Clinical Neurophysiology 116 (2005) 1858–1869

www.elsevier.com/locate/clinph

Tibial somatosensory evoked potential intraoperative monitoring:

Recommendations based on signal to noise ratio analysis of popliteal
fossa, optimized P37, standard P37, and P31 potentials
D.B. MacDonalda,*, Z. Al Zayedb, B. Stigsbya
a
Section of Neurophysiology, Department of Neurosciences, King Faisal Specialist Hospital and Research Center, MBC 76,
P.O. Box 3354, Riyadh 11211, Saudi Arabia
b
Section of Pediatric Orthopedics, Department of Orthopedics, King Faisal Specialist Hospital and Research Center, MBC 76,
P.O. Box 3354, Riyadh 11211, Saudi Arabia
Accepted 27 April 2005

Abstract

Objective: To compare the intraoperative signal-to-noise ratio (SNR), reproducibility and rapidity of popliteal fossa (PF), optimized P37,
standard P37 and P31 potentials.
Methods: Raw sweeps and 11 averages doubling sweep number from 2 to 2048 were compared in 37 patients undergoing scoliosis surgery.
Optimized (highest amplitude or SNR) P37 derivations were Cz–CPc (22), CPz–CPc (27), Pz–CPc (7), iCPi–CPc (8), CPi–CPc (1), Cz–Pz
(2) or Pz–FPz (3), and in two patients with non-decussation, Cz–CPi (1) or CPz–CPi (3). Standard P37 and P31 derivations were CPz–FPz
and FPz–C5S. Signal amplitude was measured in 2048-sweep averages; peak noise was measured in raw sweeps and G averages; SNR was
amplitude/noise. Visual superimposability and !20–30% amplitude variation determined reproducibility. Sweeps to reproducibility
determined rapidity.
Results: The SNR order was PF[optimized P37Ostandard P37OP31. Mean optimized P37 SNR advantages over the standard P37 and P31
were 2.1:1 and 4.9:1. SNR had powerful non-linear correlations to reproducibility and rapidity. Median sweeps to reproducibility were PF: 2,
optimized P37: 128, standard P37: 512 and P31: 1024. EEG noise was greatest in FPz derivations. Burst-suppression increased scalp
potential SNR and rapidity.
Conclusions: Optimized P37 and PF recordings are most rapidly reproducible due to superior SNRs and are recommended. FPz should be
avoided. Burst-suppression may be desirable.
Significance: CPz–FPz and FPz–C5S should no longer be standard.
q 2005 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.

Keywords: Tibial somatosensory evoked potentials; Signal to noise ratio; Intraoperative monitoring

1. Introduction increases with the averaged SNR, but a surgical feedback

Rapid reproducibility is fundamental to effective soma-
tosensory evoked potential (SEP) intraoperative monitoring
(IOM). Low signal to noise ratio (SNR) potentials require
averaging, which reduces noise by approximately the square
root of the number of averaged sweeps (N). Reproducibility
* Corresponding author. Tel.: C966 1 464 7272x32772; fax: C966 1 442
4763.
E-mail address: dbmacdon@yahoo.com (D.B. MacDonald).
1388-2457/$30.00 q 2005 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
doi:10.1016/j.clinph.2005.04.018
delay theoretically proportional to NZ(averaged SNR/raw
SNR)2 occurs. Hence, very low raw SNR potentials may be
too slowly resolved with the many sweeps needed for
reproducibility, or non-reproducible, and therefore, unreli-
able with the few sweeps desired for rapidity.
Non-invasive tibial SEP monitoring is challenged by low
raw SNRs. The popliteal fossa (PF) potential provides
technical control while the standard cortical P37 recorded
with CPz–FPz provides a proximal monitor (American
Electroencephalographic Society, 1994b). However, P37
 D.B. MacDonald et al. / Clinical Neurophysiology 116 (2005) 1858–1869
derivations optimized to highest amplitude seem to have
higher SNRs than CPz–FPz (MacDonald et al., 2004b). The
subcortical P31 from FPz–C5S is often recommended as
another control or monitor resistant to anesthesia (American
Electroencephalographic Society, 1994b; American Society
of Neurophysiologic monitoring, 2004; Burke et al., 1999;
Guerit et al., 1996), but is frequently noisy and sometimes
omitted (MacDonald and Janusz, 2002; MacDonald et al.,
2003; More et al., 1988; Mostegl et al., 1988; Nuwer, 1986,
1998). As no previous reports have based recommendations
on SNR properties, we have analyzed the intraoperative
SNR, reproducibility and rapidity of PF, optimized P37,
standard P37 and P31 potentials to form a sound basis for
recommending effective tibial SEP monitoring techniques.
2. Methods
2.1. Patients
The study included all 37 patients sequentially referred
for routine upper and lower limb SEP and transcranial
electric muscle motor evoked potential (MEP) monitoring
of scoliosis surgery during the study period. There were
11 males and 26 females (ages 4–22 years, median 14
years). All had clinically normal sensorimotor function
except for one patient with myopathy. Two patients had
horizontal gaze palsy and progressive scoliosis with
sensorimotor non-decussation shown by reversed lateraliza-
tion of intraoperative median and tibial cortical SEPs and
muscle MEPs (Jen et al., 2004; MacDonald et al., 2004a).
Horizontal gaze palsy was obvious in one, but subtle and
found postoperatively in the other. Both had a midline
ventral cleft of the medulla on magnetic resonance imaging
suggesting congenital absence of the sensorimotor decussa-
tions. Other diagnoses were idiopathic scoliosis (24),
neurofibromatosis (4), congenital scoliosis (2), Marfan’s
syndrome (1), achondroplasia (1), osteomalacia (1), and
diastamatomyelia (1). The surgeon obtained routine
informed consent for surgery with SEP/MEP monitoring.
2.2. Anesthesia
Anesthesia followed our established routine and was
adjusted to clinically determined surgical depth and
satisfactory blood pressure. Pre-positioning anesthesia was
either total intravenous anesthesia (TIVA) using propofol at
5–10 mg/kg/h and opioids in 20 patients or 0.5–2%
sevoflurane sometimes with nitrous oxide in 17. This choice
followed the preference of the anesthesiologist assigned to
the surgery. Post-positioning anesthesia for surgery was
TIVA in all patients and propofol occasionally reached
12 mg/kg/h. Neuromuscular blockade was omitted after
intubation. Propofol blood levels and bispectral index were
not performed and minimum alveolar concentration levels
were not specifically noted.
1859
2.3. General recording methods
The recording instrument was an Endeavor (Nicolet
Biomedical Instruments, Madison, WS, USA). Scalp and
neck electrodes were collodion-fixed gold-plated cups and PF
electrodes were adhesive silver-silver chloride discs. Impe-
dance was below 2 kO and recording leads were braided.
Interleaved tibial nerve stimuli were constant-current rec-
tangular pulses of 0.2 ms duration and fixed 4.7 Hz frequency
at supra-maximal intensity for single-sweep PF responses.
The SEP analysis time base was 100 ms.
2.4. P37 optimization
We routinely optimized P37 derivations to highest
amplitude for each side as previously reported (MacDonald
et al., 2004b). A referential recording of FPz, Cz, Pz, CP4,
CP2, CPz, CP1 and CP3-mastoid was used to identify the
P37 and N37 maximum sites for use as inputs 1 and 2. The
N37 was infrequently absent and then FPz was sometimes
optimal as input 2 instead. The Endeavor design allows
extending this reported method to simultaneously include
recording of all known potentially optimal derivations.
Thus, Cz, Pz, CPz, iCPi and CPi to both CPc and FPz as well
as Cz–Pz were also routinely recorded, where CPi and CPc
were CP3 or CP4 ipsilateral and contralateral to the
stimulated nerve and iCPi was CP1 or CP2, ipsilateral.
Ipsilateral and contralateral sites were switched in these
montages when referential recording showed non-decussa-
tion. The simultaneous bipolar recordings confirmed the
referentially inferred optimal derivations or resolved any
ambiguity. In one patient CPz–CPc was selected as optimal
bilaterally because of less noise than CPz–FPz that had
highest amplitude. The 5–10 min procedure took place after
induction. Fig. 1 illustrates the technique and Table 1 lists
the identified optimal derivations.
2.5. Other derivations and filtering
The PF was routinely recorded with a pair of electrodes
separated by 3 cm placed just above the popliteal fossa
crease. Standard P37 and P31 derivations were CPz–FPz
and FPz–C5S. The bandwidth was 30–500 Hz for scalp
potentials and 5–500 Hz for the PF (a 5 Hz high-pass filter
gives a more level baseline before PF onset than a 30 Hz
filter with our instrumentation). Notch filtering was not
used. Free-running EEG was displayed using optimized
P37, standard P37 and P31 derivations and filter settings.
2.6. SNR recordings
Bilateral PF, P37 and P31 recordings of 12 traces each
were obtained concurrently. Trace 1 contained ongoing
single sweeps and traces 2–12 were averages successively
doubling sweep number from 2 to 2048. A second trial was
superimposed. The automatic artifact rejection level was
1860 D.B. MacDonald et al. / Clinical Neurophysiology 116 (2005) 1858–1869

