod. and transient ischemic attacks (Vieth. Focal slowing of the spontaneous M E G activity has been detected in patients with brain infarctions. Pantev et aL. T h e currents causing the epileptic spikes are under the maximum signals and are approximately vertical in orientation. Hari. April 1993 Damaged hemisphere 16 days after stroke £>-AVV$K»"*A ^bc^^^wv^viU?-^ <r*$Kp*vP\tp(flfy "7 .e..Hamalainen et air Magnetoencephalography y derivatives show much smaller signals. but returned to normal during the period of recovery (Pantev et aL. 1988. 1991). the patients learn to respond to the stimuli in a similar way as they earlier reacted to real sounds. No. (1989) have suggested the interesting possiRev. Similarly. 2. the responses of their auditory cortex may resemble closely those of subjects with normal hearing. and contradictory results have been reported recently (Jacobson et aL. Makela et aL. Hoke. and some of them start to understand speech even without lip reading. Hoke et al. This type of "qualitative" measurement. Several experimental factors do. respectively. even if the source localization fails. Strokes associated with insufficient or blocked blood flow to the temporal lobe often cause marked defects in speech perception and production. In spite of this very artificial input. Other patient studies Besides epilepsy.

With these developments. applied on large patient populations. The results have confirmed features of cortical functions that earlier were observed only in animals and have also revealed information about entirely new functional principles. Inc. Nishi-ku. their locations agreed with those found in a control group of healthy subjects.. Hari. Polhemus Navigational Sciences. San Diego. Better knowledge of normal brain functions is the necessary basis for advances in understanding the mechanisms of different brain disorders as well. Finland. VT 05446. Goetheallee 1. must be taken into account.486 Hamalainen et a/. Emphasis will be put on combining information from diverse brain imaging methods. In patients with progressive myoclonus epilepsy. Neuromag.6 times larger when compared with the response of a normal subject. but the source locations agree with activation of the somatosensory hand area (Karhu.. CONCLUSIONS M E G has already proven itself a useful tool in studies of human neurophysiology and information processing. 02150 Espoo. to assess reliably the clinical value of M E G . Box 1360. Port Coquitlam. 1 Hercules Drive. 9727 Pacific Heights Boulevard. Kajolae? ah. characteristic for cortical signal processing. the actual shape of the subject's cerebral cortex. Advanced signal-analysis techniques can be applied to reduce coherent noise. Since both electric and magnetic signals are generated by the same neuronal activity. Friedrich-Schiller-University. 2.and psychophysiology. Quantum Design. and future clinical applications of M E G will depend strongly on experience obtained with these devices. for reading the entire manuscript. Germany. BC. Further attempts to develop M E G ' s clinical applications include dc recordings on patients suffering from migraine to detect magnetic signals associated with the spreading depression (Barkley et aL. as determined by magneticresonance imaging. 15-1750 McLean Avenue. 2-4-3 Okano. Colchester. Furukawa Electric Corporation. in a totally noninvasive way. With the new whole-head-covering instruments. CA 92121. and both the accuracy and the reliability of the data will improve.. In precise work. and to Professor Heikki Seppa for studying parts of it. We also thank Ms. c / o Low Temperature Laboratory. P A 19124-2705. With M E G it is possible to study neurophysiological processes underlying mental acts in healthy and awake humans. since patients can then be studied more Rev. We believe. Yokohama 220. since it can follow changes on a millisecond time scale. ACKNOWLEDGMENTS This work was financially supported by the Academy of Finland.O. APPENDIX Amuneal Corporation. Helsinki University of Technology. 5 . 11578 Sorrento Valley Road. High-quality instruments will be installed in many leading hospitals.. however. the activity in different parts of the brain can be monitored simultaneously. Mod.3 ms. a completely new era of clinical M E G should emerge. The cortical hyperreactivity of these patients is not extended to the auditory system. Karin Portin for preparing many of the illustrations. When such magnetometers are fully operational. the spatial accuracy of M E G will improve and the method will thereby complement P E T studies. No. Germany. Multi-SQUID magnetometers are being actively developed in several countries with considerable industrial backing. that the understanding that has resulted from M E G recordings is useful for everybody working on E E G . Phys. P. and by the Sigrid Juselius Foundation. 1991).: Magnetoencephalography Abnormalities have been observed in SEFs of patients with clinically diagnosed multiple sclerosis (Karhu. D-7990 Friedrichshafen. in agreement with evoked potential data. Furthermore. Additional abnormalities were common in the middle-latency responses: the 60-ms deflection was often of larger amplitude than that in control subjects. C T F Systems. independent of one's access to M E G facilities. Inc. Riitta Salmelin and Professor Claudia Tesche. Jena. M E G will probably play an important role in studies of basic neurophysiology. SEFs are 3 . VII. Hari. one should pay attention to the available M E G data when making conclusions about cerebral functions on the basis of E E G recordings. recordings of both electric and magnetic signals under similar conditions have been important in identifying the neural sources of various evoked responses and of spontaneous brain rhythms. Biomagnetic Technologies. Japan. San . which have resolved brain areas activated during different behavioral tasks and during language processing. Therefore. It is certain that during the present Decade of the Brain considerable progress will be made in human neuroscience. Dornier Systems G m b H .. Ltd. 4737 Darrah Str. We express our gratitude to Dr. Canada V3C 1M9. including simultaneous stimulation of different sensory modalities. Neuromagnetic studies are thus increasing considerably our knowledge of brain activity. 65. the Korber-Stiftung (Hamburg). April 1993 rapidly and more conveniently. and more sophisticated solutions of the inverse problem than the dipole models will become feasible. The N20m deflection was reduced and delayed by 2 . Field mappings indicated that the sources of N20m and P60m were in the somatosensory hand area. E E G recordings are routinely made in thousands of laboratories of both clinical and basic neuro. but the number of M E G installations is less than 50. the responses and especially their electric counterparts can be utilized safely in clinical testing. 1992). Philadelphia. Makela et ah> 1992). Vol.

R." Vol. W. T. A. M. Buchanan. Buser. Moran. "A consensus statement on relative merits of EEG and MEG. M. Kajola. M. 2G Enterprises. Costa Ribeiro. "Effect of inhomogeneities on the apparent location and magnitude of a cardiac current dipole source. G. Phys. S. J. and M. I. Ben-Jacob." Science 223. Williamson." J. Rosen. Beatty. "Neuromagnetic localization of epileptiform spike activity in the human brain." Science 190. Paul (Wiley. R. M. J. and M." Electroencephalogr. A.. and Y. Lett.." Science 199.. Bertrand. M. II. 705-708. edited by A. F.. Varpula. Pernier. C. and R. Chuo-ku. Neurosci. I. and M." IEEE Trans. R. A. F. Lett. Springer Proceedings in Physics. C . 1978. L. E." Brain Res. J. and P. A. V. Japan. pp. Barrett. Ahonen. 1989. Magn. Atsumi. A. G. J. Williamson. BME-17. Cahill. Neurophysiol. Williamson. Knuutila. R. D. P.. 64.. P. CA 94043. C. and V. Vol. and E. and S. O. 1992. New York). W. New York). Maugui'ere (Elsevier. pp. Weisskoff. L.. J. 1992. Knuutila. I. Kaufman. D. G. Takenaka Corporation. and D." Phys. Hamalainen. Erne. D. T. "Designing planar gradiometer arrays: Preliminary considerations. M. R. Biomagnetism 987 (Tokyo Denki University. T. Hari. Tesche. F. Hommachi. 716-719. M. and J. D.. Vilkman. T. Lounasmaa. "Multichannel SQUID systems for brain research. Kotani (Plenum. 480-482. Brenner. April 1993 487 Barkley. R. O. Lounasmaa. (in press). 1967. 1975. 1991. "Single-unit activity in the auditory cortex of monkeys selectively attending left vs. E. J.. Lynn. 6806-6815. Froment (IEEE Engineering and Medicine and Biology Society.. edited by A. Hoke. J. Phys. C. O. Vacuumschmelze G m b H . S. Vol. J. 595-621. S. Lehmann. 1991. W. 1-6-5 Higashi-Nihonbashi. S. DC). 2. D. M. D. Vevea. D." in Soc. S. Y. G. Cabrera. S. I. 1987. Williamson. P. Knuutila. 1-13. J. G. 17. S. "Neuromagnetic evidence of spatially distributed sources underlying epileptiform spikes in the human brain. 189-199. C. Ilmoniemi. J. Biomed. "The application of electromagnetic theory to electrocardiography. Perrin.. D-6450 Hanau. Eng. Kotani (Plenum... Ahonen.. Dittmar and J. 3-18. Clin. M. Simola. 41 of New Trends and Advanced Techniques in Clinical Neurophysiology. 1992. 1987. T. McKinstry. pp. J. Liibbig (Springer. Vol. E." IEEE Trans. Amsterdam). R. Duck. J. Williamson.. Abstr. Berlin). A. Ueno. "Technical requirements for evoked potential monitoring in the intensive care unit. Halliday. Williamson. "Magnetometer position indicator for multichannel MEG. M. Seppa. 1984. S. M. Neurophysiol. 6. Phys. Lyon). 1981. R. Kajola. Jr. M. "Activation of auditory cortex with visual stimulation through sensory-sensory conditioning. Sutherling." Electroencephalogr. 317-319. 1978.. New York). "Neuromagnetometer calibration using an array of small coils. P. Wheatley. 463-491. 159. 52. REFERENCES Ahlfors. D. L. "Recent advances in Josephson junction materials. eds. Laine. 1378-1381.. S. Vilkman." in Advances in Biomagnetism. "A 122-channel magnetometer covering the whole head. Katila. Hamalainen. R. Tokyo). p. G. Lewis. Ryhanen. Drung. Stroink. Aittoniemi. "Sampling theory for neuromagnetic detector arrays (Helsinki University of Technology. Anogianakis. Ilmoniemi. Belliveau. IEEE Trans Biomed. 1970. 1989. Tokyo 103. pp. S. Osaka 541. Butler.: Magnetoencephalography Diego. Koch. 1991. CA 92121. C . Flynn. S. Barth. Finland). 63-67. J. S... Germany. 1992. . Griiner Weg 37. 891-894. Clin. Kuusela.Hamalainen et a/. Washington. K. T. pp. Hoke. edited by P. Stroink. Koch and H. 897 Independence Avenue. Medwick. 78. J. Ahlfors. and J. Rev." J." in Advances in Biomagnetism.. B. Fenici. Salinger. B.-L. Barnard. Buchbinder. J. right ear stimuli. C.." Biophys. "First results from a superconductive detector for moving magnetic monopoles." Phys. Kennedy. Arthur. P. L. Paulson. 1978. R. R. M. pp." J. 693-696. and T. 1989. and V. P. E. S. Chuo-ku. 2786-2792. and M. J. J. Albert. and B. No. Kotani (Plenum. Scheich. 82. EEG Suppl. Simola. Ilmoniemi. T." Electroencephalogr. New York). 1988. A. Kotani. R. Mauguiere. Karp. and P. J. Neurophysiol. J. 27.. Timlake. 16-20. A. J. J. T. Beatty. Geselowitz. L.. J. Eng. P. 225-236. pp. Peters. A.. S. D. Part 2 (Society for Neuroscience. K. Stroink. "Somatically evoked magnetic fields of the human brain. A. 8. Jr. Takanashi. M. Sutherling. J. Vilkman. "Thalamocortical mechanisms underlying synchronized EEG activity. 65. Numerical solution of the integral equations. Regression and the Moore-Penrose Pseudoinverse. 293-296. "Techniques for dc magnetoencephalography. B. edited by S.L.. A. 141-146. Japan. Imry. Tokyo). 1993. D. J. "Functional mapping of the human visual cortex by magnetic resonance imaging. Superconducting Sensor Laboratory of Japan. G. Mountain View. and J.. Hoke. Rev. 1982. R. Neurophysiol." in Superconducting Devices and Their Applications. (Paris) 39 (C6) 1223-1225. M. and J. Rossini. Lipton.." in Extended Abstracts of 1987 International Superconductivity Electronics Conference (ISEC 987) (Japan Society of Applied Physics." in Proceedings of the Satellite Symposium on Neuroscience and Technology. 1961. A. "Visually evoked magnetic fields of the human brain." in Advances in Biomagnetism. M. 1991. Hamalainen. 81-83. Hienz. and N. Anderson. "Estimates of visually evoked cortical currents. Fenwick. P. J. 1982. 1991. 1972. Brenner. G. A. edited by S. "A neuromagnetic study of selective auditory attention. 443_446. Simola. "On balancing superconducting gradiometric magnetometers. Appl. Bruno. "Thermal conductivity of liquid He 3 . edited by S.. Hamalainen. "Toward a technology of electronic circuits with high-r c supeconductors. J. Vol. Cantor. Tepley. J. I. H. and D. Braginski. M. J." in A Textbook of Clinical Neurophysiology. L. M. 4-chome. Badier. Clin. 48. C. 307-320. Cohen. Arthur." Science 254. and H. "Dynamics of the dcSQUID. Barth.. and V. Ahonen. 2 G / H y p r e s . Kajola.. V. Tesche. edited by H. M. D. Brady. 14th Annual Conference of the IEEE Engineering in Medicine and Biology Society. and R. Mod. 649-652. pp. and L. I." Science 218. N.. T. Benson. Rossini and F. N. E. Bohorquez.-M. Braginski. Engel. and W. Chichester). Clin. S.. 82. 94 of Mathematics in Science and Engineering (Academic. M. 659. 51-70. Kaufman. T. D. Rev. and H. Grandori. 7.. and R. M-L. S. D. Katila. Ilmoniemi. Engel. 1990. 348-360. Romani. A. and S. J.

