A spatio-temporal detective quantum efficiency and its application to fluoroscopic

systems
S. N. Friedman and I. A. Cunningham

Citation: Medical Physics 37, 6061 (2010); doi: 10.1118/1.3495968

View online: http://dx.doi.org/10.1118/1.3495968
View Table of Contents: http://scitation.aip.org/content/aapm/journal/medphys/37/11?ver=pdfcov
Published by the American Association of Physicists in Medicine

Articles you may be interested in

Cascaded-systems analyses and the detective quantum efficiency of single-Z x-ray detectors including
photoelectric, coherent and incoherent interactions
Med. Phys. 40, 041916 (2013); 10.1118/1.4794495

A theoretical and experimental evaluation of the microangiographic fluoroscope: A high-resolution region-of-

interest x-ray imager
Med. Phys. 38, 4112 (2011); 10.1118/1.3599751

A moving slanted-edge method to measure the temporal modulation transfer function of fluoroscopic systems
Med. Phys. 35, 2473 (2008); 10.1118/1.2919724

Solid-state fluoroscopic imager for high-resolution angiography: Physical characteristics of an 8 cm×8 cm

experimental prototype
Med. Phys. 31, 1462 (2004); 10.1118/1.1750992

The detective quantum efficiency of medical x-ray image intensifiers

Rev. Sci. Instrum. 69, 3953 (1998); 10.1063/1.1149205
A spatio-temporal detective quantum efficiency and its application
to fluoroscopic systems
S. N. Friedmana兲
Sackler School of Medicine, Faculty of Medicine, Tel Aviv University, Ramat Aviv 69978, Israel
I. A. Cunninghamb兲
Imaging Research Laboratories, Robarts Research Institute and Lawson Health Research Institute,
100 Perth Drive, London, Ontario N6A 5K8, Canada; and Department of Medical Biophysics,
The University of Western Ontario, 1151 Richmond Street, London, Ontario N6A 5B8, Canada
共Received 1 April 2010; revised 11 September 2010; accepted for publication 13 September 2010;
published 29 October 2010兲
Purpose: Fluoroscopic x-ray imaging systems are used extensively in spatio-temporal detection
tasks and require a spatio-temporal description of system performance. No accepted metric exists
that describes spatio-temporal fluoroscopic performance. The detective quantum efficiency 共DQE兲
is a metric widely used in radiography to quantify system performance and as a surrogate measure
of patient “dose efficiency.” It has been applied previously to fluoroscopic systems with the intro-
duction of a temporal correction factor. However, the use of a temporally-corrected DQE does not
provide system temporal information and it is only valid under specific conditions, many of which
are not likely to be satisfied by suboptimal systems. The authors propose a spatio-temporal DQE
that describes performance in both space and time and is applicable to all spatio-temporal quantum-
based imaging systems.
Methods: The authors define a spatio-temporal DQE 共two spatial-frequency axes and one temporal-
frequency axis兲 in terms of a small-signal spatio-temporal modulation transfer function 共MTF兲 and
spatio-temporal noise power spectrum 共NPS兲. Measurements were made on an x-ray image
intensifier-based bench-top system using continuous fluoroscopy with an RQA-5 beam at
3.9 ␮R/frame and hardened 50 kVp beam 共0.8 mm Cu filtration added兲 at 1.9 ␮R/frame.
Results: A zero-frequency DQE value of 0.64 was measured under both conditions. Nonideal
performance was noted at both larger spatial and temporal frequencies; DQE values decreased by
⬃50% at the cutoff temporal frequency of 15 Hz.
Conclusions: The spatio-temporal DQE enables measurements of decreased temporal system per-
formance at larger temporal frequencies analogous to previous measurements of decreased 共spatial兲
performance. This marks the first time that system performance and dose efficiency in both space
and time have been measured on a fluoroscopic system using DQE and is the first step toward the
generalized use of DQE on clinical fluoroscopic systems. © 2010 American Association of Physi-
cists in Medicine. 关DOI: 10.1118/1.3495968兴

Key words: detective quantum efficiency, DQE, spatio-temporal, spatiotemporal, fluoroscopy,

x-ray imaging, performance, dose efficiency, noise power spectrum, NPS, modulation transfer
function, MTF, temporal MTF, lag, small signal

I. INTRODUCTION not change with image position. It is defined in terms of the

Fluoroscopic systems provide “real-time” imaging of inter-
nal patient anatomy and are commonly used for interven-
tional procedures. Typical skin dose rates during fluoroscopic
imaging can be 20–50 mGy/min 共more for larger patients兲
and some procedures can last 60–90 min resulting in large
patient doses.1 For all procedures, regardless of dose, it is
crucial that the highest possible image quality be obtained
for a given patient dose to minimize risks from radiation
exposure.
The detective quantum efficiency2–5 共DQE兲 is a metric
that has gained widespread use in radiography to assess sys-
tem performance and as a surrogate measure of “dose effi-
ciency.” The DQE is applicable to linear 共or linearizable兲 and
shift invariant 共LSI兲 systems with wide-sense stationary
noise processes,6 in which the mean and autocovariance do
modulation transfer function 共MTF兲, describing spatial reso-
lution, and the Wiener noise power spectrum 共NPS兲, describ-
ing image noise5
d̄ 2MTF2共u, v兲
DQE共u, v兲 = , 共1兲
q̄NPS共u, v兲
where u and v are spatial-frequency variables corresponding
to the spatial variables x and y, d̄ is the average dark-
subtracted flat-field pixel value, and q̄ is the average input
x-ray density with units quanta per mm2.
The DQE as expressed by Eq. 共1兲 cannot be applied di-
rectly to fluoroscopic systems due to temporal consider-
ations. As early as 1974, Swank7 described the noise-
reducing effects of system lag that causes a sharing of image