Fig. 1. P37 optimization. M, mastoid; PF, popliteal fossa. All traces for both sides are recorded simultaneously, but left and right examples from separate
patients are shown to emphasize the variety of possible results. The left and right P37 maximums are at Cz and Pz. The N37 maximum is contralateral to the
stimulated nerve at CP4 or CP3, which along with the ipsilateral P37 scalp field demonstrates normal decussation (reversed lateralization of these potentials
demonstrates non-decussation). The bipolar recordings confirm the referentially inferred Cz–CP4 and Pz–CP3 optimal derivations that have substantially
larger amplitude, and therefore, SNR than CPz–FPz. PF responses are recorded to ensure correct stimulus lateralization. Optimal derivations vary considerably
between individuals and sides due to variations of P37/N37 topography.

200 mV. Most recordings took place before positioning; a 2.8. Quantitative assessment
few were performed during closure or another non-critical
segment of surgery (e.g. surgeon break). They used our The PF peak-trough, P37–N45 or P31–N35 estimated
routine electrode set and general methods and did not potential amplitude was measured in each trace with a
modify or compromise monitoring. All recordings con- distinctly visible potential. Traces without a visible
tained PF, optimized P37 and P31 potentials and the last 21 potential were assigned a value of zero. True potential
recordings included the standard P37. The EEG pattern amplitude was defined as the mean estimated potential
was continuous during 27 recordings (15 TIVA and 12 amplitude of trials 1 and 2 measured in trace 12 (2048
sevoflurane) and burst-suppression during 10 (eight TIVA sweeps). Peak noise was measured in traces extended
and two sevoflurane). Of the 10 patients with recordings
during burst-suppression, two also had a comparative Table 1
recording during continuous EEG that was excluded from Optimal P37 derivations for the 74 tibial nerves of the 37 patients
statistical analysis.
Input 1 Input 2 (K down)
(C down) CPc CPia Pz FPzb
2.7. Visual assessment Cz 22 1 2 0
CPz 27 3 0 0
Raw noise was qualitatively compared in single sweeps Pz 7 0 – 3
iCPi 8 0 0 0
and free-running EEG. The sweep number at which each
CPi 1 – 0 0
SEP first became visibly distinct from noise was determined
in each trial. An experienced judgement of each trace’s CPc and CPi, CP4 or CP3 contralateral and ipsilateral to the stimulated
nerve; iCPi, CP2 or CP1 intermediate centroparietal sites, ipsilateral.
superimposability determined the sweep number at which a
Two patients had abnormally ipsilateral N37 potentials due to
sufficient visual reproducibility for reliable monitoring was non-decussation.
b
first established, thereby determining rapidity (Fig. 2). FPz was excluded in 96% of optimized derivations.
 D.B. MacDonald et al. / Clinical Neurophysiology 116 (2005) 1858–1869 1861

Fig. 2. SNR recording. PF, popliteal fossa. Display gain of each trace is adjusted to avoid overlap and produce similar final vertical deflections of each potential.
In this recording, PF, optimized P37, standard P37 and P31 potentials were first visible at 1, 2, 4, and 64 sweeps and reproducible at 2, 64, 256 and 1024 sweeps,
respectively.