. Romani. " A superconducting galvanometer employing Josephson tunneling. Cuffin. 27. Carelli. Japan (R&D Association for Future Electron Devices. Neurophysiol. pp. Hoenig. 2." I E E E Trans. No. Kress.. D. H. Ketchen. J." Electroencephalogr. J. 1985. 38-51. and B. I. 1981. 1970b.. 1992b." J. 13.8 1 . Neurol. K. "Minimum norm estimation of current distributions in realistic geometries. R. 1975. 211-222. H. Phys.1 4 4 . 1983. 999-1012. Cuffin. edited by Rev. Klapwijk." Rev. pp." I E E E Trans. R. pp. and V. Erne. " A second derivative gradiometer integrated with a dc superconducting interferometer. and M. 1984. Pizzella. Phys. edited by H .. Purcell. 41. F . Dossel. de Weerd. 1975. B. Clarke. 46. 1989." Appl. Stroink. Ioannides. C. Daalmans. Eng. Foglietti. Barbanera. pp. D." in Advances in Biomagnetism. Leoni. Clin. Magn. Petah Tikva. P. pp. 2797-2800. Clin. Pullano. M. Bar. Fuchs." I E E E Trans. 1987. and M. J. "Investigation of a double-loop dc S Q U I D magnetometer with additional positive feedback. 27." Electroencephalogr.. Pat. E. 1979. 1986. 5 3 . Cohen. J. 1966. 464-470.6 6 8 . Liideke. J. Schomer. Yoshizumi. and M. " A nine channel D C SQUID system for biomagnetism." in Advances in Biomagnetism. O. C . "Localized and distributed source solutions for the biomagnetic inverse problem I. "The potential distribution in a layered anisotropic spheroidal volume conductor. M. 1991. and J. SQUIDs. Hokkaido. Yang. and J. Fuchs. Maung.. Cohen. 39.488 Hamalainen et a/. Appl. 2927-2931. New York). Phys.." J. " A multi-channel S Q U I D system for current density imaging. 2573-2577. J. L. 6 6 5 . M. and G. D. pp. 5. Pasquarelli. Vol. 1968.789. J. Foglietti. Kotani (Plenum. W. 351-356. Seifert. 1989. N. J.. G." I E E E Trans. Ives. B. P. 1982. "Performance of an electronic gradiometer in noisy environments. 664-666. Kotani (Plenum. Hoke. 28. 1989.. BME-32. Daalmans. Jr. Zhang. S. 219-222. Kriiger. C . C. J. Today 23. "The estimation of time-varying dipoles on the basis of evoked potentials. G. P. A. 1977. Stroink.: Magnetoencephaiography 1991. M. 1991. and G. and D . C.5 8 . "Digital feedback loops for D. Magn. Cybern. edited by M.-Y. H. C . and G. 8 3 7 . Yunokuchi. Clin. 64. 1988. " A modular low noise 7-channel SQUIDmagnetometer.5 7 . and L Khalil. Del Gratta. 645-650. "Apparatus and method for detecting cancer in tissue. Drung.. Cohen. "Probabilistic methods in a biomagnetic inverse problem. Eng. 1989. R. Lett. Clarke. L. Shikabe. Cohen. L. pp. "Integrated thin film S Q U I D instruments.. Cuffin. 77. and R. Berlin). P. Greenblatt. Phys. K. R. Biomed.. 6 6 9 . J. . 3. 1990. and R. 27.8 4 1 . M. Neurophysiol. J. Drung. 623-627. Mag. O. C. D. Eng. 6 9 . 3 7 2 .. B. Bolton. "Demonstration of useful differences between magnetoencephalogram and electroencephalogram. Clarke. 9 9 .. D. "Low-field room built at high-field magnet lab. H. S." in Advances in Biomagnetism. 58. 59. M AG-19. 784-786. de Waal. Carelli. S. and M." Appl. Damadian." I E E E Trans. and performance. Kullmann.. V. F.. R. "Magnetic fields of a dipole in special volume conductor shapes. Phys. Cohen. V. Stroink. and K. Claassen. M. M. "Single neuron activity in the right and left human temporal lobe during listening and speaking. "Tunnel junction dc SQUID fabrication. Clarke. "Localization of biological sources with arrays of superconducting gradiometers. New York).." J. "Ultra low noise all niobium dc-SQUIDs." in Biomagnetism: Clinical Aspects. Crowley. Phys.. David. 1989. "Magnetoencephaiography: Detection of the brain's electrical activity with a superconducting magnetometer.. P. 27. 155-156. Kullmann. 56. Magn.. J.. David. Amsterdam). 1976. Hoke. and H. Hoke. M. Goubau. Romani (Elsevier. K o c h and H . Dossel. Cohen. and T. Liibbig (Walter de Gruyter. 2268-2275. D. D. K. " S Q U I D magnetometer noise measurements of Y B a 2 C u 3 0 7 thin films and input coils. Appl. 65. 1992." Inverse Problems 5. Bommel. "Geophysical applications of SQUIDs. R. "Magnetoencephaiography: Evidence of magnetic fields produced by alpha-rhythm currents. 1992a. Williamson. J. N . Introduction and conceptual basis. Appl. Pegrum. Okada." in SQUID '91: Superconducting Devices and Their Applications. 15-16.. 2997-3000. Neurophysiol. Hoke. G. D. M. 27. V. 905-910. D. G. Phys. Clarke. Berlin)." Philos. W. C. J. Williamson. and D . N. Cosgrove." Electroencephalogr." Science 175.-M.." in SQUID '91: Superconducting Devices and Their Applications. and M. B. N ." in Proceedings of the 2nd Workshop on High-Temperature Superconducting Electron Devices. "A Posteriori time-varying filtering of averaged evoked potentials I. Donaldson. Goubau. "Coupling considerations for SQUID devices. " M E G versus E E G localization test using implanted sources in the human brain." Phys. BME-24.. C. Shikabe). G. "Magnetic measurement and display of current generators in the brain. J. 1983. Modena. C. and M ." in SQUID'85: Superconducting Quantum Interference Devices and their Applications. Phys. Torrioli. Biomed. 587-590. " A comparison of moving dipole inverse solutions using E E G ' s and M E G ' s . Mod.. Cohen." U. Engel." Ann. L. 1985. A. Dilorio. 54.. Springer Proceedings in Physics. N . New York). "Thin-film dc SQUID with low noise and drift. Cuffin. 1990. operation. 1986. J. de Munck. Appl. Maniewski." Biol.. edited by S. Y. Kennedy. 64. de Munck.3 8 1 . M . Biomed." Science 161. 1983. Liideke. Cohen. Clarke. R. W. edited by S. R. Williamson. " A linear discretization of the volume conductor boundary integral equation using analytically integrated elements. J.. 603-606. S. Creutzfeldt. Vol. Hahlbohm and H . 569-572. G. Part I: The 2-D detector.." in Digest of the 12th International Conference on Medical and Biological Engineering.." J. Ilmoniemi. B. " I E E E Trans. 156-160." Cryogenics 26. Ketchen. D ." I E E E Trans. Springer Proceedings in Physics.. J. 115-127. April 1993 J. C . 1972. edited by S. W. M. Chapman. pp. G. Jerusalem (Beilinson Medical Center. de Munck. "Dc-SQUIDs for a 31-channel SQUID-gradiometer array for biomagnetic diagnosis. A. Liibbig (Springer. Low Temp. 1970a. pp. Uhl..6 7 1 . Bommel. Magn. J. 986-990. Phys. 811-817.. 1972. D.C. Kotani (Plenum. and D ." I E E E Trans. 1991. "Large-volume conventional magnetic shields. New York). P.-M. Lett. Appl. R. "Design and optimization of dc SQUIDs fabricated using a simplified four-level process. "Selective localization of alpha brain activity with neuromagnetic measurements. "Compact integrated dc SQUID gradiometer.S. J. edited by H . 25.. Bain. S. (Raven." in Fundamental Mechanisms of Human Brain Function. Israel). 5 6 . 288-294. Romani. O. E. B. M. B. 41. Magn. S. L. G. Drung. W. 729-759." J. 1992. Carelli. B. G.832. M. 6065-6067.

