EPID IMAGE QUALITY: LAS VEGAS PHANTOM AS AN OBJECTIVE ...

Witold Skrzyński 1
Andrew J. Reilly 2
David I. Thwaites 2
Wojciech Bulski 1
1 Centre of Oncology, Warsaw, Poland
2 Edinburgh Cancer Centre, Western General Hospital, Edinburgh, UK
w.skrzynski@rth.coi.waw.pl
.
EPID IMAGE QUALITY: LAS VEGAS PHANTOM
AS AN OBJECTIVE TOOL
Abstract: This work is a comparison of three approaches for
EPID image quality control using the Las Vegas phantom:
contrast-detail counting by human observers, an automatic
detail counting technique and the direct calculation of performance
indices (SNR and MTF). The standard detail
counting procedure is observer-dependant and insensitive to
artifacts, whilst the automatic method is robust and fully
objective. Results of MTF calculations were in excellent
agreement with those obtained using a commercial QC-3V
phantom. Basic image analysis software thus enables the
Las Vegas phantom to be used as an objective tool for
evaluating the quality of electronic portal images.
1. INTRODUCTION
EPIDs (Electronic Portal Imaging Devices) are
widely used for the assessment of geometrical accuracy
in modern radiotherapy [1]. Portal images of a high and
consistent quality are essential for this task.
The Las Vegas phantom (Figure 1) is a contrast-detail
phantom supplied with most EPIDs for acceptance
testing [2] and routine image quality control. The
control procedure suggested by manufacturers is usually
limited to a simple check of the number of details visible
in the image by a human observer.
Increasing contrast
(hole depth)
Increasing detail size
(hole diameter)
Fig. 1. Schematic view of Las Vegas phantom
Several other phantoms and test methods are also in
common use. One of the most popular is the commercial
QC-3V phantom with high-contrast bar resolution patterns
and uniform regions of interest [3]. This is supplied
with the PIPSpro [4] software for automated analysis of
image quality. PIPSpro calculates the MTF (modulation
transfer function) and SNR (signal to noise ratio). Recently
it has been suggested that a very simple phantom
and home-made software might be used to objectively
measure image quality parameters [5].
This work considers two objective approaches for
image quality control with the Las Vegas phantom: detail
Warsaw, 29-30 Sept.
2005
counting by a computer observer, and utilizing the phantom
in a new mode to calculate SNR and MTF.
2. MATERIALS AND METHODS
2.1 EPIDs
Measurements were performed using Varian PortalVision
aS500 (amorphous silicon) EPIDs in Edinburgh
Cancer Centre and aS500 and LC250 (liquid ion
chamber) EPIDs installed in the Centre of Oncology in
Warsaw. All of the EPIDs were installed on Varian Clinac
linear accelerators.
2.2 Software
All the images were analyzed using the ImageJ
software platform [6]. Several complex image analysis
routines were required for this work and custom ImageJ
‘plugins’ were developed to implement these.
2.3 Detail counting
Las Vegas phantom images for five aS500 EPIDs in
Edinburgh Cancer Centre were acquired on a weekly
basis over a period of 9 months. For each image, the
number of visible details was counted immediately following
acquisition. Three trained observers performed
the task on a rotational basis and the results have been
investigated for intra- and inter-observer variability.
A statistical algorithm for automated detail counting
([7], modified) was also implemented. Results obtained
by human and computer observers were then compared.
2.4 MTF and SNR calculation
Pairs of images of Las Vegas and QC-3V phantoms
were acquired for various beam energies and dose rates.
For the Las Vegas phantom the MTF was calculated
as the magnitude of the Fourier transform of the
edge spread function (ESF) [8]. The difference between
the mean pixel value in a uniform region of the phantom
and that of a region outside the phantom was taken as the
signal for the SNR calculation.
Images of the QC-3V phantom were first analyzed
with the PIPSpro software, which outputs MTF normalized
to unity at the frequency of 0.1 line-pairs per mm
(lowest frequency bar test pattern). These were then reanalyzed
using a modified method [9] which outputs the
“correct” MTF (normalized to unity at zero frequency).