to 200 ms by including a 100 ms pre-stimulus epoch.
The maximum and minimum noise peaks of the entire
trace were measured and noise was defined as the mean
maximum-minimum difference of trials 1 and 2. Raw noise
was measured in visually representative single sweeps
excluding any stimulus artifact or visible signal. Traces 2–
12 noise was measured in G averages to cancel out evoked
potentials, including any late components of the preceding
stimulus and those evoked by the interleaved contralateral
stimulus. The SNR of each trace was defined as true
potential amplitude/noise. Inter-trial SEP amplitude vari-
ation was calculated to quantify reproducibility in each
trace. It was defined as the estimated potential amplitude
difference expressed in percent of the larger amplitude of
the two trials; traces without a visible SEP in either trial
were assigned the maximum value of 100%. This definition
was relevant because warning criteria often consider the
percentage by which an SEP is smaller than a previous one
(Burke et al., 1999).
Statistical comparisons between potentials used paired
t-tests for naturally paired observations and unpaired t-tests
otherwise. Differences were considered significant when
P!0.05 on each side tested separately. Because all
variables were skewed, mean values were calculated after
zero skew transformations and then re-expressed in the
original units (see addendum). This was done for each
side and then summarized by the mean of right and left
derived values after demonstrating no inter-side difference
(PO0.05). Linear regression between variables was per-
formed using logarithmic transformations except between
trace SNR and amplitude variation for which zero skew
transformation was used. Pearson correlation coefficients
were expressed as r2 values.
3. Results
3.1. Visual assessment
PF traces exhibited by far the least noise consisting of
low voltage extraneous interference, although intermittent
EMG contaminated two recordings. Relatively low voltage
EEG and extraneous noise characterized optimized P37
traces. Standard P37 and P31 traces had obviously greater
EEG and extraneous noise and the P31 contained ECG and
also intermittent EMG in two recordings. Scalp derivations
1862 D.B. MacDonald et al. / Clinical Neurophysiology 116 (2005) 1858–1869
Fig. 3. Continuous (A) and burst-suppression (B) EEG anesthesia patterns
(30–500 Hz bandwidth). The bottom two traces in each panel are the right
and left optimized P37 derivations. With continuous EEG, optimized P37
derivations have less EEG noise than FPz derivations. With burst-
suppression, EEG noise is reduced in all derivations, but FPz–C5S still
contains ECG and greater extraneous noise.
had clearly less noise during burst-suppression and this was
most pronounced for the standard P37 and least for the P31
(Fig. 3).
Overall, the PF had by far the fewest sweeps to visibility
and to visual reproducibility, followed by the optimized
P37, standard P37 and finally P31 (Fig. 4). Median sweep
numbers to reproducibility were PF: 2, optimized P37: 128,
standard P37: 512 and P31: 1024. One standard P37 and five
P31 recordings were not reproducible at 2048 sweeps. Scalp
potentials required fewer sweeps during burst-suppression,
which improved rapidity most for the standard P37 and least
for the P31 (Fig. 5); median sweep numbers to reproduci-
bility fell to 64, 128 and 768 for the optimized P37, standard
P37 and P31.
Comparing sweeps to visual reproducibility in individual
recordings, the PF consistently required fewer sweeps than
the optimized P37 (median 32 times less). In turn the
optimized P37 required fewer or infrequently equal sweeps
compared to the standard P37 (median four times less) and
especially P31 (median eight times less). There was only
one exception of a more rapidly reproduced P31, but this
occurred during induction with sevoflurane and nitrous
oxide. Finally, the standard P37 usually required fewer or
equal sweeps compared to the P31 (median four times less),
but five exceptions of more rapidly reproduced P31
potentials occurred.
Fig. 4. Sweep number (N) to first visibility (A) and visual reproducibility
(B) expressed as a percentage of recordings. PF, popliteal fossa potential;
oP37, optimized P37; sP37, standard P37. At each averaging stage, the
percentage of recordings in which signals first appear or show visual
reproducibility is highest for the PF, followed by the optimized P37,
standard P37 and finally P31. The standard P37 and P31 were not
consistently reproducible at 2048 sweeps.
3.2. Quantitative assessment
3.2.1. Potential amplitude, noise and SNR
Mean values of true potential amplitude, raw noise and
raw SNR considering any EEG anesthesia pattern were
significantly different between the four potentials (P!0.01)
except for slightly lower standard P37 than P31 noise. These
values were most favorable for the PF followed by the
optimized P37, standard P37 and finally P31 (Table 2).
Amplitudes were not significantly different between con-
tinuous EEG and burst suppression, but scalp recordings
had lower noise and higher SNR during burst-suppression
(P!0.02), most noticeably for the standard P37 and least
for the P31. Amplitudes, noise and SNR showed no
significant difference between TIVA and sevoflurane, and
did not correlate to age.
 D.B. MacDonald et al. / Clinical Neurophysiology 116 (2005) 1858–1869 1863

Fig. 5. The effect of burst-suppression. N, averaged sweep number. Display gain is adjusted to minimize overlap and produce similar final vertical deflections of
each potential. During continuous EEG under sevoflurane, the P31 was not reproducible at 2048 sweeps and the optimized P37 (CPz–CP4) was reproducible at
256. During burst-suppression under propofol, the P31 became reproducible at 1024 sweeps and the optimized P37 became visible in single sweeps and
reproducible at only 16, or even 8. Note the larger non-EEG P31 noise content during burst-suppression and the higher optimized P37 amplitude with propofol.

Table 2
Mean true potential amplitude, raw noise and raw SNR

EEG Mean values

PF oP37 sP37 P31
Amplitude (mV) Any 4.4 2.7 2.0 1.1
Noise (mV) 2.7 19 31 37
SNR 1.6 0.15 0.07 0.03
Noise (mV) Continuous 21 35 41
Burst-suppression 10 14 24
SNR Continuous 0.13 0.06 0.03
Burst-suppression 0.25 0.14 0.04

PF, popliteal fossa; oP37, optimized P37; sP37, standard P37; SNR, signal to noise ratio.