" in Biomagnetism '87. "Experimental gradiometer with a digital feedback loop. Brittenham. 1990. "Low noise YBa 2 Cu 3 0 7 _ 8 grain boundary junction dc SQUIDs. 319-322. Tracht. T. "On the magnetic field generated outside an inhomogeneous volume conductor by internal current sources. 1992. Herwig. G. Temp. pp. Huggins. 9. Auditory Evoked Magnetic Fields and Potentials.. Lett. Warnick. Berlin). R. Gallagher. H. Drung. J. Shirae. S. Stroink. C. GifFard.. M. M. 1983. A. Fenici. 7. "The positioning problem in biomagnetic measurements: A solution for arrays of superconducting sensors.. L. E. W. 1990. pp. D. W. 542-546. Kiviranta. Instrum. 2nd ed. Sandstrom. R." Biophys. W. Williamson (Tokyo Denki University. 27. Suni.. edited by H. pp. M AG-6. 25. Pizzella. Flynn. Sandstrom. Magn. van Loan. and G. "Magnetic activity of a single peripheral nerve in man. Sci. Foglietti. H. and M." Appl.. Koch and H. Pasquarelli. "Digital-analog magnetometer utilizing superconducting sensor. Gross. Foglietti. D. B. 38. N. S." Rev. "Low-noise high-speed dc superconducting quantum interference device magnetometer with simplified feedback electronics. Kleinsasser." Scand. J." J. KleinRev. Tripp. Karp. 1989. H. and H. Wheatley. Buxton. Tokyo).. "High quality refractory Josephson tunnel junctions utilizing thin aluminum layers. and A. 1987. 1762-1768. 3141-3147. Duret. 1987. Gurvitch. Ketchen. Freeman.. M. L.. 2565-2568. "Model studies of the magnetocardiogram. R. P. "Magnetic auditory responses from the human brain: A preliminary report. S. and A. Fujimaki." in Advances in Biomagnetism. Herwig. 1965.1-18. Kawasaki. edited by H. "A multi-channel biomagnetometer. 2. Crocoll. Berlin). R. 27. C. pp. Torrioli. 56. 281-285. M. B.. A." in SQUID '85 Superconducting Quantum Interference Devices and their Applications. and S.. and W. Res. Erne. and Z. Magn. J.. H. H. 1967. J. 1980. Erne. E. B. Katayama." J. E. The Feynman Lectures on Physics. Koch. "Characteristics of YBa 2 Cu 3 0 7 _ 8 grain boundary junction dc-SQUIDs. Kofoed. edited by (Japan Society of Applied Physics.. 65. 57. Scheer. Gronberg. "Potential. Lett. Erne. 3 (Addison-Wesley. and J. pp. Elberling. D. Gallagher. Seppa. 37-60. H. 1981. P. Lett. M. Feynman. Drung. M. P. "28-channel hybrid system for neuromagnetic measurements. B. "Highresolution magnetic measurements of human cardiac electrophysiological events. "On bioelectric potentials in an inhomogeneous volume conductor. S. edited by S." IEEE Trans. Phys. Gross. Berlin). Grandori. Ranken. Jackson. W. M. Reading. 406-408. J. 445-471. (The Johns Hopkins University.. Springer Proceedings in Physics. and A. M. 1986. L. Erne. 1972. "Instrumentation for biomagnetism. L. Phys. N. "Performance of dc SQUIDs with resistively shunted inductance. 2959-2962. pp. "A single-chip SQUID magnetometer. D. edited by H. Karp. "Performances of higher order planar gradiometers for biomagnetic source localization. B. and R. Ketchen. Third International Workshop on Biomagnetism." in Advances in . H.-D. 1988. Hasuo.. M. 1983." J.. Hahlbohm and H. Leighton.. and H. R. Sands. Aust. Knuttel. P. Romani. V. P. edited by K. N. Tokyo). M. A. Grummich. 123-141." IEEE Trans. "Principles and methods of low-frequency electric and magnetic measurements using an rf-biased point-contact superconducting device. Appl. F. No. A.. and P." Appl." IEEE Trans. B. E. M. 80. Salmi. Goldman. A. Trontelj. Del Gratta.-D. Guy. "The Berlin magnetically shielded room (BMSR): Section B—performances. N. Saermark. L. D. Foglietti. S. R. D." Appl. Webb. R. J. MAG-23. Hahlbohm. pp. Liibbig. H. N. Chaudhari. Vol. J. S. Phys. Erne. A. Jutzi. D. "Superconducting magnetometer measurements of geomagnetic activity in the 0. H. Geophys... Liibbig (Springer. Z. 346-347. Lett. Gupta. J. 1975. R.-D. Katila. 21 -24. 6 of Advances in Audiology (Karger. Farrell. W. 1975. N. Berlin). S. Danish." in Biomagnetism. Audiol.. Hahlbohm and H. L. Kotani.. Liibbig (Walter de Gruyter. and S. P." in SQUID '80: Superconducting Quantum Interference Devices and their Applications. and W. Leedham-Green. and P. Neuhaus. "Measured performance parameters of gradiometers with digital output. Fraser-Smith. 1983. G. MA). Raider." Nuovo Cimento D 2. Ketchen. 1988. Physiol. Vol. P. Cantor. F. W. 1991. and E. and R. H. J. M. and K. 1967.." Biophys. 79-87. and K. 166-169. 1451-1453. Berlin). Sandstrom. G. Liibbig (Walter de Gruyter. Magn. L. 1393-1395. 6. Proceedings. and M. Gallagher. 1986. "Low-frequency noise in low \/f noise dc SQUID's. A. Pizzella. Cantor. 1989. 1034-1037. George. 472-474. Hoke. Basel).. edited by S. Peters. I. 1992. 951-961. Neuhaus. impedance. N.. 582-593." Appl. Crocoll. R. Phys. R. 64. 1984. and H. Gen. 42. Erne. Curio. "A low noise dc SQUID based on Nb/Al-A10 x /Nb Josephson junctions. and I. New York).. J. M. Kotani (Plenum." IEEE Trans. "Three-dimensional volumetric reconstruction for neuromagnetic source localization. Gupta." J. E.. GifFard. J. and D." Appl. Tamura.. 1991. M. Kawasaki. H. pp. Eds.. Narici. 1989. 1990.. Tsukuba). Golub. M. W. Grynszpan. unpublished. F. L. Vol. Washington.: Magnetoencephalography edited by H. pp. Magn. D. Berlin). "Figure of merit and spatial resolution of superconducting flux transformers. 737-740. "Vector measurement of the heart magnetic field.Hamalainen et a/. New York). 21. D. Duret. 1985. R." IEEE Trans. Biomagnetism. T. S. April 1993 489 sasser. 533-610.. Cayles. "Fundamentals for SQUID applications. D. pp. Phys. R. Romani.. 1319-1322. Koch and H. and P. M. W." in Proceedings of the 6th Sensor Symposium. 1970. T. Jutzi. Erne. C. G. 185-190. L. H. 727-729. A. 1989. Berlin (Walter de Gruyter. Emoto. M. 911-925. Mod. L. Electron Devices 35. R.. Imamura. S. 1943. Geselowitz. B. Scheer. edited by S.214-220. 49. L. M.. 1981.." Nuovo Cimento D 2. J. and G. and E. 1983. and H. Trahms. and A. Liibbig (Springer. J. Bak. C. Koch. H. Mass Action of the Nervous System (Academic. G. B. Herrmann. Matrix Computations. Phys. and S. B. Walker.. 27. 1989. Vol.-D." in SQUID '91: Superconducting Devices and Their Applications. and R. 1-11. 55. A. Phys.1 to 14 Hz frequency range. J. Hoke. B. L. R. V. Liibbig (Walter de Gruyter. Chaudhari. Romani. V.. Hahlbohm. Romani. D. and K. Lebech. N. Ueno. Geselowitz. R. V. Ketchen. 1973. S. Ketchen. G. Romani. Low. Phys." IEEE Trans. Kataoka (Institute of Electrical Engineers in Japan. 57. E. 1980. 13. 2412-2418. Eds. Hahlbohm. and G. Furukawa. M. C . 231-247. Kleinsasser. N. Ryhanen. Forgacs. L. G. V. Geselowitz.. A. Lett. Trontelj. Koch. "A clinical system for accurate assessment of tissue iron concentration. and rectification in membranes. Baltimore). Magn. Atsumi. H. and C. R." in Extended Abstracts of 1987 International Superconductivity Electronics Conference (ISEC '87). Drung. Williamson. and J. R. Harris.. C ." Nuovo Cimento D 2.

edited by J. Neurophysiol. Teszner. T.. Hamalainen. Hamalainen. and D . R. von. Seifert. Hamalainen. 1989. 2.. Brain Res. M. and R. V. 64-72. J. and M. 1984. Koyanagi. D . 1988. Ishibashi. A 12. and T. Tiihonen. Phys. Bommel.." Ear and Hearing 9.. Katila (Helsinki University of Technology. Huopaniemi. Hamalainen. J. 1991. Lounasmaa. J. "Utilization of reconstruction algorithms in transmission and emission computed tomography. 45. G. H. M. Vol. 561-569.1 2 6 . Clin." I E E E Trans. "Interferences about sources of neuromagnetic fields using Bayesian parameter estimation" (Helsinki University of Technology. Baltimore. T. Clin. Hinshaw. Pelizzone. Neurol. Pelizzone. Hari. Eds. Ilmoniemi. "Separate finger representations at the human second somatosensory cortex. E. and V. and T. Technical Report T K K . R. " O n brain's magnetic responses to sensory stimuli. and J. Leinonen. R. and M. Romani.801. M. "The neuromagnetic method in the study of the human auditory cortex. and J. H. V. edited by E. Hari. B.." in Electroencephalography. 1989. "Realistic conductivity geometry model of the human head for interpretation of neuromagnetic data. H.. Joutsiniemi. P. M . Hamalainen. 277-280. 1993. Ilmoniemi. H .-L. Makela. 27. R. J. Reinikainen. M.A 5 5 9 . 1993. Tiihonen. 1992. Grandori. Ishikawa. and J. Weinheim).. Hari. Helmholtz. Magn. 74. and O. Hoke.. "Functional organization of the human first and second somatosensory cortices: A nueromagnetic study. K. Lounasmaa. L." Appl. Phys. S. Knuutila. 1988. 1993. E. R. Kotani (Plenum. M. M.and offsets of noise bursts. 1987. Hari. 1987. M. and S. Hallstrom. Uhl. S.. 1983. R. M. Hari." Neurosci. R. Y.. S. Niedermeyer and F. J. Hamalainen. Ilmoniemi. L.. Higuchi. J. Rev. 661-664. 82. 1983. pp.4 3 . N . April 1993 Hari. Eng. 1991. 287-335." in Biomagnetic Localization and 3D Modelling. W. Hoke. Hari. Hoke. 245-249. G. Joutsiniemi. A. P. H.. "Computation of the return current in encephalography: the auto solid angle. J. and Related Fields. R. Tuomisto. Mod. 432-436. H. 237-240. and J. Kekoni. J.. " A n introduction to N M R imaging: F r o m the Bloch equation to the imaging equation. 42." Proc." Ann.. 1989. "Magnetoencephalography as a tool of clinical neurophysiology. Vol.. "Anatomical correlates for magnetoencephalography: integration with magnetic resonance images. J. Sarvas. Kaila. H. Hari. S. edited by F . Hari.. Kasai.. 8. 1853. 257-299. and A. pp. Paulus. Phys. Sams." Audiology 26. in press). 222-282. " M E G versus E E G localization test (Letter to the Editor). and K. and T. M. Aittoniemi. 148-152. Neurophysiol. I E E E 71. pp. mit Anwendung auf die thierisch-elektrischen Versuche. O.490 Hamalainen et a/. Erne. Varpula. 1984.-L. 36.. Phys. 65. and Y." in Auditory Evoked Magnetic Fields and Electric Potentials." Clin.: Magnetoencephalography Biomagnetism. Schneider. F . Hari. R. C. Ilmoniemi. edited by S. Abraham-Fuchs. 463-470. Kaukoranta. "SQUID-based measuring techniques—a challenge for the functional diagnostics in medicine. H. Idell." Electroencephalogr. Katila. and M. 701-704." Phys. 54. Clarke. Ilmoniemi. K. Mauno. SPIE 1351. 799 . J. Finland)." Electroencephalogr. "Somatosensory evoked cerebral magnetic fields from SI and SII in man.F . J. J. R. L. Lehtinen. New York). Stroink." Neurosci. Knuutila. 1989. T.. J. Biomed. J. Salonen. J. Clin. Tiihonen. R.. Williamson. Technical Report TKK-F-A620. and K. Basel). Feldman. M. "Input impedance of an amplifier based on a dc superconducting quantum interference device. and J. S. 1982. M. 222-224. Neurophysiol. 1990. "Multichannel detection of magnetic compound action fields of median and ulnar nerves." in Digital Image Synthesis and Inverse Optics. Reichenberger. Hari. and J. "Neuromagnetic mismatch fields to single and paired tones. M. Neurophysiol. Clin. S. J. 2777-2785. K a r h u . K. "Cerebral magnetic fields. "Interstimulus-interval dependence of the auditory vertex response and its magnetic counterpart: Implications for their neural generation.2 3 3 . Hari. M.8 2 . Clinical Applications. and R. Hari. M. Makela. Sarvas. M. Biomed. Chem." Electroencephalogr. J. Hamalainen. Hallstrom. Kaukoranta." Electroencephalogr. Heller." Ann. Biomagnetism: Clinical Aspects (Elsevier. Hari. Pantev. 40.. R.-L. Hari.. Hamalainen. Gmitro. T. Proc. R. "Spatial resolution of neuromagnetic records: theoretical calculations in a spherical model.3 2 . 165-171. Chinone. pp. edited by A. T. Okada." in Advances in Biomagnetism. Rajala. "The positioning of magnetometer pickup coil in dewar by artificial signal source. Brain Res. G. 353-377. Neurosci. Hilbert. H. Physiol. and O. . Hari. Makela." Exp. R. 37. Meas. L." in The Art of Measurement: Metrology in Fundamental and Applied Physics. and G. and R. Romani (Karger. pp. 1992. Espoo) pp. "Interpreting measured magnetic fields of the brain: Estimates of current distributions" (Helsinki University of Technology. 57. M. Finland).. Daalmans. edited by S. Hoenig." Eur. J. "Ueber einige Gesetze der Vertheilung elektrischer Strome in korperlichen Leitern. and J. Liitkenhoner. 7 7 . Penttinen. 254-263.. New York). E. Vilkman. "Auditory evoked transient and sustained magnetic fields of the human brain: Localization of neural generators. 376-390. Weisse. 1988. 338-350. Lent.. Lounasmaa. Hari. 72. Sams M. 71. J. 1986. S... 3 1 . Sams. E. M. Clin. Amsterdam). pp. Biol. Reinikainen. Hamalainen. K a d o . Suppl. Kotani (Plenum. R. 1984.." in Imaging Techniques in Biology and Medicine (Academic. R. K. R. Rif. and O. and G. R. Basic Principles. S. Stroink. K. Nenonen. (in press). N . " A neurophysiologist's view on biomagnetic source localization. K r a m e r (VCH Verlagsgesellschaft.. Hoke. 9 3 . Nakanishi. No. V. Lett. H. 1988. Huttunen.-M. M . J. 1990. C . 157-169. "Neuromagnetic localization of cortical activity evoked by painful dental stimulation in man. 1991b. M." C R C Crit. R. Williamson.. Joutsiniemi. LaHaie (SPIE International Society for Optical Engineering. Hoke. R. C. N . "Neuromagnetic responses of the human auditory cortex to on. M. "Selective listening modifies activity of the human auditory cortex. "Interpreting magnetic fields of the brain: Minimum-norm estimates. 1991a." Science 244. and M. 2 9 ." J. 3 2 . Med. Jarvinen. edited by B. Hallstrom. S. Hoke. Rev.4 3 . L." Electroencephalogr." I E E E Trans. S. 1990. Hari. Hari. 1991. Lett. and I. S. 2 1 1 .. 1989. Lopes da Silva (Williams and Wilkins. New York). 152-154. P. M. Varpula. Eng. 14... Kaukoranta. A. M . "Neuromagnetic responses from a deaf subject to stimuli presented through a multichannel cochlear prosthesis. Neurophysiol. F. H. Tiihonen. Bar. J.. M. M . J. Comp. 6 of Advances in Audiology. 89. Clin. Katila. 1980. R. Hamalainen. Haario. Bellingham).-L. "Multi channel dc SQUID sensor array for biomagnetic applications. "Recording and interpretation of cerebral magnetic fields. Salminen." Exp. Neurophysiol. Hinz. (in press). Hamalainen. 30.