6061 Med. Phys. 37 „11…, November 2010 0094-2405/2010/37„11…/6061/9/$30.00 © 2010 Am. Assoc. Phys. Med. 6061
6062 S. N. Friedman and I. A. Cunningham: Spatio-temporal DQE and its application to fluoroscopic systems
signal between subsequent fluoroscopic images 共called
frames兲 and therefore reduced NPS and inflated DQE values.
Several investigators have studied these effects and intro-
duced lag-correction factors using both temporal8–10 and
temporal-frequency11–13 domain descriptions. This has led to
the definition of lag-corrected metrics,8–11,14,15 including a
lag-corrected DQE, DQE␤共u , v兲, applicable to fluoroscopic
systems11,15
d̄ 2MTF2共u, v兲 d̄ 2MTF2共u, v兲
DQE␤共u, v兲 = = ␤,
to¯q̇NPS␤共u, v兲 to¯q̇NPS共u, v兲
where to is the time between subsequent images 共1/frame
rate兲, and ¯q̇ is the average x-ray density rate with units
quanta per mm2 s. The lag-corrected NPS, NPS␤, is obtained
by dividing the 共spatial兲 NPS measured from fluoroscopic
frames by the temporal correction factor ␤, where13
␤ = at 冕
−⬁
⬁
MTF2t 共␯兲d␯
and where at is the frame integration time, MTFt is the tem-
poral MTF, and ␯ is the temporal-frequency variable corre-
sponding to the temporal variable t. It should be noted that
0 ⱕ at ⱕ to, where the difference between at and to can be due
to the time required for the readout of the integrated signal
and formation of the corresponding frame.
However, there are several reasons why these lag-
corrected metrics are expected to be inadequate to assess
fluoroscopic systems. For example, to date, lag-corrected
metrics8–11,14,15 have required an assumption of deterministic
lag, implying that the system will consistently produce iden-
tical temporal outputs given identical inputs, characterized in
terms of a linear temporal filter applied to input signals. This
is not the case for quantum-based lag, such as might occur in
a phosphor, where lag results from a random delay in the
emission or release of individual optical quanta. This random
process is instead characterized in terms of a probability den-
sity distribution describing the delayed emission.
The implications of quantum-based lag were explored
theoretically by Akbarpour et al.13 They showed that the ef-
fect of quantum lag on image noise in a single fluoroscopic
frame may not be as simple as a scaling of all noise compo-
nents by a linear filter. For example, in their simple cascaded
model of a fluoroscopic system consisting of a phosphor op-
tically coupled to an active matrix sensor, lag only affects the
correlated 共i.e., spatial-frequency-dependent兲 component of
image noise. They also showed that lag can introduce uncor-
related noise if the conversion gain to the secondary quanta
共such as light in a phosphor兲 is not sufficiently large. This
result was unexpected and implies that lag can introduce a
secondary quantum sink similar to that caused by the spatial
scatter of secondary image quanta. An assumption of deter-
ministic lag may also be inappropriate in the presence of
additive noise, or noise aliasing in either space 共affected by
detector-element width and spacing兲, or time 共affected by
frame-integration time and rate兲. The implications of these
considerations may be substantial; lag-corrected metrics may
Medical Physics, Vol. 37, No. 11, November 2010
6062
not be a good representation of fluoroscopic systems with
suboptimal performance and it is exactly these systems for
which a DQE metric may be most important. To accurately
assess these systems and enable quantification of the errors
introduced by any inappropriate use of lag-corrected metrics,
the introduction of a new metric not limited by assumptions
such as deterministic lag is required.
The utility of lag-corrected metrics is further limited by
the spatial-only description of performance. Fluoroscopic
systems produce a time sequence of images that are used in
共2兲
spatio-temporal detection tasks. It seems necessary, there-
fore, that the DQE be extended to include signal and noise
performance in the temporal domain to describe the perfor-
mance of fluoroscopic systems. In the future, this may lead
to the use of a spatio-temporal noise-equivalent quanta
共NEQ兲 that might describe time-dependent detection tasks,
similar to the use of NEQ for describing detection tasks in
spatial images.16 Spatio-temporal NEQ might build on the
共3兲
work of Wilson and colleagues17–24 who developed a Fourier
model to predict observer performance in spatio-temporal
tasks.
We propose a spatio-temporal DQE, defined in terms of a
small-signal spatio-temporal MTF 共Ref. 15兲 and a spatio-
temporal NPS,12 that describes system performance and dose
efficiency in both space and time. A small-signal approach is
used for the temporal MTF to accommodate exposure-rate-
dependent lag in some systems.15 This metric does not re-
quire assumptions of deterministic-like behavior of lag and is
applicable to fluoroscopy and other LSI imaging systems that
exhibit correlations in both space and time. We describe the
theoretical development of a spatio-temporal DQE in this
article and illustrate its use to evaluate a bench-top fluoro-
scopic system operating in continuous 共nonpulsed兲 mode.
Lag-corrected DQE measurements are acquired under the
same conditions and compared to the spatio-temporal DQE
results.
II. THEORY
We define a three-dimensional 共3D兲 spatio-temporal DQE
based on previously established definitions of 共small-signal兲
spatio-temporal modulation transfer function15 and spatio-
temporal noise power spectrum.12 It is expressed as11
d̄ 2MTF⌬2 共u, v, ␯兩q̇
¯兲
DQE共u, v, ␯兲 = . 共4兲
¯q̇NPS共u, v, ␯兲
It is assumed that the spatio-temporal MTF is separable 共dis-
cussed in Sec. V兲 and can be expressed as the product of
spatial and 共exposure-rate-dependent small-signal兲 temporal
components, MTF共u , v兲 and MTF⌬共␯ 兩 ¯q̇兲, respectively, giving
MTF⌬共u , v , ␯ 兩 ¯q̇兲 = MTF共u , v兲MTF⌬共␯ 兩 ¯q̇兲.15
III. METHODS
Measurements of the spatio-temporal DQE were deter-
mined using a bench-top laboratory fluoroscopic system op-
erated in continuous 共nonpulsed兲 mode, consisting of a