There was a consistent order of raw SNR in individual stage. The rapidity to reach any given mean SNR level was
recordings with PF[optimized P37Ostandard P37OP31 markedly fastest for PF recordings and substantially faster
(Table 3). Sweep number to reproducibility had a powerful for optimized P37 than standard P37 and especially P31
negative non-linear correlation to raw SNR [NZ5.6(raw recordings.
SNR)K1.5, r2Z0.94] (Fig. 6). A similar negative correlation Each potential’s mean estimated amplitude decreased
to true potential amplitude (r2Z0.56) and stronger positive with averaging when calculated from the sweep number at
correlation to raw noise (r2Z0.80) were both weaker than
the correlation to raw SNR. Table 3
Mean ratios of true potential amplitude, raw noise and raw SNR in
pMean
ffiffiffiffi noise decreased with averaging by approximately
pffiffiffiffi individual recordings
1= N and mean SNR increased by approximately N (r2O
0.99) (Fig. 7A and B). Individual recordings showed Mean (range)
random positive and negative deviations from these curves. PF/oP37 oP37/sP37 oP37/P31 sP37/P31
The mean values remained significantly different between Amplitude 1.6 1.4 2.3 1.8
the four potentials at each averaging stage (P!0.01) except Noise 0.16 0.65 0.49 0.79
for slightly lower standard P37 than P31 noise. The SNR SNR 9.3 (1.9–115) 2.1 (1.1–6.0) 4.9 (1.0–18) 2.3 (1.1–15)
order and correlation between sweep number to reproduci- PF, popliteal fossa; oP37, optimized P37; sP37, standard P37; SNR, signal to
bility and SNR values were maintained at each averaging noise ratio.
1864 D.B. MacDonald et al. / Clinical Neurophysiology 116 (2005) 1858–1869

Fig. 6. The relationship between sweeps to visual reproducibility and raw

SNR. PF, popliteal fossa; N, averaged sweeps. To avoid overlap, the
optimized P37 and standard P37 markers have been shifted down and up by
10%. The line is the regression mean. The six imaginary data points at 4096
sweeps represent standard P37 and P31 recordings that were not
reproducible at 2048 sweeps and were excluded from the regression.

which all of its traces had visible signals. The decrement

with each doubling of sweep number was 0.6% for the PF
(r2Z0.77) and 3.1–3.3% for the scalp potentials (r2O0.86).