1991. Surrey). edited by C. 1991. and J. S. R. 1986. Neurosci. Backonja. Josephson. Ilmoniemi. edited by K. R.. 967-979. R. T. S. 4 4 0 . J. Ilmoniemi. 4 5 9 . Clin. edited by D . "Localized and distributed source solutions for the biomagnetic inverse problem I I . Skills 37. 523-542.. Ilmoniemi. and W. 80. O. 281-286. Abstr. New York). G. Suppl. R. S. " N e w method for the study of spontaneous brain activity. 2 7 1 . Schwartz. 83. M. Iversen. "Digital model for studies in magnetocardiography. M. Cleeland. Rogalla. R. Tokyo). "Continuous probabilistic solutions to the biomagnetic inverse problem. Neurophysiol. M . "The forward and inverse problems in the spherical model. Williamson. Knuutila. A.. Williamson. Liibbig (Springer. Penttinen. G. Hoke. O. 3. " A 24channel magnetometer for brain research. R.. M . 1988. pp. Brain Res. J. C." Exp.. Newman.. E. Makela. Bolton. and C.. 27. "General principles. D . Thurlow. Clin. J. 1985. "Sub-micron Y B a 2 C u 3 0 7 _ 6 grain boundary junction dc S Q U I D ' s . Ahlfors. Boston). Washington." in Biomagnetism: Applications & Theory. Ilmoniemi. Jack. Liitkenhoner. Mercereau. and W. and S. and S. J. Lett. M . 241. Newman. J. Kaukoranta. and T. Peltonen. J. Jaszczuk. R. Part 1 (Society for Neuroscience. 1079-1094.9 . B. and J. Vol.. J. C. 192-200. Brain Res. A. R.. Katila (Pergamon." Sci. Martinez." Electroencephalogr. M. Hoke. Hasuo. Hasson. M." I E E E Trans.3 2 1 . p. and T. Clarke.-T. L. 1986. 12 of Progress in Low Temperature Physics. and S. and M . Rif. S. DC). 1992. V. J. " O n the current multipole presentation of the primary current distributions.Hamalainen et a/. pp. W. Kaufman. B. ter Brake. J. Hari. M. Tepley. Gupta. Foster. S. 16. Katila. H. J. C. and H. Phys.. 347-349. "Late pain-related magnetic fields and electric potentials evoked by intracutaneous electric finger stimulation. J. 660-664. R. Houwman. E. Springer Proceedings in Physics. J. P. Jacklevic. P. C. Hari. 64. S. Knuutila. " D c . A. "Responses of the human auditory cortex to vowel onset after fricative consonants. M . P. F. Knuutila. C. J. and H. 591-594. R. 2. Wakai. Magn. and A. J.-L. Ilmoniemi. and M. J. P.. R. Res. R. J. and J." Hearing Res. Wharton. "Cerebral magnetic fields to lingual stimulation. Ueno. 64. Ehnholm. K a r h u . J.. Neurosci. E. Ioannides. E. 124-132. New York)." J." Eur..2 3 . and M. L. pp. O... edited by H . and J. Lett. 1979. 4 5 7 . 1962. P.. Stroink. 63. M. J." Phys. and M. pp." in Biomagnetism '87. Rev. 568.. Salustri." Electroencephalogr. Kelha. W. R. New York). B. L. and W. 4 6 . Kotani (Plenum." Percept. 4 4 . Hari. and C. "Cortical reactivity in patients with progressive myoclonic epilepsy. M . Magn. edited by M. Veldhuis. " in Superconducting Devices and Their Applications. "Visually evoked magnetic fields: Retinotopy and cortical magnification factor. Schwartz. April 1993 491 netic field ( m 1 0 o / m 20o) measurements in tinnitus and normal groups. Neurophysiol. D . Clin. Joseph. B. J. Krusius (Butterworth. E. pp. Jaycox. No. T. Pukki. Am. E. E." Neurology 42. Collan. Vol.. Principles of Neural Science. J. 3rd ed.. Barber and T. Berlin). 1989. M . K a r h u ... Hari. Kaukoranta. Kotani (Plenum.. Berglund. E. Knuutila. "Mixed and sensory nerve stimulations activate different cytoarchitechtonic areas in the human primary somatosensory cortex SI. Katila. Hamalainen. 152-170. 1981. 2 5 1 . 1990. 1989. Mot. J. Hoke. and K. Foster and J.5 2 . J. A. Stroink. 1987. Huttunen. 1983. Hamalainen. M . and R. a magnetoencephalographic study in man. C . and S. 159-160. 1988. H. edited by S. Williamson. K. 1992. pp. " A submicrometer N b / A 1 0 x / N b Josephson junction. Kobal. S. Hari. "Magnetic fields from the auditory cortex of a deaf human individual occurring spontaneously or evoked by stimulation through a cochlear prosthesis. " I E E E Trans.. Hamalainen. Hari. A. Hallstrom. Phys. "Recent developments in neuromagnetism. Vilkman. Neurophysiol. and R. Kajola. Sarvas. 12. J. Jaszczak. and J. T. M. R. 1-48. Huttunen. Pantev. Neurophysiol. 1 9 . Kirviranta. S. 2. (Elsevier. "Objective evidence of tinnitus in auditory evoked magnetic fields. G. "Cortical responses to painful C 0 2 stimulation of nasal mucosa. 1989. Tesche. A. J. Lehnertz. Jacobson. MAG-17. V. pp. M . 1990. E. A. Amsterdam). K a r h u . Oxford). Ilmoniemi." Phys. I E E E 76. J. 64. "Somatosensory evoked magnetic fields in multiple sclerosis... 150-154.3 3 9 . 1. Magn. S. J. 80. Hutchinson. 353. and J.. and R. G. K o c h and H . "Quantum interference effects in Josephson tunneling." in Soc. Kaukoranta.. Lounasmaa. 100-113." Audiology 28. Hari. Amsterdam). A. "Planar coupling scheme for ultra low noise dc S Q U I D s . J. and M. J. T. "Tomographic radiopharmaceutical imaging. 6. 278-282. 1991. Lounasmaa. B. Howland. Blum (Butterworth. Guildford. Huttunen. 37. Clin. J. M.. Silver." in Proceedings of the Tenth International Cryogenic Engineering Conference. Moran. 524-528." Electroencephalogr. 1989. 1989. 1990. Lambe. R. Clarke. "Modulation of spontaneous brain activity during mental imagery. W. 673-676. M.4 6 0 . Neurosci. "Multi-SQUID devices and their applications. M. Kajola. 3. R. "The chemistry of the brain. E. and J. 1992. M o d .5 2 . M. 56.2 5 3 . M A G . 4 6 7 . M. Lu. 2955-2958. Joutsiniemi. 1991." Hearing Res. J. R. Ketchen. J. N . 1973.. J. Kaufman. 1986. Reinikainen.. 1586-1588. 118-129. 3 0 9 .. Williamson." Nuovo Cimento D 2. H. F . J. Hari. and O.6 6 . M . R. Atsumi. 182-185. 1964. Appl." in Advances in Biomagnetism. 1987. Neurophysiol. " A fourchannel S Q U I D magnetometer for brain research. G. Kawasaki. Horacek.. edited by S. Potenti. "Auditory evoked cortical mag- Rev." Somatosens. Paetau. E. Bolton. Williamson. Kaukoranta. Siirth. edited by H." I E E E Trans.. A h m a d ." Proc. J. Hari. "Omissions of auditory stimuli may activate frontal cortex. E. " in Advances in Biomagnetism. Hamalainen. A. R." Inverse Probl. S. Williamson (Tokyo Denki University. 1988. and S. J. pp. T. 1. S." Vol. Hutchinson (IRL. Imamura. Clin." Electroencephalogr." Practical NMR Imaging. S. R. D. 1989." Exp. Baffa. 69. P. K. 1984.S Q U I D sensor system for multichannel neuromagnetometry. J.. J." in Evoked Potentials III: The Third International Evoked Potentials Symposium. and V. 1991. Ilmoniemi.. S. L. Ioannides. Ahlfors.4 6 8 . 6 0 .. Vol. R. Mervaala. Stroink. . "Possible new effects in superconductive tunneling.: Magnetoencephalography Lehnertz. Flokstra. 1987. Paetau. Ryhanen.. Kandel. 400-403. 65. 1973. "Cerebral magnetic fields evoked by peroneal nerve stimulation. V. pp. and C. Kotani. R. Seppa. J. G. Hostetler. B. "Effects of degree of visual association and angle of displacement on the 'ventriloquism' effect. 1984.4 7 3 . Brewer (North-Holland." J. Knuutila. R. Chaudhari. "Remarks on the design of a four-channel S Q U I D magnetometer for brain research. and R.4 4 4 . Weinberg. R. edited by H ." Electroencephalogr. Cogn. Jessel. 58.