7-year-old Thales x-ray image intensifier 共model TH9432
6063 S. N. Friedman and I. A. Cunningham: Spatio-temporal DQE and its application to fluoroscopic systems
QXH304VR, P-20 output phosphor with a thickness between
4 and 8 ␮m according to the manufacturer25兲, and a Pros-
ilica digital CCD camera 共model EC1350兲 operated in frame-
transfer mode26 at 29.97 frames per second with a frame
integration time of at = 32.0 ms. Measurements were made
using an RQA-5 input x-ray beam 共70 kVp, 7.1 mm Al half-
value layer, 21 mm Al additional filtration14兲 at
3.9 ␮R/frame, as well as a 50 kVp beam with 0.8 mm added
Cu filtration 共4.6 mm Al half-value layer, hereafter referred
to as a hardened 50 kVp beam兲 at 1.9 ␮R/frame. The image-
plane pixel size was 0.28 mm⫻ 0.28 mm and all images
were dark-field subtracted and corrected for geometric
distortions.27 While a 3D analysis in x, y, and t was per-
formed, only 共y , t兲-direction results are displayed for simplic-
ity and are representative of 共x , t兲-direction results.
III.A. Spatio-temporal modulation transfer function
The 共small-signal兲 spatio-temporal MTF was calculated
assuming separability 共previously confirmed on the test
system15,28兲. The slanted-edge29,30 and semitransparent mov-
ing slanted-edge15,28 methods were used to calculate the spa-
tial and temporal components where both stationary and
moving edges are used to assess spatial and 共velocity-
dependent兲 spatio-temporal blurring, respectively. The spa-
tial transfer function was measured using a stationary 3 mm
radiopaque tungsten sheet. The small-signal temporal MTF
was measured using a 0.381 mm semitransparent copper
edge using the method described previously,15 which re-
quires images of both a stationary and moving 共image-plane
velocity vo = 37⫾ 1 cm/ s兲 edge 共discussed in Sec. V兲. The
temporal component of the spatio-temporal measurement
was determined by removing spatial contributions. All metal
sheets used to form the edges were precision machined,
mounted in a custom-made assembly, placed against the de-
tector cover, and moved under servomotor control. All mea-
surements were made with the edges approximately centered
in the region of interest 共ROI兲. The oversampled edge pro-
files were differentiated to determine the line-spread func-
tion. After the application of a Hann window,31 the fast Fou-
rier transform 共FFT兲 was used to determine the MTF
component.28 Open-field normalization was used to obtain
accurate MTF values32 with the zero-spatio-temporal-
frequency value set to unity 共discussed in Sec. V兲.
III.B. Spatio-temporal noise power spectrum
A 3D spatio-temporal approach was taken to the NPS cal-
culation, as described elsewhere.12 Open-field spatio-
temporal ROIs, 36 mm⫻ 36 mm⫻ 2.1 s corresponding to
128 pixels⫻ 128 pixels⫻ 64 frames, were formed using
100 four-second exposures. Only frames collected after
steady-state exposure was achieved 共⬃1.5 s after x-ray turn
on兲 were used. Each 4 s exposure was used to produce 25 3D
ROIs with 50% overlap in space and no overlap in time.
Spatial detrending was achieved by subtracting the differ-
ences between a given frame pixel value and the correspond-
ing average pixel value across all analyzed frames. Temporal
detrending of the frame pixel value was achieved by sub-
Medical Physics, Vol. 37, No. 11, November 2010
6063
tracting the difference between the average pixel value in the
frame and the average pixel value across all analyzed frames.
A Hann window was not applied to the data prior to calcu-
lating the 3D FFT.
III.C. Spatio-temporal detective quantum efficiency
The flat-field dark-subtracted average pixel value from
image data d̄ was measured and the average density rate of
incident x rays ¯q̇, corresponding to the open-field images
above, was calculated from the measured exposure using the
conversion of 30 174 quanta/ mm2 per unit air kerma for the
RQA-5 spectrum.33 Equation 共4兲 and the spatio-temporal
MTF and NPS measured above were used to calculate the
spatio-temporal DQE.
III.D. Lag-corrected detective quantum efficiency
The same image data were used to calculate the lag-
corrected DQE using Eq. 共2兲, where to = 33.3 ms, and ␤ was
calculated by numerically integrating the temporal MTF be-
tween ⫾100 Hz.15,28
IV. RESULTS
IV.A. Spatio-temporal modulation transfer function
The spatio-temporal MTF for the RQA-5 beam in the
共y , t兲-frequency directions is shown in Fig. 1共a兲. Zero-
frequency values 共both spatial and temporal兲 were set to
unity without scaling any other value 共discussed further in
Sec. V兲. This accurately displays the MTF and the system
spatial low-frequency drop 共LFD兲 of ⬃12%, shown in Fig.
1共b兲. The spatial MTF in the y-frequency direction decays to
0.1 at ⬃1.3 mm−1. The distinctive sinc-like shape is noted in
the t-frequency direction, displayed in Fig. 1共c兲, with zero
values at integer multiples of ⬃31.25 Hz 共multiples of 1 / at兲.
Similar results were obtained for the hardened 50 kVp beam.
IV.B. Spatio-temporal noise power spectrum
The spatio-temporal NPS in the 共y , t兲-frequency directions
for the RQA-5 beam is shown in Fig. 2共a兲. The y-component
is displayed in Fig. 2共b兲 and decays to ⬃3% of the peak
value by 1.8 mm−1. This suggests that uncorrelated and ad-
ditive noise components are negligible. In contrast, the
t-component, displayed in Fig. 2共c兲, displays only a weak
frequency dependence. It is closely approximated by the
aliased temporal MTF 共calculated using aliases centered at
⫾30, ⫾60, and ⫾90 Hz, see discussion in Sec. V兲, and this
suggests that temporal uncorrelated and additive noise are
also negligible. Similar results were obtained for the hard-
ened 50 kVp beam.
IV.C. Detective quantum efficiency
The spatio-temporal DQE in the 共y , t兲-frequency direc-
tions for the RQA-5 beam is shown in Fig. 3共a兲 and the
y-frequency component along with the lag-corrected DQE
values are plotted in Fig. 3共b兲. Both show a zero-frequency
value of 0.64 with a substantial low-frequency drop 共dis-
6064 S. N. Friedman and I. A. Cunningham: Spatio-temporal DQE and its application to fluoroscopic systems 6064