3.2.2. Potential amplitude variation

Mean amplitude variation of all potentials progressively
decreased with averaging (Fig. 7C). PF mean values were
markedly lowest at each averaging stage (P!0.001).
Optimized P37 mean amplitude variation was substantially
lower than standard P37 (P!0.05) and especially P31
(P!0.01) mean values below 1024 sweeps. The optimized
P37 mean had fallen to 12% at only 128 sweeps, while
standard P37 and P31 mean values reached this level at 1024
and 2048 sweeps. Standard P37 mean amplitude variation
was consistently below P31 mean values, but this was not
statistically significant after 16 sweeps due to substantial
variability of both potentials.
Considering all traces together as an SNR continuum,
there was a non-linear negative correlation between trace
SNR and amplitude variation (r2Z0.64, P!0.001) with
steep mean amplitude variation reduction occurring from
very low SNRs up to about one and a more gradual decline
thereafter (Fig. 8). The meanC1SD regression line fell to Fig. 7. Mean values of noise (A), signal to noise ratio (B) and potential
30 and 20% amplitude variation at SNRs of 1.8 and 4.0. amplitude variation (C) at each averaging stage. N, averaged sweep
number; PF, popliteal fossa, oP37, optimized P37; sP37, standard P37. The
Median sweep numbers to reach this range were PF: 2,
error bars are 95% confidence intervals for the mean in A and B and C1SD
optimized P37: 128 and standard P37: 1024, while only 36% in C. The dotted lines are regression curves through mean values. For
of P31 recordings reached this range at 2048 sweeps. the PF, optimized P37, standard P37 and P31, the slopes in A were NK0.48,
A 20% amplitude variation level approximately matched NK0.50, NK0.51, and NK0.51 (r2O0.99), the slopes in B were N0.47, N0.50,
the judgement of visual reproducibility. The relationship N0.49, and N0.51 (r2O0.99) and the slopes in C were NK0.18, NK0.31, NK0.35,
and NK0.39 (r2O0.92).
was imperfect because some traces by chance had !20%
 D.B. MacDonald et al. / Clinical Neurophysiology 116 (2005) 1858–1869
Fig. 8. The relationship between percentage amplitude variation and trace
signal to noise ratio (SNR). The solid and dashed curves are the regression
mean and meanC1SD.
variation when not visually reproducible and some had
O20% variation when judged reproducible. At visual
reproducibility, the scalp potentials’ meanC1SD amplitude
variation was 19–21% and maximum was 25–32% and there
was no significant difference between the three potentials,
while meanC1SD PF variation was lower at 14%
(P!0.01). The six non-reproducible standard P37 and
P31 recordings had an average amplitude variation of 41%
at 2048 sweeps.
4. Discussion
This study addressed SEP amplitude rather than latency
because amplitude is the primary intraoperative consider-
ation and its estimation is more affected by SNR than
latency. True potential amplitude and noise estimates were
possible sources of error. Noise and response jitter interfere
with the determination of true SEP amplitude (Nishida et al.,
1993; Vincent, 1992). The small mean estimated potential
amplitude reductions observed with averaging likely reflect
jitter and affected the physiologically complex scalp
potentials equally (about 3% with each doubling of sweep
number) and more than the simpler PF potential (0.6%).
Thus, mean trace 12 amplitude underestimated true
potential amplitude, but was fair for the three scalp
potentials and the best available estimate because noise
was least. This finding also suggests that averaging should
not extend too much beyond reproducibility to avoid further
jitter. Thus, higher SNR potentials that reproduce more
rapidly will have less jitter degradation during monitoring.
Noise is non-stationary and this challenges its estimation.
By inverting every second sweep, the G average cancels
evoked potentials while not affecting random noise. While pffiffiffiffi
not entirely accurate (Stecker,p2000),
ffiffiffiffi the observed 1= N
decrease of mean noise and N increase of mean SNR
agrees with general averaging theory and supports this
estimate. Individual recordings roughly follow these curves,
but with random positive and negative deviations because
noise fluctuates. Peak noise was easily measured
1865
and intuitively relevant. Root mean square noise measure-
ment (Stecker, 2000) was not available in our instrument
and would have reduced noise magnitudes, but probably not
relative differences. We widened the noise assessment
window to 200 ms and kept this the same in all traces for
consistency. Still, this was brief sampling for single sweeps
and might have caused raw noise underestimates that would
presumably diminish with averaging. However, pwe ffiffiffiffi
attempted to measure representative sweeps and the 1= N
conformity of mean noise at each N argues against such a
systematic error. Also, the strong relationship between raw
SNR and rapidity would have broken down had marked raw
noise errors been frequent, and the same relationship was
maintained using SNR values from each averaging stage.
Therefore, our raw noise estimates appear sufficiently valid.
Finally, noise estimates did not consider noise and signal
frequency content that may be relevant when substantially
different. However, reproducibility and rapidity assessments
were independent of this consideration and followed
consistent patterns with SNR, suggesting that this factor
did not play a critical role between the potentials we
examined.
Several alternative signal-processing techniques promise
faster signal estimation than averaging (e.g. Lam et al.,
2004). Our results will apply equally to these techniques
when they become clinically available because they must
also be enhanced by superior raw SNRs.
4.1. Reproducibility
Monitoring involves a series of noise-distorted esti-
mates that must be sufficiently reliable to identify true
change and avoid misinterpretation (Burke et al., 1999).
Demonstrating replication is the fundamental requirement
for reliability (American Electroencephalographic Society,
1994a). Visual reproducibility is our preferred method
because residual noise is considered by evaluating the
entire sweep’s closeness of fit. At least the components of
interest should superimpose ‘almost exactly’ (Nuwer et al.,
1994), although subjectivity is involved. Amplitude
variation compliments visual assessment by quantifying
amplitude replication, but disregards residual noise so that
any two estimates may be similarly inaccurate or
maximally dissimilar by chance according to SNR. This
explains the imperfect relationship observed between
amplitude variation and visual reproducibility. One IOM
guideline recommends !20–30% variation (Burke et al.,
1999), which agrees with the meanC1SD amplitude
variation of 19–21% and maximum of 25–32% observed
at our judgement of scalp potential reproducibility. That
there was no significant difference of these values between
the three scalp potentials suggests that reproducibility was
evenly judged and that differences of rapidity to visual
reproducibility were not due to bias. Lower mean PF