S. S. 58. Hari. Cunningham. Instrum. construction. Scheer. and V.492 Hamalainen et a/. C. 64. 1992. D . New York). J. J. Ahonen. No." J. 1989. "Design. J. V. and J. 4 2 5 . R.2916-2919. April 1993 Stroink." I E E E Trans. J. Tesche. O. Zeki. pp. Lett. and M. 58. J. 1989. and D . J. H." J. Kohonen. 287-290. 22. M . Cope. J." Invest. Parkkonen. J. Berlin). R. K. 64 (Springer. Ritala. Williamson (Tokyo Denki University. " S Q U I D sensors. 1981. P. J.. and M. 300-312. I E E E 71. and R. Ketchen. P. R. edited by H. edited by S. 572. J. A. 1650-1657. 27. 79. 27. J. Knuutila. M. 182-196. G. M. 1988. J. Tiihonen. 1992. J. Hesse. Atsumi. Deiber.. Vol. "Low noise dc SQUIDs fabricated in N b . 1984a. J. Williamson. Koch. and M. Phys. edited by S. G. Appl." Springer Proceedings in Physics. Tesche. and A. Weinheim)." I E E E Trans. 1985. J. Koch. Phys. B." in Advances in Biomagnetism.. Joutsiniemi. "Largearea low-noise seven-channel dc S Q U I D magnetometer for brain research. J.. Rev. Drung. H. 1993. C . 256-264. L. R. Biomed. Magn. J. and A.. Ilmoniemi.. Phys. " Proc. I E E E 78. J. "Evoked responses of human auditory cortex may be enhanced by preceding stimuli.. 1978.. and W. 1991. 1989. M. C. M. Knuutila. 43. W. Kajola. Lounasmaa. Knuutila. New York). "Integrated thin-film dc SQUID sensors.. and R. P. Kessler. (Sinauer Associates. S. B. I. and H.2 4 1 . Sams. Ilmoniemi. "Image formation by induced local interactions: Examples employing nuclear magnetic resonance." J.. 1989. R. Liibbig (Walter de Gruyter. Kroger. 1983. 269-284. 207-224. "Superconducting Devices and Their Applications. Hamalainen. " A n introduction to neural computing. L. A. M." I E E E Trans." Nature 242. Ketchen. "Design of improved integrated thin-film planar dc SQUID gradiometers." Rev." Neurosci. and H. M AG-18. M. S.4 2 8 . Knoll. M. van Harlingen. Ritala. 49." Electroencephalogr. 1985." in Advances in Biomagnetism. C. and S. J." in Superconducting Devices and Their Applications. and W. Loveless. J. Hahlbohm. Z. J. Knuutila. Frackowiak. Lii. 635-638. Phys. pp. "Flicker ( 1 / / ) noise in tunnel junction dc SQUIDs. Phys. optimization. H. M. and R. W. Kajola. Gallagher.: Magnetoencephalography Ilmoniemi. 3757-3764. S. F. Knuutila. S... J. Kaplan. Kurkijarvi. Ueno. S. 2 3 6 . and performance of a large-volume magnetic shield. P. Eds. Smith. H. Ahonen. Ketoja. K. V. "Positron emission tomography: Prospects for clinical utility. 1990." I E E E Trans. Springer Proceedings in Physics. "Design. Jillie.O.." in Magnetic Sensors. Leinonen. Vol. Ketoja. J." Electroencephalogr. " A 37 channel dc SQUID magnetometer system. K. V. Rev.." Brain Res. 713-716. A.. R. Berlin). Ketchen. pp. 1989.. James. 51. 1984b. Vaughan. "Selective niobium anodization process for fabricating Josephson tunnel junctions. O. M . and C. T. Hamalainen. 3 . Cantor. 1991. Ahonen. 316-325. 2793-2796. Liibbig. pp. " A 122-channel whole cortex SQUID system for measuring the brain's magnetic fields. 1464-1480... L. 2.. Kuriki. edited by H. "The colour centre in the cerebral cortex of .. B. M . I. Friston. Matsuda. London). "Seeing faces activates three separate areas outside the occipital visual vortex in man. G. Low Temp. Martin. 68.. Kraut. 369-392. Magn. Hamalainen. Katila. Lu. G.." Neural Networks 1. 1992. 1988.-L.. Knuutila. Ryhanen. C. S. Matachi. Hamalainen. J. Hahlbohm and H ." I E E E Trans. Seppa. M. BME-34.. 186-189. and J. M... Kotani. P. 1991. Radiol. Mod. A. Hamalainen. Peters. 5 of Sensors—A Comprehensive Survey.A l 2 0 3 N b trilayer technology." Phys. J. Vilkman.. Magn. K. J. A. Lounasmaa. 260-270. Knuutila. Martinis. Ilmoniemi. Kotani (Plenum. Williamson. Donaldson.. 27. Low Temp. and D . S. Ketchen. M. B. edited by W. M. Vol. M. "Effects of alternating bias current on the low-frequency noise in dc SQUIDs. Vilkman." Phys. " H u m a n auditory primary and association cortex have differing lifetimes for activation traces. Ahonen. A. E. 190-191. and R. and J." I E E E Trans. Lounasmaa. O. "Intracortical generators of the flash VEP in monkeys. S. M. R. and C.. T. 64. Neurophysiol. H. Kuffler. 1987. J. Price. R. Zemel (VCH Verlagsgesellschaft. 74. and M . Knuutila. L. 62. Arezzo. and L. 381-445. 39. and E. Clin. 529-537. T. Liibbig (Springer. Nicholls. and C. B 30. "Design considerations for dc SQUIDs fabricated in deep submicron technology.-T. G. From Neuron to Brain. Kajola. Joutsiniemi. V. Ahlfors. Kajola. S. Partain. J. Kemppainen. Mutikainen. pp. "Auditory evoked potentials and magnetic fields in patients with lesions of the auditory cortex. Salmi. R." Appl. 1983. "Generator sites of spontaneous MEG-activity during sleep. N . R." Electroencephalogr. (submitted). T. M. 1991. and S. pp. R. Matthies. M. H. Simola.-L. 1985. Appl. O. "Design considerations for multichannel SQUID magnetometers." J.. 1988. 1987. S. "The self-organizing m a p . M. 280-282. 82. C. Goubau.4111-4116." I E E E Trans. T. N. A 105. Clin. Kurkijarvi. Appl.-P. Lammertsma. D . and H. A. Laine. M. Sunderland). A. 2 3 9 . J. Neurophysiol. and R. S.-T. " O n the spatial locating accuracy of multichannel magnetometers. Lett.. N . Phys. Tesche.2 4 1 . Lauterbur. Ketchen. V. and G. Williamson. "Channel capacity of multichannel magnetometers. 1987. edited by K. Vol. Lueck. M. W.. Phys." J. and J." J. Phys. Dallas. Tokyo). Kotani (Plenum. S. J. J. 939-944. Kohonen. Lounasmaa. Goubau. V. Koch. W. S. Hari. and M . M." Acta Neurol. T. J. Eng. Low Temp. J. D . 65. Magn. Sci. "Instabilities and chaotic solutions of the current biased dc superconducting quantum interference device. J. Hallstrom. 1989.1 6 . B. 1987. Stroink. Gopel. J. Kennard. 320-329. 3005-3008. Clarke.. B. "Design and fabrication considerations for extending integrated dc SQUIDs to the deep sub-micron regime. Clin. "Single-photon emission computed tomography. M. Magn. T. M. "Fourier imaging of electrical currents in the human brain from their magnetic fields. 1984. 1987. Erne. M AG-23.-L. Kullmann. 2145-2156. J. J. "Tangential bifurcation and critical indices in a Josephson junction. Scand. Experimental Principles and Methods Below IK (Academic.. A. 4322-4325. Koch. Hoke. 1992. K. 2nd ed. M. "Effects on dc S Q U I D characteristics of damping of input coil resonances. and construction of a D C SQUID with complete flux transformer circuits. Clarke. M. Hamalainen. Neurophysiol. Koch and H . 7 1 . Pegrum. Kaufman.. 837-842. in SQUID '85: Superconducting Quantum Interference Devices and their Applications. "Characterization of brain noise using a high sensitivity 7-channel magnetometer. Ketchen. Bhushan. D . Magn. Hari. B. Hoke. Kajola. 1974. J. Lu. S. S. W." in Biomagnetism '87. "Superconducting thin-film gradiometer. 217-227. 1982. 1988. Berlin). 1973. B. M. H .. N . Heino.." Proc.

Ono. W. and R. 1989." in Biomagnetism '87. P. Section A: Design and construction. W. 39. M. J. Finland). H. and M. Proc. J." SIAM J. L. J. 5. Hoke. edited by K. R. pp. "Human auditory cortical mechanisms of sound lateralization.. Williamson (Tokyo Denki University. Williamson. Appl. tightly coupled. Purcell. pp. 622-628. R.. H. Control 17. Leinio. and M. Johnson. Muhlfelder. M. Shoji..D. 303-322. 22-36. "Effect of ac impedance on dc voltage-current characteristics of superconductor and weaklink junctions. Appl. On the Detection of the Magnetic Heart Vector—An Application of the Reciprocity Theorem. "Hearing lips and seeing voices. 1988. 1991b. J. J. Montonen. Vol. 1964. 1984. W. Pizzella. 16. 1050-1055. A. R. M. Atsumi. 423-434. Lewis. and S.. pp. Hari. J. Hari. "Thick-walled conducting shield in biomagnetic experiments.. 495-503. Ricci. L. Y. Peters. Suppl. 729-732. L. "Noninvasive functional localization by biomagnetic methods. W. J. Montonen. Amsterdam). Phys. Idell. and F. H.. Comput. No. McCumber. McGurk. and T. 30. edited by M. McEvoy. Okada. 520.. B. 1993. Y. J. J. P. part I. "Noninvasive functional localization by biomagnetic methods." Biomagnetism: Clinical Aspects. R. J. "Attention and Brain Function" (Lawrence Erlbaum." in Advances in Biomagnetism. and T. and R. "The influence of head geometries on electroand magnetoencephalograms. 257-264. 201-288. 13. Katila. and A. 423-430. Irisawa.. 94-96. and L. Peters. Electron... Yahara. Katila. R. P. Naatanen. M. J. Valanne. SPIE 1351. Neurophysiol. Montgomery. S. Picton. Nenonen. 863-868. Neurophysiol. J. dc SQUID amplifiers. Math. G. Traversa. Meijs. Kotani (Plenum." J. and T. W. Katila. M." Med. and G. 386-389. pp. Okada. and H." Ph. Ahonen. Bellingham). J. M. and T. 23. 286-298. 16. Matsuba. Cromar. 426-429. Beall. part II. and M. Maclin. F. G. Erne. Numer. J. 69. April 1993 493 Boom. G. R. J. 1992. 1992. 49. M. Erne. J.. I. 1992. H." Jpn. Malmivuo.. and J. K." NeuroReport 3. "Biomagnetic measurements of spontaneous brain activity in epileptic patients. Mosher. 1992. Romani (Elsevier. J." Med. Beall. Seppanen. Directional selectivity. "Thermal noise in a magnetically shielded room. "The application of the Richardson extrapolation in simulation series of EEG's. Makela. L. "The Berlin magnetically shielded room (BMSR). 65. M. "Thermal noise of a biomagnetic measurement dewar." in Digital Image Synthesis and Inverse Optics.-D. Tokyo). 427-429. Meijs." in Advances in Biomagnetism *91: Clinical Aspects. R. J. V. 1990. Bosch. M.. and R. and H. Eng. R. Carelli. L. edited by S. 26. Kaufman. Shinoki. C. Phys. P. Kosaka. 39. H. LaHaie (SPIE International Society for Optical Engineering. and H. Nenonen. 746-748. 1985. Aoyagi. G. 364-375. P. Anal. "Neuromagnetic auditory evoked responses after a stroke in the right temporal lobe. Imada.. A. Lett. Phys. "An algorithm for least-squares estimation of nonlinear parameters.. 437-447. L. D. edited by A. Linnankivi. Acta Poly technica Scandinavica.: Magnetoencephalography man. "In search for short-duration memory trace of a stimulus in the human brain." Behav. T. pp. Sams. J.." Nature 340. (in press). and P." Psychophysiology 84. J. B. 1991." Electroencephalogr. Nenonen. and W. 410-419. Leoni. Ryhanen. Hari. C. 1992. 1983. Lekkala. thesis (University of Twente. Nakanishi." Appl." Electroencephalogr. S. Horacek. 107-112. Pulkkinen and P. A. M. S. S. "Magnetic responses of the human auditory cortex to noise/square wave transitions." J.-P. Y. and G. B. Nenonen. Clin. Mager. Katila." Electroencephalogr. Derka. 1976. Berlin). Marquardt. 1968. D." Ann." J. Liibbig (Walter de Gruyter. and S. A." J. E. 1988. "Auditory evoked magnetic fields after ischemic brain lesions. R. Heinonen. J. Biol. W.. Timlake. Ueno. low noise.. R. Magn. "Integrated dc-SQUID magnetometer." in Biomagnetism.. Helsinki). J." Hearing Res. and M. The Netherlands). J. M. C . Clin. Eng. Soc. 1987. S." IEEE Trans. Leahy. Lewis. dc SQUIDs. 1118-1120. "The use of multiple deflations in the numerical solution of singular systems of equations with applications to potential theory. Hahlbohm. Eng.. 54.431-441. 1990. edited by M. Biol. Williamson. 541-557. Lyytinen (University of Jyvaskyla.. "On the magnetic field distribution generated by a dipolar current source situated in a realistically shaped compartment model of the head. Okada.. J. P." in Human Action and Personality.. Hoke. Liibbig (Walter de Gruyter. Ono." in Biomagnetism. Makela. E. P. L. H. A..Hamalainen et a/. Malmivuo. T. Mod.and jU-rhythm in humans: a neuromagnetic study.. 1987.-D. G.. Electrical Engineering Series No. J. and A.. Naatanen. van Oosterom." Science 264. L. Erne.. "Multiple dipole modeling and localization from spatio-temporal MEG data. Hari. Torrioli. Katila. "Very low noise. F. Makela. 66. Stroink. P. J.9-10. and T. in press.. . J. G." Nuovo CimentoD 2. S. edited by S. Rev. 2. Phys. 1987. 1982. Naatanen. S. Nenonen. 25. Ind.. Hahlbohm. Lynn. Kotani. "Evaluation and reduction of the thermal noise of a biomagnetic measurement dewar. 1988. Clin. "Evoked a. 1990. "Well coupled. 51-78. 1987. Hari. R. Leahy. and I. M. Romani (Elsevier. W. Singh. Stroink. 39 (The Finnish Academy of Technical Sciences. pp. R. 1976. C. Neurol. T. 1968. da Silva. Tuomola. Neurophysiol. II. "Retinotopic map on the visual cortex for eccentrically placed patterns: First noninvasive measurement. N. Eng. Appl. N. and J. H. Hoke. H. Clin. I. Romani. New York). Amsterdam). 11. Meijs. "Superconducting shield for biomagnetism measurements coupled with ferro-magnetic. C. 1986. pp. Gmitro. Naatanen.. C . "The role of attention in auditory information processing as revealed by event-related potentials and other brain measures of cognitive function. J. Biomed. edited by L. J. Makela. Cromar. and J. "DC-SQUID system for high-resolution biomagnetic measurements. 76-82. Katila. Part 1. M.. D. N. T. B." J." Brain Res. 1985. 3113-3118. D. "Multiple dipole modeling of spatio-temporal MEG data. and F." IEEE Trans. 1981. and W.. 375-425. "Different analysis of frequency and amplitude modulations of a continuous tone in the human auditory cortex: A neuromagnetic study. Rossini. G. 27. Muhlfelder. H. Romani. "The Nl wave of the human electric and magnetic response to sound: a review and analysis of the component structure. Comput. Eng. "The gradient in the sampling of TVdimensional band-limited functions. R. Katila.. 222-226. Hillsdale). and P. Erne. S. Modena. Nenonen. Berlin). N. Leinonen. M. B. Barbanera. C. Maniewski. Narici. 1991a. 1963. 1987. Brain Sci. MacDonald. S. and A. 222-231. M AG-21. Hovi. T. 1981. edited by S. L. Clin. Koyanagi. Mosher. and T." Hearing Res..