modulation transfer function

0.8

0.6

0.4

0.2

0
0 0.5 1 1.5 2 2.5
spatial frequency (mm−1)

(a) RQA-5, 3.9 µR/frame (b) y-frequency direction

1
modulation transfer function

0.8

0.6

0.4

0.2

0
0 20 40 60 80 100
temporal frequency (Hz)

(c) t-frequency direction

FIG. 1. The measured spatio-temporal MTF for the RQA-5 beam at 3.9 ␮R/frame is displayed in 共a兲, and values along each of the axis are displayed in 共b兲
and 共c兲 corresponding to the system spatial and temporal MTFs, respectively.

cussed in Sec. V兲. The lag-corrected DQE is slightly less
共⬃10%兲 than the spatio-temporal DQE 共discussed in Sec.
V兲.
The t-component values are shown in Fig. 3共c兲 and start
decreasing at ⬃6 Hz. There is a ⬃50% decrease to a value
of ⬃0.3 at 15 Hz. These results could not be obtained using
a lag-corrected approach and show nonideal system temporal
behavior.
Similar DQE plots for the hardened 50 kVp beam are
shown in Fig. 4 with low-frequency values of 0.64. The lag-
corrected DQE is also ⬃10% smaller than the spatio-
temporal DQE, and a ⬃50% decrease in the
t-frequency-component values by 15 Hz is noted.
V. DISCUSSION
The detective quantum efficiency values of the bench-top
XRII-based fluoroscopic system were presented in this ar-
ticle to illustrate the spatio-temporal DQE. Measurements of
the lag-corrected DQE were also provided for comparison
and to give some insight to the relationship between the two
metrics.
V.A. Spatio-temporal modulation transfer function
The assumption of spatio-temporal separability in the
MTF was required for the moving slanted-edge method15,28
and remains a requirement in the present study. It was con-
firmed for the test system previously15,28 and is known to be
a good assumption for P-20 phosphors.34 Separability may
be a good assumption for many systems as different physical
processes are responsible for spatial blur and temporal lag.
MTF values along the x and y axes were measured using
perpendicular edges,29,30 and values in the 共x , t兲 and
共y , t兲-frequency planes were calculated.
The spatial MTF component was calculated using a radio-
paque edge to minimize the influence of scattered radiation

Medical Physics, Vol. 37, No. 11, November 2010

6065 S. N. Friedman and I. A. Cunningham: Spatio-temporal DQE and its application to fluoroscopic systems 6065

0.4

noise power spectrum (mm2s)

0.35

0.3
0.4
0.25
noise power spectrum

0.3
0.2
(mm s)

0.2
0.15
2

0.1
0.1
0
0
0
0.05
5 0.5
10 1
15 1.5 0
spatial frequency 0 0.5 1 1.5
temporal frequency
(Hz) (mm−1) spatial frequency (mm−1)

(a) RQA-5, 3.9 µR/frame (b) y-frequency direction

0.4
NPSt
noise power spectrum (mm2s)

0.35 MTFt2 (aliased)

MTFt2
0.3

0.25

0.2

0.15

0.1

0.05

0
0 2 4 6 8 10 12 14
temporal frequency (Hz)

(c) t-frequency direction

FIG. 2. The measured spatio-temporal NPS for the RQA-5 beam at 3.9 ␮R/frame is displayed in 共a兲, and values along each of the axis are displayed in 共b兲
and 共c兲 corresponding to the system spatial and temporal NPSs, respectively. The temporal MTF and aliased temporal MTF are also displayed in 共c兲.