amplitude variation at reproducibility was due to frequent
high reproducibility in single sweeps.
1866 D.B. MacDonald et al. / Clinical Neurophysiology 116 (2005) 1858–1869
4.2. The importance of signal to noise ratio
Our results demonstrate that increasingly smaller ampli-
tude variation and sweep number to reproducibility occurs
towards higher SNRs. Also, higher raw SNR potentials are
much more likely to rapidly reach the 1.8–4.0 SNR range,
where meanC1SD amplitude variation falls to 30–20% and
the great majority of amplitude estimates will have
sufficient reliability. Furthermore, SNR correlates more
powerfully with rapidity than potential amplitude or noise
alone and of these two determinants, noise has the stronger
independent correlation. Thus, SNR has clinically important
non-linear positive relationships to reproducibility and
rapidity and IOM should select the highest SNR methods
emphasizing both higher amplitude and even more impor-
tantly lower noise. Lower SNR methods compromise
reproducibility and rapidity and should be rejected.
4.3. Anesthesia
We found no SNR differences with type of anesthesia,
but did not aim to evaluate this. Anesthesia was not formally
controlled and potentially confounding factors included
more frequent burst-suppression with TIVA, late timing of
some TIVA recordings and inconsistent nitrous oxide
supplementation of sevoflurane. We did not include
isoflurane that may be common in other settings. Previous
studies establishing that TIVA favors cortical response
SNR form a solid basis for this anesthetic choice during
monitoring (Boisseau et al., 2002; Kalkman et al., 1991a,b;
Langeron et al., 1999; Taniguchi et al., 1992).
Thirty Hertz low frequency filtering removes most EEG
slow activity, but our results show that fast activity remains
a major source of scalp SEP interference and contaminates
FPz more than centroparietal sites. Therefore, the use of FPz
reduces SNR under anesthesia and should generally be
avoided, unless shown to be optimal. This might also be true
for Fz because anesthetic fast activity is well known to be
frontal dominant.
Anesthetic EEG suppression increases scalp SEP SNR
and rapidity (Rytky et al., 1999) and the results show that
this is also true of burst-suppression that in our experience is
commonly encountered with TIVA. The noise reduction is
greatest for the standard P37 because of its large FPz EEG
noise content and least for the P31 because of its large
remaining non-EEG noise content. Thus, burst-suppression
may be desirable for tibial SEP monitoring if the associated
dosing is tolerated by the patient and does not excessively
depress concurrently monitored muscle MEPs. We had no
difficulty monitoring MEPs during TIVA burst-suppression,
which concurs with our previous experience (MacDonald
et al., 2003). Inhalational anesthesia may be more
detrimental (MacDonald et al., 2003), but we did not
attempt MEPs during sevoflurane.
Response and noise fluctuation might theoretically cause
instability during burst-suppression. In practice,
the optimized P37 quickly resolves with high stability (e.g.
Fig. 5) due to low and actually more stationary (suppression)
than non-stationary (burst) noise in this state. We do not think
that separating bursts from suppression is necessary, although
it would be possible to set rejection levels low enough to
average only during suppressed segments.
It must be emphasized that anesthesia was adjusted to
routine clinical parameters, so our experience does not
necessarily extend to burst-suppression titration that would
involve higher dosing than used for most of our patients. It
should also be noted that satisfactory rapidity is frequently
possible with continuous EEG and optimized P37 derivations.
4.4. The PF potential
The PF potential has the greatest SNR, reproducibility
and rapidity. Its low noise is due to the absence of EEG
and ECG as well as the close inter-electrode distance
minimizing extraneous interference. When EMG contami-
nation occasionally arises, we find that increasing analgesia
will suppress nociceptive reflex muscle activity. However,
EMG can mar PF recording in lumbo-sacral surgeries when
nerve root irritation occurs. PF monitoring provides
technical control and detects leg ischemia that can
confound aortic and orthopedic surgical monitoring
(MacDonald and Janusz, 2002; Vossler et al., 2000), and
single-sweep visibility helps determine supra-maximal
stimulus intensity. The reproducibility within 32 sweeps
wastes no time.
4.5. The optimized P37
The optimized P37 has the highest SNR of the scalp
potentials. The mean 1.4:1 amplitude advantage over
CPz–FPz confirms earlier work (MacDonald et al.,
2004b), and the mean 0.65:1 noise advantage is due to
FPz omission and close inter-electrode distance. Both
factors contribute to a mean 2.1:1 SNR superiority and
even greater 4.9:1 superiority over the P31. Consequently, it
provides substantially greater reproducibility and rapidity.
The median 128 sweeps to reproducibility requires only 27 s
and 86% resolve within 256 sweeps or 1 min, when stimulus
frequency is 4.7 Hz and there are no sweep rejections.
Optimization is achieved through electrode pairs opti-
mally placed to pick up the maximal response gradient and
minimize noise. In order of frequency, the P37 maximum
may be at CPz, Cz, Pz, iCPi, CPi or even iCPc or CPc in the
case of non-decussation. The N37 maximum is usually at
CPc, but may be absent or ectopic at Pz with a Cz P37
maximum, or at CPi in the case of non-decussation
(MacDonald et al., 2004a,b). This marked natural variability
should not be ignored because managing it with optimiz-
ation clearly produces superior results. The technique
becomes routine once mastered and is facilitated by modern
recording instruments such as the Endeavor, but was
developed with older 8-channel instruments.
 D.B. MacDonald et al. / Clinical Neurophysiology 116 (2005) 1858–1869
Fig. 9. Optimizing to highest SNR. CPz–FPz was optimal for P37 amplitude
(5.2 mV), had a raw SNR of 0.16 and was visually reproducible at 128
sweeps. Although CPz–CP3 had lower amplitude (3.6 mV), it also had
visibly less noise, a larger raw SNR of 0.21 and more rapid visual
reproducibility at 64 sweeps. CPz–CP3 was thus optimal and used for
monitoring.
While intended to maximize amplitude, the results
indicate that SNR should be optimized instead. Frequently
this will be the same derivation, but when FPz yields highest
amplitude, using another input 2, usually CPc may improve
SNR and rapidity due to lower noise (e.g. Fig. 9).
Significant advantages could be achieved more simply by
using the most commonly optimal CPz–CPc derivation that
will have lower noise and frequently higher amplitude than
CPz–FPz (MacDonald, 2001; MacDonald et al., 2004b;
Miura et al., 2003). However, even this derivation is
bilaterally optimal in only 17% of patients (MacDonald
et al., 2004b), so many additional SNR advantages will be
missed (e.g. Fig. 1). Also, decussation should be assessed
before relying on CPz–CPc because non-decussation may