Baltimore/Munich). S. RechtschafFen. Natl. K. M. (Cambridge University. Kim. and A. E. Leoni. and W. "Auditory attention affects two different areas in the human supratemporal cortex. Kaufman. 1989. 1962.. Phys. and W. 567-570. and S. E. Guant. Niedermeyer and F . Brown. and R." Electroencephalogr. Okada. Magn. Liitkenhoner." I E E E Trans. London 226. I E E E 62. Makela. "Electron trap states and low frequency noise in tunnel junctions. Netzer. "Thermal agitation of electric charge in conductors. "Intrinsic signal changes accompanying sensory stimulation: Functional brain mapping with magnetic resonance imaging. Veldhuis.. Petersen. P. "Principles and clinical validity of the biomagnetic method. Teszner. IM-36." Ann. G.494 Hamalainen et aL: Magnetoencephalography Katila. G P O . J. Y. M." in Advances in Biomagnetism. J. Tanenbaum. J. "Identification of sources of brain neuronal activity with high spatiotemporal resolution through combination of neuromagnetic source localization (NMSL) Rev.. R.. 45. J. B.. 632-636. A. K. Middleton. "Magnetocardiographic functional localization using a current dipole in a realistic torso. T. 3 8 2 . Trans. "Source parameter estimation in inhomogeneous volume conductors of arbitrary shape. pp." Proc. 261-264. Stroink. A." Inf. P. C. P. Vapalahti. and J. Lassen." I E E E Trans. K. M. "The design of low-noise amplifiers. edited by H . 1991. Kumpf. 299-361. 1992. J. Instrum.3 2 3 . 75. H. Kleinsasser.. "Somatotopic organization of the human somatosensory cortex revealed by neuromagnetic measurements. Okada. Nousiainen. "Reconstruction of multidimensional stochastic fields from discrete measurements of amplitude and gradient.. 1984. M A G . and T.. 2 7 9 . 1954." Electroencephalogr. Roland. 1989. Basic Principles. Hahlbohm and H. " A new interpretation of noise reduction by matching.." I E E E Trans. Petersen. and J.. Lehnertz. 266-272. Kajola. U. Fahrendorf. Y. 5951-5955. Sams. 103. R. Neurophysiol. Anogianakis." Nuovo Cimento D2. Stober. Kajola. Richardson. Y. Robinson. pp. Hoke. R. " H u m a n auditory evoked gamma-band magnetic fields. M. Kotani. Middleton.-G. P. Pantev. M . "Environmental noise cancellation for biomagnetic measurements. M. New York). S. J. T. F. J. Y. M . D . Ohta. M." Phys. Pantev. Neurophysiol. W. pp. Vetterling. Biomed. 7 7 0 . D. Hari. 7 2 8 ." in Biomagnetism: Applications & Theory." I E E E Trans. and M. C . pp. U. 721-724. 1985. R. 519-536. R. J. Pechura. 1139-1151. R.. 184-194. and R. A. J. J. "Sampling and reconstruction of wave-number-limited functions in TV-dimensional Euclidean spaces. Kotani (Plenum. D ." Proc. and M. 2. and J." J. Liitkenhoner. I E E E 69. Acad. "Multichannel instrumentation for biomagnetism. Pantev. Washington. and B. Paetau. 1989. 32. G." Philos. Romani. Mapping the Brain and its Functions (National Academy. 1928. Numerical Recipes in C. Magn. T. Romani. Kaukoranta. Tiihonen. Jasper. Berlin). Kales. " M E G localization of epileptic cortex—impact on surgical treatment." Acta Otolaryngol. Niedermeyer. Salustri. and L. M. Flannery.. "Focal activations of human cerebral cortex during auditory discrimination. K o tani (Plenum. 1658-1661. Clin. 1992. Matsui. USA 89. and N . Tuomisto. Hoke. 1987. Stroink. and K. Neurophysiol. Hampson. 1981.. DC). S. Liibbig (Walter de Gruyter. Hari. Biomed. pp. D . A. The Cerebral Cortex of Man (MacMillan.." Hearing Res. van Oosterom. A. 69. 1018-1021. Meas. Hanasaka. 1988. "Seven-channel rf S Q U I D with 1 / / n o i s e s only at very low frequencies. "Magnetic field tomography of coherent thalamo cortical 40-Hz oscillations in humans." Electroencephalogr. 201-204. K.2 3 .4 0 6 . Penfield. P. J. F. M. edited by S. 1983. N . 32. Hoke. Kajola. 1989. M. W. C . Lehnertz. 464-472. Singh. M . and C.. USA 88.. 79. 7. "Cortical activity evoked by a multichannel cochlear prosthesis. "Tinnitus remission objectified by neuromagnetic measurements. Gallen.. 1968. B. S. "Tonotopic organization of the human auditory cortex revealed by transient auditory evoked magnetic fields. and H . Hoke. B. Williamson. DC). Komiyama. M . Boston).. Williamson. 25. " D e p t h electroencephalography. "Rule-based location of multiple current-dipole sources from biomagnetic data. L. Singh. S. Partanen. W. and U. S. Press. 1974. and C. Netzer. Duret. Mod. Rogers.. A. K a r h u . M. Ugurbil. 56. D . and R. Y. Teukolsky.." Med. Hasson. April 1993 and magnetic resonance imaging (MRI). 11. 1991." in SQUID 985: Superconducting Quantum Interference Devices and their Applications. Rev. A. Rif. Liitkenhoner. and L. R. F. Clin. Menon. Ribary." Inf. Bolton. Penfield. Lado. E. 1986. Wittkowski. Williamson. Sandstrom. L. Mogilner. Takahashi. Merkle. 1990. " A manual of standardized terminology. M. "Josephson tunnel junctions with refractory electrodes. Llinas. K. C . Brain Res. Hoke. Ellermann. Stroink. Hamalainen.. Montandon. "Magnetoencephalography in the study of epilepsy.7 7 5 ." Exp. 8996-9000. I.. 197-205. Cambridge). Y.3 9 1 . H." Neurophysiologie Clinique20. 1964. edited by E.7 4 1 . E. Paetau. and D . 5. M. R. 1982. Hoke. Epilepsy and the Functional Anatomy of the Human Brain (Little. Palfreyman. G. L. and R. Lehnertz." in Electroencephalography. Eds. P. Acad. C. 38. Acad. Swithenby. Nyquist. W. 2nd ed. T. Galambos.S. S. B. G.. "Neuromagnetic measurements at hospital: Instrumentation and preliminary tests. R. Gallagher. Pelizzone. "Optimization of the dynamic behavior of a S Q U I D system using an electronic simulation. Ioannides. C . "The deferred approach to the limit." Proc.. "Discrimination of localized and distributed current dipole sources and localized single and multiple sources. W. 110-113. 36. 1989.. New York). Natl. Washington. K. B. O . and T. Ogawa. Martin. Granstrom. Tank. and A. 1991. 1985. 11037-11041. W. G. H. I. A. 1991. and R. Rasmussen." Proc. Katila (Pergamon.. 658-664.. Progr. M. 106-109. No.. 40. techniques and scoring system for sleep stages in human subjects" (U. USA 88.. Raider. and T. Lopes da Silva (Urban & Schwarzenberg." I E E E Trans. edited by S. SkinhizJj. 4 0 4 . 1927. Vol. Magn. Control. K o r k m a n .-L. 1987. and M. R. Eng. Narici. Oostendorp. Sci. Raider. Shinada. R. H. and W. 919-932. J. Neurol. M AG-21. Hari. Clin. L. S. D . Buhrman. Flokstra.. . 173-184. Soc. 445-476. M. R. G. Hamalainen. E. M . Uchikawa. 1987. A. Neurophysiol. Natl. Hari. Weinberg. D . D . and P. 1991. G. S. New York)." I E E E Trans. Hari. 110-117. Pantev. M . 1985. 1991. New York). Control. 1992. Rillo. 1990. Yamada. Makeig. 169-187. S.." Proc. 160-170. 65. Sci.. D ." NeuroReport 2." in Advances in Biomagnetism. Takahata. Nicolas. Y. Sci. "Landau-KlefFner syndrome: epileptic activity in the auditory cortex. Eng. R. Clinical Applications and Related Fields. T. edited by H. C . 1981. Paetau. 1950. and D . Technol.