on the measurement.35 Calculation of the temporal compo- by the zero-frequency value would have resulted in a
nent using the moving slanted-edge method requires prior 共1 – 0.88兲 / 0.88⬇ 14% inflation,32 obscuring most of the low-
knowledge of the spatial MTF to separate the combined frequency drop noted.
spatio-temporal blurring in the moving slanted-edge image. It is generally known that temporal response curves as
A stationary image of the semitransparent edge, and not the measured using build-up and decay system responses, corre-
radiopaque edge, was used in the calculation to avoid an sponding to rising and falling-edge profiles, may differ.28
inappropriate scaling of the temporal MTF by the effective This violates the necessary Fourier assumption of linearity,
edge attenuation, which describes the combined effects of as these differences are a consequence of exposure-rate-
scattered x rays and scaling due to use of a nonopaque dependent lag. It has been shown that these differences are
edge.15 no longer present when using a small-signal approach in
It is known that use of a finite ROI for the calculation of time, enabling a Fourier analysis.15 It has also been observed
a MTF may result in spectral leakage and a measured zero- that falling-edge measurements agree with small-signal
frequency value less than unity due to the truncation of tails results,15 and if this is found to be generally true, small-
in the line-spread function.28,32 We used open-field normal- signal spatio-temporal MTF definitions in this article can be
ization and set the zero-frequency value to unity when mea- replaced with large-signal falling-edge ones. This has the ad-
suring both spatial and temporal components of the MTF to vantage that more robust radiopaque measurements can be
ensure accurate MTF measurements,32 as shown in Fig. 1. used in place of semitransparent ones for spatio-temporal
Incorrectly normalizing all values in the spatial component DQE calculations.

Medical Physics, Vol. 37, No. 11, November 2010

6066 S. N. Friedman and I. A. Cunningham: Spatio-temporal DQE and its application to fluoroscopic systems 6066

lag-corrected

detective quantum efficiency

1
spatio-temporal

detective quantum efficiency 0.8

1

0.8 0.6
0.6

0.4 0.4
0.2

0
0.2
0
0
0.5
5
1
10 spatial frequency 0
temporal frequency 1.5 0 0.5 1 1.5
(Hz)
15
(mm−1) spatial frequency (mm−1)

(a) RQA-5, 3.9 µR/frame (b) y-frequency direction

detective quantum efficiency

0.8

0.6

0.4

0.2

0
0 2 4 6 8 10 12 14
temporal frequency (Hz)

(c) t-frequency direction

FIG. 3. Calculated spatio-temporal DQE values of an XRII-based fluoroscopic system for the RQA-5 beam at 3.9 ␮R/frame are displayed in 共a兲, and values
along each of the axis are displayed in 共b兲 and 共c兲 corresponding to the spatial and temporal DQEs, respectively. The 共spatial兲 lag-corrected DQE for the
system is also displayed in 共b兲.

The calculation of the aliased squared temporal MTF is V.B. Spatio-temporal noise power spectrum
important to assess the contribution of aliasing to the shape
of the temporal NPS and requires a low-noise MTF measure- Calculated NPS values along the temporal-frequency axis
ment. Our calculation was achieved using the first three show close agreement with the aliased squared temporal
aliases in both negative and positive temporal-frequency di- MTF values when appropriately scaled. While the shape of
rections. This corresponds to superposition of squared mea- the NPS can result from different processes, such as aliasing
sured temporal MTF values up to ⫾120 Hz. Values of or the statistical nature of spatial blur and temporal lag, this
sinc2共at␯兲 are less than 10−4 for values greater than ⫾120 Hz result indicates that the shape along the t-frequency axis is
and represent the upper limit of squared temporal MTF val- likely dominated by temporal-aliasing effects, and additive
ues 共actual values are likely much less兲, and thus truncation and uncorrelated noise are likely negligible. This suggests
effects were assumed to be negligible. The calculation of the that the test system satisfies many of the necessary require-
aliased squared MTF was limited to these aliases to mini- ments for lag-corrected metrics 共i.e., deterministic lag, and
mize the inclusion of high-frequency positively-biased negligible additive and uncorrelated noise兲 and is consistent
noise,28 which was found to be a negligible contribution at with previous conclusions.15 However, the effect of temporal
these frequencies. aliasing on the accuracy of lag-corrected metrics is not yet

Medical Physics, Vol. 37, No. 11, November 2010

6067 S. N. Friedman and I. A. Cunningham: Spatio-temporal DQE and its application to fluoroscopic systems 6067

lag-corrected

detective quantum efficiency

1
spatio-temporal

detective quantum efficiency 1 0.8

0.8
0.6
0.6

0.4
0.4
0.2

0 0 0.2
0
0.5
5
1
10 spatial frequency 0
temporal frequency 1.5
15 −1 0 0.5 1 1.5
(Hz) (mm ) spatial frequency (mm−1)

(a) hardened 50 kVp, 1.9 µR/frame (b) y-frequency direction

detective quantum efficiency

0.8

0.6

0.4

0.2

0
0 2 4 6 8 10 12 14
temporal frequency (Hz)

(c) t-frequency direction

FIG. 4. Calculated spatio-temporal DQE values of an XRII-based fluoroscopic system for the hardened 50 kVp beam at 1.9 ␮R/frame are displayed in 共a兲,
and values along each of the axis are displayed in 共b兲 and 共c兲 corresponding to the spatial and temporal DQEs, respectively. The 共spatial兲 lag-corrected DQE
for the system is also displayed in 共b兲.