not be that rare with scoliosis and requires CPz–CPi instead
(MacDonald et al., 2004a). Decussation assessment is
inherent to full optimization.
4.6. The standard P37
The results confirm that CPz–FPz amplitude is almost
always suboptimal because it misses the maximal response
1867
gradient (MacDonald et al., 2004b). Furthermore, the large
noise content from FPz and long inter-electrode distance is
unnecessary. Thus, CPz–FPz has inferior SNR, reproduci-
bility and rapidity. Despite its mean 2.3:1 SNR advantage
over the P31, it is sometimes more slowly resolved because
of marked variability, which partly explains why the P31
can compliment this derivation. The median 512 sweeps to
reproducibility requires 2 min, 21% require 1024–2048
sweeps or 4–7 min and the potential is sometimes not
practically reproducible. CPz–FPz should clearly not be
standard.
Similar considerations apply to some other rec-
ommended derivations. For example, iCPi–FPz (American
Electroencephalographic Society, 1994b) will be off the
maximal gradient more often than CPz–FPz and be as noisy.
CPi–CPc (Burke et al., 1999) will often have less noise and
include the N37 maximum, but will frequently miss the
maximal gradient because the P37 is least often maximal
at CPi. These derivations should only be used if shown to
be optimal.
4.7. The P31
The principle rationale for P31 monitoring is its
resistance to inhalational anesthetics compared to the
standard P37 (Bernard et al., 1996; Pathak et al., 1989;
Wolfe and Drummond, 1988). However, this is not
necessarily relevant to better P37 methods. Moreover, the
fundamental P31 problem is feedback delay, exemplified by
the reported inability to provide quick enough SEP/MEP
feedback during aortic surgery when it was included
compared to when it was not (MacDonald and Janusz,
2002). Our results confirm that this problem is due to its low
SNR. It has the lowest mean amplitude and largest noise
content arising from FPz, long inter-electrode distance,
ECG and sometimes EMG. While neuromuscular blockade
may help, EMG contaminated only two of our recordings
and MEPs are a contraindication. The median 1024 sweeps
to reproducibility consumes 4 min, 24% require 2048
sweeps or 7 min, and replication is not demonstrable for
7% of tibial nerves. Consequently, the P31 should no longer
be routinely monitored in the context of optimal P37
methods that should be used instead.
This leaves the problem of identifying anesthetic P37
alterations. In our experience, these correlate with dose
changes and/or evolve gradually, while pathologic decre-
ments are more abrupt. Furthermore, upper limb cortical
SEP controls have been shown to identify systemic P37
influences through parallel amplitude change and are also
rapid (MacDonald and Janusz, 2002; MacDonald et al.,
2003; Shahin et al., 1996). Thus, combining optimal P37
methods and upper limb SEPs makes the P31 unnecessary
and detrimental. This strategy also protects against
brachial plexus or arm peripheral nerve injury
(MacDonald et al., 2003; O’Brien et al., 1994; Schwartz
et al., 2000).
1868 D.B. MacDonald et al. / Clinical Neurophysiology 116 (2005) 1858–1869
The P31 may still be useful in selected cases when CPz–
FPz and/or unfavorable anesthesia are still being used, or
when the P37 is depressed by antecedent cerebral pathology
(DiCindio et al., 2003; Ecker et al., 1996), or absent in
infants (Guerit et al., 1997). Rapid reproducibility will often
be impossible in these circumstances for which the P31
should be kept as a fallback option. Thus, it is reasonable to
be prepared for P31 monitoring, but in our practice it is
never needed.
4.8. Rapidity
The importance of rapidity is not comprehensively
addressed in the literature, but the likelihood of successful
intervention is obviously greater when neurologic compro-
mise is detected quickly. Intramedullary surgery can cause
immediate trauma, but SEPs cannot be instantaneous.
Ischemia or compression are common injury mechanisms
that can be detected by muscle MEP loss within 2 min
(de Haan et al., 1996; Jacobs et al., 1999; MacDonald and
Janusz, 2002; MacDonald et al., 2003). Therefore the target
feedback interval for all monitored modalities combined
(e.g. upper and lower limb SEPs and MEPs) should ideally
be 2 min or better—without compromising reproducibility.
Our results confirm that this can be achieved in the great
majority of patients with optimized P37 and PF recordings
(MacDonald et al., 2003) and show that this cannot be
achieved in the majority with standard P37 or P31
recordings.
4.9. Conclusion and recommendations
Non-invasive tibial SEP monitoring should select the
highest and reject lower SNR methods because SNR has
clinically important powerful non-linear correlations to
reproducibility and rapidity. FPz derivations should gener-
ally be avoided because of greater EEG noise, and therefore,
lower SNR than centroparietal derivations. Burst-suppres-
sion may be desirable because of increased scalp potential
SNR and rapidity. Optimized P37 and PF recordings are
most rapidly reproducible due to superior SNRs, and are
therefore, best for monitoring. The most commonly optimal
CPz–CPc derivation is a simpler P37 compromise, but will
still be suboptimal in the majority and requires decussation
assessment before use. Standard P37 and P31 recordings
have inferior SNRs, reproducibility and rapidity and should
no longer be standard. The P31 may still be a useful fallback
potential in a few selected cases.
4.10. Addendum
The zero skew transform was yZ ðelnðxÞpx K 1=px Þ, where
y is the transformed value, x is the raw skewed value and Px
is a transform factor for {x} solved to zero skew of {y} to
within six decimal points. The inverse transform was
xZ elnð1Cpx yÞ=px .
Acknowledgements
The following neurophysiology technologists partici-
pated in the recordings: Mohamad Al-Enazi; Judy Barclay,
RET; William Gene, RET, REPT; Brent Hedgecock, RET,
CNIM and Betty Jarvis.
References
American Electroencephalographic Society. Guideline nine: guidelines on
evoked potentials. J Clin Neurophysiol 1994a;11:40–73.
American Electroencephalographic Society. Guideline eleven: guidelines
for intraoperative monitoring of sensory evoked potentials. J Clin
Neurophysiol 1994b;11(1):77–87.
American Society of Neurophysiologic monitoring. Guidelines for
intraoperative monitoring using somatosensory evoked potentials.
Draft version 4.0 [American Society of Neurophysiologic Monitoring
web site]. Available at: http://www.asnm.org/. Accessed May 22, 2004.
Bernard JM, Pereon Y, Fayet G, Guiheneuc P. Effects of isoflurane and
desflurane on neurogenic motor-and somatosensory-evoked potential
monitoring for scoliosis surgery. Anesthesiology 1996;85(5):1013–9.
Boisseau N, Madany M, Staccini P, Armando G, Martin F, Grimaud D,
Raucoules-Aime M. Comparison of the effects of sevoflurane and
propofol on cortical somatosensory evoked potentials. Br J Anaesth
2002;88(6):785–9.
Burke D, Nuwer MR, Daube J, Fischer C, Schramm J, Yingling CD,
Jones SJ. Intraoperative monitoring. The international federation of