2. 61. L. S." Appl. Smith.. Llinas. R.Hamalainen et al. M AG-23. W. "Neocortical propagation in temporal lobe spike foci on magnetoencephalography and electroencephalography. Suk. R. Wirth. Rose. "Frequencyspecific sources of the auditory N19-P30-P50 response detected by a multiple source analysis of evoked magnetic fields and potentials. C . Clin. edited by S. D . 1991. Kotani (Plenum. Hoke. M . and R. edited by K. 25. Atsumi. J. S. and M . 4 7 0 ." Electroencephalogr. 3013-3016. 65. 1989. 1991.11-22. and H. 329-335. M. "The human auditory sensory memory trace persists about 10 s: neuromagnetic evidence. "Tonotopic organization of the human auditory cortex. J.. J. pp." Biophys. Antennas Propag. 1339-1340. Neurophysiol. Sams. "Aluminum shielded room for biomagnetic measurements. Berlin). " A c biasing of a dc S Q U I D . 1989. L. Katila. S. 1083-1086. J. J Williamson. New York). E.. "The form. "Dc-SQUID electronics based on adaptive positive feedback: Experiments. 1990. Satrapinski. Knuutila. E.. H. Lett." I E E E Trans. Vol. Kotani." Electroencephalogr. 1987." in Advances in Biomagnetism. Biomed. Neurol. Biol. Yamamoto. T. R. S. Stroink. 346-350. I. 48.6 9 . Hari." in Advances in Biomagnetism. K. Kiviranta. and J.. Ed. 78. 1992. R. Scherg. Stewart.. 4 0 . "The influence of model parameters on E E G / M E G single dipole source estimation." in Advances in Biomagnetism.. 64. and H . 1991." Rev. D . Shaw. M. M. Koch. W. U. 2 7 9 . W. G. New York). M. Seppa. "Application of a multichannel MEG-system in temporal lobe epilepsy. Magn. and W. Feistel. J." I E E E Trans. P. Hamalainen. 1983. edited by S. "Practical low-noise integrated dc superconducting quantum interference device magnetometer with additional positive feedback. 1982a. J. 1978. Barth. 1989. Silvey. and M. 1987. Simmonds. Sams. Patent N o . Sarvas. 76.. Romani. M. Cantor. A. "Effect of parasitic capacitance and inductance on the dynamics and noise of dc superconducting quantum interference devices. 2 4 4 9 .\ Magnetoencephalography 123-159. Hoke.5 0 ." in Superconducting Devices and Their Applications. 2 7 7 . M. 9 3 . Seppa. 3 7 3 . Hoke.1 0 0 . Ryhanen. Vol. Stroink. "Model order determination and limits of source resolution for multi-source neuromagnetic data: Simulation studies.. Seppa. " A coupled D C S Q U I D with low \/f noise. London). 1987. 27. Romani (Karger. L. J. M. and S. Blackford.2 8 2 . S." J. 71.. Kotani (Plenum. 1992. and L. H. Williamson. A... Vol.4 7 3 . Salmi. pp. and H. Phys. Stehling. Gronberg." Rev.2 9 . Clin. and T. Ryhanen. and A. 4 3 . 1990. Vol. Kaukoranta. 276-280. Cantor. M. Sams. C. Knuutila.. G. 6150-6166." Science 216." I E E E Trans. Sato. 1993. (in press). Stroink.. U. Hari. C . Phys. Folberth." I E E E Trans. Cappell. New York)." I E E E Trans.1 0 9 . 1989. 1968. Seppa. H. J. "Cortical activity elicited by changes in auditory stimuli: Different sources for the magnetic NIOOm and mismatch responses.. and J. Low Temp. R. Mod. Suni. M. "Current-voltage characteristics of Josephson junctions. 1989. 1993. R. 1977. Sams.. edited by S.. "Development and performance of a multichannel system for studies of biomagnetic signals of brain and heart. G. J. Neurosci. M.. H . Kaufman. Ueno. S. and S. H. 6 6 9 . B.6 7 2 ." in Biomagnetism 987.. McEvoy. "Seeing speech: visual information from lip movements modifies activity in the human auditory cortex. 1815-1845." in Advances in Biomagnetism. New York). "The magnetic field of a single axon. Magn. Appl.." Science 238. Seppa. Biomed. 4389612.. A. Roth. Neurophysiol. Horacek. Abraham-Fuchs. T. and J. and P. Aine. Roth." I E E E Trans. 1990. 287-386.2 4 5 1 . (in press). H. Magn. R." Rev. Response to changes in interaural time difference. Bauer. J. "Influence of the signal coil on dc-SQUID dynamics. R. H. No.1 9 6 . Schneider." J. 1980. and G. Neubauer. Phys. New York). Sci. J. J. R . 1991. Sci. H. R. 32. Stok. 385-402. " S Q U I D magnetometers for low-frequency applications. Ryhanen. M. W." Science 254.. voltage distribution and physiological significance of the K-complex.. 1991. 12. K. and R.2 8 0 . Wikswo. "Biomagnetic instrumentation. G.. pp. Instrum. J. and M . "Basic mathematical and electromagnetic concepts of the biomagnetic inverse problem. " U. O. and I. " A high open-loop gain controller. pp. "Dc-SQUID electronics based on adaptive noise cancellation and high open-loop gain controller. T.. Tokyo). 1993. 4 6 3 . M. 54 (Raven. Scott. Seppa. pp." Appl. J. Cogn. (in press). Drung. Hari. Hoenig. 1991. M. "Proposal of multichannel S Q U I D amplifiers for biomagnetic measurement. J. 1988. and D . Daalmans. G. and J. 9 7 . J. J. Ilmoniemi. D . G. 1989. R. S. M. Koch. 141-145. Green. Hoke. Rev. T." I E E E Trans. Stroink." Ann. K. 1981. Sato. 127. "Effect of parasitic capacitance on dc SQUID performance. AP-34. L. "Echoplanar imaging: Magnetic resonance imaging in a fraction of a second. Romani. J. and R. Phys. Statistical Inference (Chapman and Hall. and H." Psychophysiology 28. Kotani (Plenum. Seifert. 6 of Advances of Audiology. Seppa. O.S.-T. Williamson. Williamson. Hamalainen. Kaufman. "Multiple emitter location and signal parameter estimation. Shirae. M. and J. Huk. Williamson. Basel). J. G. edited by S. 185 . edited by F.. and M . S. Drung. S. Mansfield." J. Abraham-Fuchs. H. Seppa. K o c h and H ." in Auditory Evoked Magnetic Fields and Electric Potentials. O. 2 1 . Vilkman. Katayama. Ryhanen. and L. BME-34. 1985. "Fundamentals of dipole source potential analysis. Simola. 121-664). T. Knuutila. H. M. Ribary. Ahonen. 289-296. J. Sutherling. Magnetoencephalography. M. Springer Proceedings in Physics. . Brown. B. 2488-2490. H. J. Lett. K.. Simola.. pp.. Hamalainen. Reichenberger. (in press). Rohrlein. Hoke. and M. Phys. Proceedings of the Fourth International Workshop on Biomagnetism (Nuovo Cimento D 2. and C. Naatanen. G. Hari..4 6 8 . Hamalainen. Furukawa. Med. S. 228-230. D . Grandori. Yamamoto. Williamson. B. 1956. Lounasmaa. L. "Anatomical localization revealed by M E G recordings of the human somatosensory system. 1987. G. Cappell.. Instrum.3 8 1 . P. Scherg. 1991. H. Llinas. Schmidt. edited by H . Stefan. G. Lett. 1986. Sipola. 8. April 1993 495 Schneider. and M. Aulanko. Supek. V.. J." Hearing Res. "Magnetic localization of somatically evoked responses in the human brain. Eng. 27. Sci. M. New York). and L. Suk." Phys. and S.. T. U. T. Turner. Kotani (Plenum.. Williamson (Tokyo Denki University. Rif. "Magnetoencephalography and epilepsy research.. pp.. and V. Magn. Liibbig (Springer. Katayama. " H u m a n auditory cortical mechanisms of sound lateralization. 59. Lu. 165-168. Eng. M. Jr. Williamson. 1993." Neurosci. J. Stroink. 53. 1982b.. 52. Romani. Eds. H . Instrum. Ribary.. Neurophysics (Wiley. R..