known. It is important to note that lag-corrected and spatio- above 6 Hz 关Figs. 3共c兲 and 4共c兲兴 could indicate reduced
temporal NPS values have different units 共mm2 and mm2 s, signal-to-noise ratio of quickly changing structures in image
respectively兲 and are not compared directly. data. This nonideal behavior in time shows that a determin-
All NPS measurements are limited by the sampling cutoff istic approach to lag does not accurately describe fluoro-
frequency, as determined by the sampling rate in both space scopic temporal performance and is likely due to the previ-
and time, and corresponded to ⬃1.8 mm−1 and 15 Hz, re- ously described temporal noise aliasing.
spectively. Many manufacturers allow the user to increase lag
through the system console and thus decrease the measured
noise in each frame through the resulting frame-averaging
V.C. Detective quantum efficiency and fluoroscopy
effect. However, it is important to note that this is typically
We have presented a comprehensive spatio-temporal de- achieved using postprocessing frame averaging that can be
scription of fluoroscopic system performance and dose effi- modeled as a linear 共deterministic兲 temporal filter. Such de-
ciency, achieved under fluoroscopic exposure-rate condi- terministic filters affect the temporal components of the MTF
tions, using the spatio-temporal DQE. The spatio-temporal and NPS in a similar manner and, thus, are not expected to
DQE showed nonideal performance in both space and time, have any effect on the resulting spatio-temporal DQE.
a result that could not be determined using lag-corrected A low-frequency drop was noted in the measured spatio-
metrics. This suggests that reduced temporal performance temporal DQEs along the spatial axes. This is an unusual