clinical neurophysiology. Electroencephalogr Clin Neurophysiol Suppl
1999;52:133–48.
de Haan P, Kalkman CJ, Ubags LH, Jacobs MJ, Drummond JC. A
comparison of the sensitivity of epidural and myogenic transcranial
motor-evoked responses in the detection of acute spinal cord ischemia
in the rabbit. Anesth Analg 1996;83(5):1022–7.
DiCindio S, Theroux M, Shah S, Miller F, Dabney K, Brislin RP,
Schwartz D. Multimodality monitoring of transcranial electric motor
and somatosensory-evoked potentials during surgical correction of
spinal deformity in patients with cerebral palsy and other neuromus-
cular disorders. Spine 2003;28(16):1851–6.
Ecker ML, Dormans JP, Schwartz DM, Drummond DS, Bulman WA.
Efficacy of spinal cord monitoring in scoliosis surgery in patients with
cerebral palsy. J Spinal Disord 1996;9(2):159–64.
Guerit JM, Verhelst R, Rubay J, Khoury G, Matta A, Dion R. Multilevel
somatosensory evoked potentials (SEPs) for spinal cord monitoring in
descending thoracic and thoraco-abdominal aortic surgery. Eur
J Cardiothorac Surg 1996;10(2):93–103.
Guerit JM, Witdoeckt C, Rubay J, Matta A, Dion R. The usefulness of the
spinal and subcortical components of the posterior tibial nerve SEPs for
spinal cord monitoring during aortic coarctation repair. Electroence-
phalogr Clin Neurophysiol 1997;104(2):115–21.
Jacobs MJ, Meylaerts SA, de Haan P, de Mol BA, Kalkman CJ. Strategies
to prevent neurologic deficit based on motor-evoked potentials in type I
and II thoracoabdominal aortic aneurysm repair. J Vasc Surg 1999;
29(1):48–57.
Jen JC, Chan W-M, Bosley TM, Wan J, Carr JR, Rb U, Shattuck D,
Salamon G, Kudo L, Ou J, Lin DDM, Salih MAM, Kansu T, Dhalaan H,
Al Zayed Z, MacDonald DB, Stigsby B, Plaitakis A, Dretakis EK,
Gottlob I, Pieh C, Traboulsi EI, Wang Q, Wang L, Andrews C,
Yamada K, Demer JL, Karim S, Alger J, Geschwind DH, Deller T,
Sicotte NL, Nelson SF, Baloh RW, Engle EC. Mutations in a human
Robo gene disrupt hindbrain axon pathway crossing and morphogen-
esis. Science 2004;304:1509–13.
Kalkman CJ, ten Brink SA, Been HD, Bovill JG. Variability
of somatosensory cortical evoked potentials during spinal surgery.
 D.B. MacDonald et al. / Clinical Neurophysiology 116 (2005) 1858–1869
Effects of anesthetic technique and high-pass digital filtering. Spine
1991a;16(8):924–9.
Kalkman CJ, Traast H, Zuurmond WW, Bovill JG. Differential effects of
propofol and nitrous oxide on posterior tibial nerve somatosensory
cortical evoked potentials during alfentanil anaesthesia. Br J Anaesth
1991b;66(4):483–9.
Lam BSC, Hu Y, Lu WW, Luk KDK. Validation of an adaptive signal
enhancer in intraoperative somatosensory evoked potentials monitor-
ing. J Clin Neurophysiol 2004;21:409–16.
Langeron O, Vivien B, Paqueron X, Saillant G, Riou B, Coriat P, Lille F.
Effects of propofol, propofol-nitrous oxide and midazolam on cortical
somatosensory evoked potentials during sufentanil anaesthesia for
major spinal surgery. Br J Anaesth 1999;82(3):340–5.
MacDonald DB. Individually optimizing posterior tibial somatosensory
evoked potential P37 scalp derivations for intraoperative monitoring.
J Clin Neurophysiol 2001;18(4):364–71.
MacDonald DB, Janusz M. An approach to intraoperative monitoring of
thoracoabdominal aneurysm surgery. J Clin Neurophysiol 2002;19(1):
43–54.
MacDonald DB, Al Zayed Z, Khoudeir I, Stigsby B. Monitoring scoliosis
surgery with combined multiple pulse transcranial electric motor and
cortical somatosensory-evoked potentials from the lower and upper
extremities. Spine 2003;28(2):194–203.
MacDonald DB, Streletz L, Al-Zayed Z, Abdool S, Stigsby B. Intraopera-
tive neurophysiologic discovery of uncrossed sensory and motor
pathways in a patient with horizontal gaze palsy and scoliosis. Clin
Neurophysiol 2004a;115(3):576–82.
MacDonald DB, Stigsby B, Al-Zayed Z. A comparison between derivation
optimization and Cz 0 -FPz for posterior tibial P37 somatosensory
evoked potential intraoperative monitoring. Clin Neurophysiol 2004b;
115:1925–30.
Miura T, Sonoo M, Shimizu T. Establishment of standard values for the
latency, interval and amplitude parameters of tibial nerve somatosen-
sory evoked potentials (SEPs). Clin Neurophysiol 2003;114(7):
1367–78.
More RC, Nuwer MR, Dawson EG. Cortical evoked potential monitoring
during spinal surgery: sensitivity, specificity, reliability, and criteria for
alarm. J Spinal Disord 1988;1(1):75–80.
Mostegl A, Bauer R, Eichenauer M. Intraoperative somatosensory potential
monitoring. A clinical analysis of 127 surgical procedures. Spine 1988;
13(4):396–400.
Nishida S, Nakamura M, Shibasaki H. Method for single-trial recording of
somatosensory evoked potentials. J Biomed Eng 1993;15(3):257–62.
1869
Nuwer MR. Spinal cord monitoring. In: Nuwer MR, editor. Evoked
Potential Monitoring in the Operating Room. New York: Raven Press;
1986. p. 49–101.
Nuwer MR. Spinal cord monitoring with somatosensory techniques. J Clin
Neurophysiol 1998;15:183–93.
Nuwer MR, Aminoff M, Desmedt J, Eisen AA, Goodin D, Matsuoka S,
Mauguiere F, Shibasaki H, Sutherling W, Vibert JF. IFCN rec-
ommended standards for short latency somatosensory evoked poten-
tials. Report of an IFCN committee. Electroencephalogr Clin
Neurophysiol 1994;91(1):6–11.
O’Brien MF, Lenke LG, Bridwell KH, Padberg A, Stokes M. Evoked
potential monitoring of the upper extremities during thoracic and
lumbar spinal deformity surgery: a prospective study. J Spinal Disord
1994;7(4):277–84.
Pathak KS, Amaddio MD, Scoles PV, Shaffer JW, Mackay W. Effects
of halothane, enflurane, and isoflurane in nitrous oxide on multilevel
somatosensory evoked potentials. Anesthesiology 1989;70(2):
207–12.
Rytky S, Huotari AM, Alahuhta S, Remes R, Suominen K, Jantti V. Tibial
nerve somatosensory evoked potentials during EEG suppression in
sevoflurane anaesthesia. Clin Neurophysiol 1999;110(9):1655–8.
Schwartz DM, Drummond DS, Hahn M, Ecker ML, Dormans JP.
Prevention of positional brachial plexopathy during surgical correction
of scoliosis. J Spinal Disord 2000;13(2):178–82.
Shahin GM, Hamerlijnck RP, Schepens MA, Ter Beek HT, Vermeulen FE,
Boezeman EH. Upper and lower extremity somatosensory evoked
potential recording during surgery for aneurysms of the descending
thoracic aorta. Eur J Cardiothorac Surg 1996;10(5):299–304.
Stecker MM. Generalized averaging and noise levels in evoked responses.
Comput Biol Med 2000;30(5):247–65.
Taniguchi M, Nadstawek J, Pechstein U, Schramm J. Total intravenous
anesthesia for improvement of intraoperative monitoring of somato-
sensory evoked potentials during aneurysm surgery. Neurosurgery
1992;31(5):891–7.
Vincent A. Methods for improving the signal-to-noise ratio of endogenous-
evoked potentials. Integr Physiol Behav Sci 1992;27(1):54–65.
Vossler DG, Stonecipher T, Millen MD. Femoral artery ischemia during
spinal scoliosis surgery detected by posterior tibial nerve somatosen-
sory-evoked potential monitoring. Spine 2000;25(11):1457–9.
Wolfe DE, Drummond JC. Differential effects of isoflurane/nitrous oxide
on posterior tibial somatosensory evoked responses of cortical and
subcortical origin. Anesth Analg 1988;67(9):852–9.