" J. J. Vvedensky. Altrocchi." in Advances in Biomagnetism. "Inductive monopole detector employing planar high order superconducting gradiometer coils. S. 1991. 27. 159-170. D . M. C. Torrioli. V. McCubbin. J. 665-675. van Harlingen. Notarys. Chen. D . C. "Improvement of the performance of a jU-metal magnetically shielded room by means of active compensation. Phys. Eng. Kotani (Plenum. 683-692. Burbank. G.." I E E E Trans.. and B. Erne. "Assessment: Magnetoencephalography (MEG). Romani (Elsevier. 1993.. 733-736. Foglietti. P.9 8 . Phys." J. 60. and D . Stroink. M AG-21. 1977. Kotani (Plenum. Kajola. J. Comput. 1982. 55.. and M. G. J. "Magnetic field fluctuations arising from thermal motion of electric charge in conductors. H. 49. "Inverse problems = quest for information. Lee. Clin. Hoffman. D . New York). Vrba. J. K. ter Brake. Valette. Hari. Stroink. L. and L. Ketchen. Biomed. 1982. Okada. A. Report of the therapeutics and technology assessment subcommittee of the American Academy of Neurology. 1990. N . Tesche. 4 5 . April 1993 van den Noort.6 0 1 . Neurol. Sci. Chi. Ketchen. 1987. F . J. Speech Signal Process. 1983.. and T. Modena (Plenum. 384-386.." Clin.-L. Burbank. C. J. Strackee. Tiihonen. Jacobs. S.. J. H. 1992." in Magnetoencephalography. Wang. W. Suppl. Kaufman. Tiihonen. C. New York). van Oosterom. 1985. S. Biomed. 1990. P. Cohen. Kubik. Weinberg. edited by S.. Brown. A. "Magnetic mu rhythm in man. Tripp. G.6 0 . D.. 793-800. 50. Wang.. Lett. M. B 81-86. Thomas. T. 54 of Advances in Neurology. New York). G. I." I E E E Trans. and M. Del Gratta. ter-Pogossian." Neurosci. and J. J. Naurzakov. Biol. 43. S. "High resolution single channel neuromagnetometry and minimal approach for multichannel systems. Phys. A. A. B." in Biomagnetism: Applications & Theory. George. J. 32. J. A. and T. Burbank. Wax. Lett.. Williamson. Ahlfors. A. Rogalla. P. L. J. pp.. "Localization of epileptic foci using a large-area magnetometer and functional brain anatomy. L. T. and S. F. 303-305. Tsuei. K. Phys. and H. "Second order gradiometer and dc S Q U I D integrated on a planar substrate. A. Kajola. P. 1032-1035. Vvedensky. 1985. and D . and H. 1985. 4015-4021. Physiol. T. "28 channel hybrid neuromagnetometer.. 12. "Noise cancellation in biomagnetometers. J. van Ancum. Hoke. L. J. Weinberg. 387-392. R. Teyler. Vieth." Nuovo CimentoD 2.. pp. P. J. L. Callegari. G. New York).496 Hamalainen et a/. "Study of the alpha rhythm with a 4-channel S Q U I D magnetometer.." Meas. 1985. H . 1985. Kim. Vrba. J.-Z. 593-595. Varpula. Cheung. M. nonexponential kinetics in condensed matter. Sato (Raven. Rev. Low Temp. Williamson. V. and V. A. Duret.." in Biomagnetism: An Interdisciplinary Approach. 17. K. Hari. A. Crandall. Phys. 3 0 1 . M. Kleinsasser. pp. Clarke. de Waal. R. . Katila (Pergamon. Vol. P. Technol." Radiology 114. 60." Med. U." J. and P. J. H. Hari. A. Acoust. 689-692. V. M. M. Sandstrom. T. Boundary Elements 7. A. 46." Appl. 1988. M . and D . 129. R. "The L A N L gradiometer orientation device. T. Hari.. Tesche. J. C. 12. W.5 0 . J. V. Phys.. 1989. M. "Magnetoencephalographic 10-Hz rhythm from the human auditory cortex. "The visual evoked magnetoencephalogram. Ilmoniemi. "Biomagnetometers for unshielded and well shielded environments. Vrba. (in press). 1983." I E E E Trans." in Advances in Biomagnetism. Lee. " A positron emission transaxial tomograph for nuclear medicine imaging (PETT). Ferguson. M. L. W. A. J. Darcey. Williamson. Fife.. and N . 39. Weinberg. and R. Y. S. Heller. Martinez. F. R. 124-129. 1985. 1991. Greiner. B. B. Gallagher. V. G. Suppl. Physiol. H. ASSP-33. A. J. edited by S. J. A. "Detection of signals by information theoretic criteria. edited by S. 5 7 ." Ann. New York). Flynn. Williamson. 29." Can. Weinberg. C.." I E E E Trans.. Rogalla. Poutanen. Tarantola. Betts. No. and T. Amsterdam). J. W. 1988. Tindall. J. C. Burbank... Cheyne. B. F. Equations of Mathematical Physics (Marcel Dekker. "The magnetic and electric fields agree with intracranial localizations of somatosensory cortex/' Neurology 38. Phys. Mullani. edited by H . 8 9 . M . Tuomisto. H. A. Taylor. Phelps. Urankar.noise and other slow.. Kaufman. G. . ter Brake. Anal. Pizzella. H. Becker. Spear.3 3 1 . M . B.. Jones. S. 64 channel SQUID biomagnetometer system. van Nieuwenhuyzen. S. Brickett. C. S. R. 101-139. Meas. The Brain—An Introduction to Neuroscience (Freeman. 5 9 6 . B. E.. Fife. Appl. Lett." in Biomagnetism: Clinical Aspects. 1988. Tissari. G. 1975. G. Hoke. H. A. 23. G. and H. Wieringa.: Magnetoencephalography Sutherling. "Measurement of the tangential component of the magnetic field associated with rhythmic alpha activity in the human brain. Magn. Stammis. K. W. P. M. Eng. Shabanov. "Studies of auditory evoked magnetic and electric responses: Modality specificity and modelling. 261-269. "The U T 19-channel dc squid based neuromagnetometer. 1060-1073. C. H. J. R. McCubbin. Kajola. C. Jaszczuk. Vol. A. and E. Inverse Problem Theory (Elsevier." I E E E Trans. . H. Part 1. 1992. Poutanen. V. 1986. Brin. D. M. 1992. Proto. and T. G. Vapalahti. and P. Folkstra. C. "Low-frequency noise and discrete charge trapping in small-area tunnel junction dc SQUID'S" Appl. J. edited by M. and M. pp. and S. H . Stroink. New York). "Physical concepts and mathematical models. " D c SQUID: Noise and optimization. J. M . "Magnetoencephalography in the study of stroke (cerebrovascular accident). M. 1983. N . Levesque." Cryogenics 28. "Spatial discrimination in SQUID gradiometers and 3rd order gradiometer performance.. and G. Kailath. Katila. Meas. and R. Katila.. Weissman. Casciardi. L." Eng. L. T. J. Tillotson." Appl. Eng. and I. M. S. Stroink. J. S. Pasquarelli.. Phys. Kajola. 1985. T. McKenzie. K a r h u . Wakai. M.. D . B. 1990. B. A. 11-12. S. Biomagnetism: Applications and Theory (Pergamon. M. T. 4 3 9 . R. Taylor. Geophys. Barth. M. Kubik. pp. K. 1989. Medvick. " 1 / / . and M . J. Thompson. 7 8 3 . Nousiainen. 1983. Fife. 1991. Chaudhari. 2. Kittredge. Vrba. J. BME-30. and M. 1705-1714. New York). Eds. Y. and M. 1989. Varpula. 843-846. G. "The solid angle of a plane triangle. Hoke. 65.283-290. 1985. Romani. Magn. J." Life Sci. Ozhogin. Vladimirov.471-483. New York). "Practical dc SQUIDs with extremely low \/f noise. P. Markham. H. M. R. 125-126. and J." Neurosci. M c K a y . Yogi.. Lee. Cufiin. E. Lett. M. J. and T. and T. 1971. 1991. T. Mod. 1984. M.4 4 3 . McCubbin. B.. Kleinsasser. B. Romani. Phys. M. "Magnetic source images determined by a lead-field analysis: the unique minimum-norm least-squares estimation. D. Tesche. Fife. Haid. Haid. "Whole cortex.. Tarantola. J. L. Suppl.7 9 5 . pp. New York)." Rev. Phys. J. and D . "Common compact analytical formulas for computation of geometry integrals on a basic Cartesian subdomain in boundary and volume integral methods. Tiihonen. A. R. Greenberg.. edited by S.. C. W. and R.... L. J. Nuwer. McGavran. 2.. Mod." van Hulsteyn. B..

. F . Barach. pp. 22. C." Rev. Williamson. P. Stroink.5 5 . 1987. Williamson. "Magnetic localization of neuronal activity in the human brain. Williamson. J. "Macroscopic quantum interference effects through superconducting point contacts. Kingston. Series A: Life Sciences (Plenum. 41.. 4 6 . New York). J. "Current probe system for measuring cellular action currents. Katila (Pergamon. 30. "Five channel S Q U I D installation for unshielded neuromagnetic measurements... Stroink. and V. Weinberg. S. Suppl. J. Appl. Weinberg. 3 6 7 ." I E E E Trans. Llinas. S.. Acad. Comput. and P. H. Sci. "Integrated D C S Q U I D magnetometer with a high slew rate. 121-125. Phys. P. 8 8 . 772-774.. Heiden. 702-710. Clarke. 48. L.3 7 5 . and I. 1981. Stroink. " O n some nonparametric methods for detection of the number of signals. Jr.." in Biomagnetism: Applications & Theory. Vol..6 . "Lowfrequency noise in dc superconducting quantum interference devices below 1 K . J. Pelizzone. Okada. No. P. Romani. 1983. 1977." in Biomagnetism: Applications & Theory. Hoke. Phys. L. 8 3 ." Phys. C. "Biomagnetism. Natl. Urbina. Henry. 27. Wikswo. Kaufman. Roth. Clarke. Eng. Yin. Wikswo. W. pp. and L. and J. Rev. and T.. S. G. A. and T. and J. Phys. Vol. G. and J. P. Speech Signal Process." Science 208. C. Q. G. 1572-1580. Neurol. P. Acoust. C. B. Williamson. Sci. Zimmerman. "Thin-film flux transformers of YBa2Cu30 7 _ x . Jr. 1988. Zimmerman. P. 1991. J. Clarke. 1984. Crum. J. J.. 129-139. pp. S. C . New York). Lett. 3 . Magn. Advances in Biomagnetism (Plenum. J. Zimmerman.. Jr. "Magnetic measurement of propagating action potentials in isolated. 1985. Stroink. Weinberg. " M E G versus E E G localization test (Letter to the Editor). J.. Modena. Williamson. edited by H.. Abraham. " Appl. Kaufman. New York). 222. C . J. Y. 1989. Kaufman.Hamalainen et aL: Magnetoencephalography 537-571.. New York). J. 55. edited by H. edited by H . Rev. R. and J. S. 65. L.9 2 . 1987. Stroink. 1533-1538.. J. Katila (Pergamon. F . E. and R. 2. 1985. 23. L. S." Proc. J. and R. and A. 2. Wikswo. Freeman. Kotani. G. Magn. 5 3 ." J. USA 85. ASSP-35. 8732-8736. Y.. 1989. Jr. Mat." in Biomagnetism: Applications & Theory. Magn. Ferrari. F. pp. P." J. Williamson. 141. "Magnetic measurements on single nerve axons and nerve bundles. 129-201. Marsden." Ann. "Magnetic field of a nerve impulse: First measurements. R.. G. Instrum. Krishnaiah. T. P.. Phys." J. New York). S.8 7 . 1966.. Thiene. Nicholson. Wellstood. onedimensional cardiac tissue preparations. M . J. Jr. and T. Biol. T. "Advances in neuromagnetic instrumentation and studies of spontaneous brain activity. J. G." Brain Topogr. 1985. Eds. New York). D . S. 1970. E. P. " S Q U I D instruments and shielding for low-level magnetic measurements. J.5 1 ." I E E E Trans. Harding. Samson." Med. 1985. 1985. Kaufman. "Design and operation of stable rf-biased superconducting pointcontact quantum devices and a note on the properties of perfectly clean metal contacts. 50.. and B. 952-957. Appl. J. G. J. Williamson. 66 of NATO Advanced Science Institutes Series. and J. M . P. and L. Hentz. 1980. edited by H ." in Biomagnetism: Applications & Theory. Part 1. and J. Kaufman. Mod. Silver. Katila (Pergamon. Biomagnetism: An Interdisciplinary Approach. Katila (Pergamon. April 1993 497 Eds. M.. . and M .. 2569-2572. J. "Magnetic assessment of regeneration across a nerve graft. Yamamoto. E.. and T. Wikswo. Weinberg.. Wikswo.. Wellstood. 1991.. R. GifFard. Wellstood.

FIG. of course. 10. in a helium dewar and the brain is inside the skull! The measuring head of the gradiometer and the tip of the dewar are concave so that they fit as closely as possible to the subject's head. Ahonen et ah. 1989. During an actual measurement the instrument is. . 1991) at a typical measurement distance above a model of the brain. The 24-channel neurogradiometer in Helsinki (Kajola et ah.

many important brain structures are clearly seen.0-T MRI system at the Helsinki University Central Hospital. . The images were acquired with the Siemens 1. Coronal section (a) at the level of the ears and sagittal (longitudinal) section (b) near the head midline of MRI's made on one of the authors.FIG. 13.

. Each region G. 18. has uniform conductivity a. .FIG. Multicompartment conductivity model. Unit vectors normal to the surfaces are denoted by n».

All source locations are within a sphere of 15 mm in diameter (Hamalainen. . 20. 1991).FIG. The white lines indicate transsection levels of the companion slice. Example of source locations superimposed on coronal (a) and sagittal (b) MR images. The white dots show positions of equivalent current dipoles corresponding to the 100-ms deflections of auditory evoked responses in independent repeated measurements.

F.8. (1991). (b) Auditory evoked responses to tone bursts recorded with the BTi 37-channel instrument over the left hemisphere.-100 bOO ms FIG. are attached to the subject's head and to the Dewar (courtesy of Biomagnetic Technologies. Inc. discussed in Sec. . The approximate positions of the sensors are shown on the schematic head. The interstimulus interval was 4s and 384 responses were averaged. BTi "Magnes" system. From Pantev et al. 39. The coils of the probe-position indicator. (a) 37-channel gradiometer of the BTi "Magnes" system in a measurement situation.). V.

Stimuli employed in the MEG study of the McGurk effect. Articulation: pa pa ka pa Sound: pa pa pa pa Perception: pa pa ta pa . For details. 49.FIG. see text.

Study of an epileptic patient. The y channel of unit 3 was not functioning properly and is omitted. 1 . The upper traces reflect the vertical and the lower ones the horizontal gradients. •' i " y " n *.. The inset illustrates ten superimposed single spikes from the y derivative of unit 8. (1991). (Bottom) ECD locations. The white lines indicate transsection levels of the two other slices. . •. -V\^ VVv^^-^^^V^^^</V^ A > W V WVVVVV From left From below From front FIG. (Top) Examples of spontaneous spike-and-wave discharges from the left hemisphere of a child with the Landau-Kleffner syndrome. shown by white squares of ten single spikes superimposed on MRI slices.5 -y—r fv^*• -r.*>«*. 63. i . Modified from Paetau et al.*** A * ^ » \ *•*.«*> 9 -y-^*-W*s..

6). The room is a cube of 2. and three layers of aluminum. 9. The subject has her head inside a 122channel neuromagnetometer (see Fig.F. . 41 and Sec. which attenuate very well the high-frequency band. 1982). The shielding factor is 103-104 for low-frequency magnetic fields and about 105 for fields of 10 Hz and above. which are effective for shielding at low frequencies of the external magnetic noise spectrum (particularly important for biomagnetic measurements). During actual measurements the subject is usually left alone and the double doors are closed. View of the interior of the magnetically shielded room at the Helsinki University of Technology (Kelha et al. V..FIG.4-m inner dimensions with three layers of /z-metal.