Medical Physics, Vol. 37, No. 11, November 2010

6068 S. N. Friedman and I. A. Cunningham: Spatio-temporal DQE and its application to fluoroscopic systems
result and corresponds to the LFD in the measured spatial
MTFs as obtained using the open-field MTF normalization,32
which preserves this drop better than zero-frequency normal-
ization. It is important to note that a LFD in the DQE is only
present when a LFD is present in the MTF without a corre-
sponding one in the measured NPS, as occurred for the test
system. However, since the correlated component of the NPS
is known to be proportional the squared MTF, it is reason-
able to expect a LFD to be present in correlated-noise domi-
nant systems such as the test system. To the best of our
knowledge, a NPS measurement containing a low-frequency
drop has not been described in the literature, and it is not yet
known whether this is because a NPS with a LFD is non-
physical or accepted methods for calculating the NPS pre-
vent the measurement of a LFD. The latter may be true as
accurate measurements of the NPS at low frequencies are
hard to obtain and thus the calculated low-frequency drop in
the DQE may be an artifact. It should also be noted that for
uncorrelated-noise dominant systems, no LFD is expected in
the NPS and a LFD is therefore predicted in the resulting
DQE.
Calculation of the lag-corrected DQE are presented to il-
lustrate the relationship with the spatio-temporal DQE and
required the temporal correction factor ␤. The exposure-rate
dependence of lag, and therefore ␤, has previously been
demonstrated on the test system, requiring a small-signal
approach.15 The temporal MTF as measured using the small-
signal semitransparent edge method was used to calculate ␤
to ensure the accuracy and validity of the results.
In the presence of uncorrelated noise or additive noise, the
lag-corrected DQE will systematically overcorrect the mea-
sured DQE values.13 However, since the effect of spatial and
temporal aliasing on lag-corrected measurements is not yet
known, the relationship between lag-corrected and spatio-
temporal DQE values cannot yet be predicted.
Agreement was previously found between fluoroscopic
lag-corrected DQE measurements and simulated radio-
graphic 共conventional兲 DQE measurements acquired under
similar conditions on the test system, suggesting that the lag-
corrected DQE provides accurate measurements of system
performance.15 We expected similar agreement between lag-
corrected results and the spatial component of the spatio-
temporal DQE. However, a small ⬃10% difference was
noted between the results 关Figs. 3共b兲 and 4共b兲兴. It is not yet
known whether this is due to noise aliasing in the t direction,
experimental error, or other causes, whether this difference is
indicative of all systems or just the test system, and the
spatio-temporal DQE should always be used to ensure accu-
rate measurements in both space and time.
Spatio-temporal calculations require more data than cor-
responding lag-corrected ones, as the ROIs must be created
in both space and time. For the test system, each spatio-
temporal ROI used for calculation extended over a set of 64
frames. Lag-corrected calculations only require spatial ROIs
and thus separate ROIs were formed on each of the corre-
sponding 64 frames. This resulted in an eightfold increase in
precision for the lag-corrected DQE values compared to the
Medical Physics, Vol. 37, No. 11, November 2010
6068
spatio-temporal values, corresponding to the 64-fold increase
in frames available for individual spatial ROI creation.
V.D. Clinical applications
Only continuous fluoroscopy results were obtained on a
bench-top system and it is not yet known whether these val-
ues are indicative of other fluoroscopic systems. Clinical sys-
tems are designed to operate in pulsed mode for improved
temporal resolution,36 and the evaluation of these systems
may require extension to pulsed fluoroscopy.
The effect of pulsed fluoroscopy on spatio-temporal DQE
measurements will require careful consideration. Temporal
MTF values for short pulses are expected to be larger than
corresponding values for longer pulses. NPS values may not
scale directly with these increases in MTF values in the case
of quantum-based lag. Thus, systems measured under short-
pulse conditions may result in larger DQE values along the
temporal-frequency axis than values obtained on the same
system using longer-pulse conditions or continuous fluoros-
copy. This may result in misleading DQE values should val-
ues of two systems that were measured under differing pulse
conditions be compared. Extension of the spatio-temporal
DQE to pulsed fluoroscopy is expected to be possible by
representing x-ray pulses as a cyclostationary random
process,5 but is beyond the scope of the present article.
While fluoroscopy has been presented as the motivation
for a spatio-temporal DQE and is a straightforward applica-
tion of the metric, spatio-temporal metrics are not limited to
fluoroscopy and could be used for any quantum-based sys-
tem exhibiting temporal correlations. Spatio-temporal cas-
caded theory may also play a role in understanding signal
and noise propagation in computed tomography or other im-
aging systems involving fast data acquisition.
VI. CONCLUSIONS
We have introduced a spatio-temporal detective quantum
efficiency that describes the signal and noise performance of
quantum-based imaging systems in two spatial domains and
one temporal domain. It was illustrated by an application to a
bench-top XRII-based fluoroscopic system. Specific conclu-
sions from this work include
共1兲 The spatio-temporal NPS was measured in three dimen-
sions while the spatio-temporal MTF was only measured
in the 共x , t兲 and 共y , t兲-frequency planes. Thus, values of
the spatio-temporal DQE were calculated in these 2D-
frequency planes only.
共2兲 The bench-top 共nonpulsed兲 fluoroscopic system showed
suboptimal performance due primarily to noise aliasing
in the temporal domain. This type of performance deg-
radation was not identified by spatial metrics including
the lag-corrected DQE. Lag can also degrade the perfor-
mance by a spatio-temporal secondary quantum sink,
but this was not observed in the test system.
共3兲 Temporal noise aliasing causes a “folding” of the presa-
mpling temporal NPS about the sampling frequency of
30 Hz and harmonics. This tends to flatten the temporal
6069 S. N. Friedman and I. A. Cunningham: Spatio-temporal DQE and its application to fluoroscopic systems
NPS and decreases the temporal DQE with increasing
frequency. The shape of the measured temporal NPS
was consistent with the aliasing of the squared temporal
MTF, indicating a drop in the temporal DQE starting at
6 Hz and increasing to more than 50% at the sampling