3. RESULTS AND DISCUSSION
3.1 Detail counting
In Figure 2 the performance distribution of two of
the Edinburgh observers is plotted and illustrates interobserver
variability. Figure 3 presents a comparison of
results obtained by human and computer observers for
one EPID. Results obtained by human observers appear
to show that image quality is improving over time. This
positive trend is rather unexpected (detectors are ageing)
and is thought to be as a result of observers becoming
progressively more confident in identifying details.
Computer analysis of the images suggests that image
quality is stable over time.
The number of human or computer-detected details
was not significantly affected by image artifacts (e.g.
horizontal banding, defective pixels). Detail-counting is
therefore is not a good measure of overall image quality.
number of counts
30
20
10
observer1
observer2
0
0,9 1,0 1,1
Las Vegas test result
(number of visible holes normalized to mean)
Fig. 2. Number of visible structures in images
of Las Vegas phantom for two observers
number of visible holes
24
22
20
0 10 20 30 40
24
22
20
human observer(s)
computer observer
0 10 20
week
30 40
Fig. 3. Las Vegas phantom test results for human and
computer observers
3.2 Modulation Transfer Function
Typical results of the MTF calculations are presented
in Figure 4. The curves obtained from the analyses
of the edge profile (Las Vegas phantom) and bar test
patterns (QC-3V phantom) are in excellent agreement.
As expected, the results obtained with PIPSpro do not
agree with other methods due to non-standard MTF normalization.
(The PIPSpro curve is shifted to the right.)
For routine quality control purposes the SNR and
the frequency at which the MTF is 50% (f50) are typically
recorded and it was found that these quantities are sensitive
to image artifacts. They are therefore a useful indicator
of when technical maintenance may be required.
MTF
1,0
0,5
0,0
0,0 0,5 1,0
Las Vegas - edge profile
QC-3V (PIPSpro results)
QC-3V (corrected MTF)
resolution (line pairs per mm)
Fig. 4. MTF obtained with three methods
4. CONCLUSIONS
The standard detail counting procedure is observerdependant
and insensitive to artifacts. However, basic
analysis software enables the Las Vegas phantom to be
used as a fully objective tool which is sensitive to artifacts
and hence suitable for inclusion in a quality control
program.
LITERATURE
[1] P. Munro, ‘‘Portal imaging technology: Past, present and
future,’’ Semin. Radiat. Oncol. 5, 115–133, 1995
[2] Low D.A., Klein E.E., Maag D.K., Umfleet W.E., Purdy
J.A., Commissioning and periodic quality assurance of a
clinical electronic portal imaging device, Int. J. Radiation
Oncology Biol. Phys. 34(1), p.117, 1996
[3] Shalev S., Test phantoms for EPIDs: A comparison
between the Las Vegas and the QC-3 phantoms, Masthead
Imaging Corporation, 2003
[4] Shalev S., PIPSpro, version 3.2.2, Masthead Imaging
Corporation, British Columbia, Canada, 2002
[5] Reilly A.J., Skrzyński W., Thwaites D.I., EPID Quality
Assurance: A Fresh Approach, Clinical Oncology
17(supp. 1), p.219, 2005
[6] Rasband W.S., ImageJ, U. S. National Institutes of
Health, Bethesda, Maryland, USA,
http://rsb.info.nih.gov/ij/, 1997-2005.
[7] Dong L, Boyer A.L., An objective method for evaluating
electronic portal imaging devices, Med. Phys. 21(6),
p.755, 1994
[8] International Electrotechnical Commission, International
Standard IEC 62220-1: Medical electrical equipment –
Characteristics of digital imaging devices – Part 1: Determination
of the detective quantum efficiency, Geneva,
2003
[9] Droege R.T., Morin R.L., A practical method to measure
the MTF of CT scanner, Med. Phys. 9(5):758-760, 1982
This work has been supported by the IAEA (Fellowship No.
C6/POL/03016). The authors gratefully acknowledge the
contribution of L Cherry, A Doig and W Ronaldson of
Edinburgh Cancer Centre in analyzing the Las Vegas Phantom
images.