cutoff frequency of 15 Hz. It is expected that this re-
duced DQE would compromise the detection of chang-
ing image structures.
ACKNOWLEDGMENTS
The authors gratefully acknowledge the financial support
of the Canadian Institutes of Health Research. S.N.F. ac-
knowledges a scholarship from The University of Western
Ontario. The authors would also like to thank Mike Mc-
Donald and David Paribello for their help with experiments,
and Nicolas Jankovic for designing and building the moving
slanted-edge apparatus.
a兲
Electronic mail: saul.friedman@gmail.com
b兲
Electronic mail: icunning@imaging.robarts.ca
1
L. K. Wagner, M. D. McNeese, M. V. Marx, and E. L. Siegel, “Severe
skin reactions from interventional fluoroscopy: Case report and review of
the literature,” Radiology 213, 773–776 共1999兲.
2 enhancement due to unsharp-mask processing in x-ray fluoroscopy,” J.
R. Shaw, “The equivalent quantum efficiency of the photographic pro-
cess,” J. Photogr. Sci. 11, 199–204 共1963兲.
3 Y. Jiang and D. L. Wilson, “Optimization of detector pixel size for stent
J. C. Dainty and R. Shaw, Image Science 共Academic, New York, 1974兲.
4 visualization in x-ray fluoroscopy,” Med. Phys. 33, 668–678 共2006兲.
R. F. Wagner and E. P. Muntz, “Detective quantum efficiency 共DQE兲
analysis of electrostatic imaging and screen-film imaging in mammogra-
phy,” Proc. SPIE 173, 162–165 共1979兲.
5
I. A. Cunningham, “Ch 2: Applied linear-systems theory,” in Physics and
Psychophysics, edited by J. Beutel, H. L. Kundel, and R. V. Metter 共SPIE,
Bellingham, 2000兲, Vol. 1, pp. 79–159.
6
A. Papoulis and S. U. Pillai, Probability, Random Variables and Stochas-
tic Processes, 4th ed. 共McGraw-Hill, New York, 2002兲.
7
R. K. Swank, “Measurement of absorption and noise in an x-ray image
intensifier,” J. Appl. Phys. 45, 3673–3678 共1974兲.
8
P. R. Granfors and R. Aufrichtig, “DQE共f兲 of an amorphous silicon flat
panel x-ray detector: Detector parameter influences and measurement
methodology,” Proc. SPIE 3977, 2–13 共2000兲.
9 30
P. R. Granfors, R. Aufrichtig, G. E. Possin, B. W. Giambattista, Z. S.
Huang, J. Liu, and B. Ma, “Performance of a 41⫻ 41 cm2 amorphous
silicon flat panel x-ray detector designed for angiographic and R&F im-
aging applications,” Med. Phys. 30, 2715–2726 共2003兲.
10
B. Menser, R. J. M. Bastiaens, A. Nascetti, M. Overdick, and M. Simon,
“Linear system models for lag in flat dynamic x-ray detectors,” Proc.
SPIE 5745, 430–441 共2005兲.
11
I. A. Cunningham, T. Moschandreou, and V. Subotic, “The detective
quantum efficiency of fluoroscopic systems: The case for a spatial-
temporal approach 共or, does the ideal observer have infinite patience?兲,”
Proc. SPIE 4320, 479–488 共2001兲.
12
J. H. Siewerdsen, I. A. Cunningham, and D. A. Jaffray, “A framework for
noise-power spectrum analysis of multidimensional images,” Med. Phys.
29, 2655–2671 共2002兲.
13
R. Akbarpour, S. N. Friedman, J. H. Siewerdsen, J. D. Neary, and I. A.
Cunningham, “Signal and noise transfer in spatio-temporal quantum
based imaging systems,” J. Opt. Soc. Am. A Opt. Image Sci. Vis 24,
B151–B164 共2007兲.
14
IEC, “Characteristics of digital x-ray imaging devices—Part 1–3: Deter-
Medical Physics, Vol. 37, No. 11, November 2010
6069
mination of the detective quantum efficiency—Detectors used in dynamic
imaging,” Medical Electrical Equipment IEC 62220-1-3 共International
Electrotechnical Commission, 2008兲.
15
S. N. Friedman and I. A. Cunningham, “A small-signal approach to tem-
poral modulation transfer functions with exposure-rate dependence and its
application to fluoroscopic detective quantum efficiency,” Med. Phys. 36,
3775–3785 共2009兲.
16
ICRU, “Medical imaging—The assessment of image quality,” ICRU Re-
port No. 54 共International Commission of Radiation Units and Measure-
ments, Bethesda, MD, 1995兲.
17
D. L. Wilson, P. Xue, and R. Aufrichtig, “Perception of fluoroscopy last-
image hold,” Med. Phys. 21, 1875–1883 共1994兲.
18
D. L. Wilson, K. N. Jabri, P. Xue, and R. Aufrichtig, “Perceived noise
versus display noise in temporally filtered image sequences,” J. Electron.
Imaging 5, 490–495 共1996兲.
19
K. N. Jabri and D. L. Wilson, “Detection improvement in spatially filtered
x-ray fluoroscopy image sequences,” J. Opt. Soc. Am. A Opt. Image Sci.
Vis 16, 742–749 共1999兲.
20
D. L. Wilson, K. N. Jabri, and R. Aufrichtig, “Perception of temporally
filtered x-ray fluoroscopy images,” IEEE Trans. Med. Imaging 18, 22–31
共1999兲.
21
D. L. Wilson and R. Manjeshwar, “Role of phase information and eye
pursuit in the detection of moving objects in noise,” J. Opt. Soc. Am. A
Opt. Image Sci. Vis 16, 669–678 共1999兲.
22
F. J. Sanchez-Marin, Y. Srinivas, K. N. Jabri, and D. L. Wilson, “Quan-
titative image quality analysis of a nonlinear spatio-temporal filter,” IEEE
Trans. Image Process. 10, 288–295 共2001兲.
23
K. N. Jabri and D. L. Wilson, “Quantitative assessment of image quality
Opt. Soc. Am. A Opt. Image Sci. Vis 19, 1297–1307 共2002兲.
24
25
Thomson Tubes Electroniques, Specification sheet TII/1.087/91 for Tha-
les image intensifier model TH9432 QXH304VR.
26
H. Lai and I. A. Cunningham, “Noise aliasing in interline-video-based
fluoroscopy systems,” Med. Phys. 29, 298–310 共2002兲.
27
R. Fahrig, M. Moreau, and D. W. Holdsworth, “Three-dimensional com-
puted tomographic reconstruction using a C-arm mounted XRII: Correc-
tion of image intensifier distortion,” Med. Phys. 24, 1097–1106 共1997兲.
28
S. N. Friedman and I. A. Cunningham, “A moving slanted-edge method
to measure the temporal modulation transfer function of fluoroscopic sys-
tems,” Med. Phys. 35, 2473–2484 共2008兲.
29
P. F. Judy, “The line spread function and modulation transfer function of
a computed tomographic scanner,” Med. Phys. 3, 233–236 共1976兲.
E. Samei, M. J. Flynn, and D. A. Reimann, “A method for measuring the
presampled MTF of digital radiographic systems using an edge test de-
vice,” Med. Phys. 25, 102–113 共1998兲.
31
R. B. Blackman and J. W. Tukey, The Measurement of Power Spectra
共Dover, New York, 1958兲.
32
S. N. Friedman and I. A. Cunningham, “Normalization of the modulation
transfer function: The open-field approach,” Med. Phys. 35, 4443–4449
共2008兲.
33
IEC, “Characteristics of digital x-ray imaging devices—Part 1: Determi-
nation of the detective quantum efficiency,” Medical Electrical Equip-
ment IEC 62220-1-1 共International Electrotechnical Commission, 2003兲.
34
L. Levi, “Spatiotemporal transfer function: Recent developments,” Appl.
Opt. 22, 4038–4041 共1983兲.
35
U. Neitzel, E. Buhr, G. Hilgers, and P. R. Granfors, “Determination of the
modulation transfer function using the edge method: Influence of scat-
tered radiation,” Med. Phys. 31, 3485–3491 共2004兲.
36
P. Xue, K. N. Jabri, and D. L. Wilson, “Adaptive reference/test forced-
choice method with application to fluoroscopy perception,” Proc. SPIE
3036, 298–307 共1997兲.