Multivendor, Multisite DCE-MRI Phantom Validation Study
Edward Jackson1, Edward Ashton2, Jeffrey Evelhoch3, Michael Buonocore4, Greg Karczmar5, Mark Rosen6, David Purdy7,
Sandeep Gupta8, Gudrun Zahlmann9
of Texas M.D. Anderson Cancer Center, 2VirtualScopics, Inc., 3Merck Research Laboratories, 4University of California Davis, 5University of Chicago, 6University of
Pennsylvania, 7Siemens Medical Systems, 8GE Global Research Center, 9F. Hoffman - La Roche, Ltd.
While the data from Sites A and B were quite consistent, data from Site C
demonstrated dramatic departures from the trends seen for Sites A and B.
INTRODUCTION
DCE signal intensity vs. IR R1 measures were not linearly related, and VFA R1
Figure 2: Phantom and cuboid positioned
The QIBA initiative seeks to advance quantitative imaging (QI) and the use of
in a 4-channel torso phased-array coil.
measures did not correlate well with IR R1 measures. The underlying issues
imaging biomarkers in clinical trials and clinical practice by: 1) collaborating to
The phantom is scanned five times, before
are now under investigation.
These inconsistencies demonstrate the
identify needs and solutions to develop and test consistent, reliable, valid, and
and following each of four 90o rotations.
importance of the QIBA initiative to “identify needs and solutions to develop
achievable QI results across imaging platforms, clinical sites, and time, and 2)
and test consistent, reliable, valid…quantitative imaging results across imaging
accelerating the development and adoption of hardware and software
platforms, clinical sites, and time”.
standards needed to achieve accurate and reproducible QI results from
Corrected
Uncorrected
Series
Acquisition Details
Time (min)
Mean correlation coefficient: 0.979
imaging methods [1]. The QIBA DCE-MRI technical committee has initially
Mean correlation coefficient: 0.818
Mean slope / intercept: 45.2 / -4.2
Scout scan
5
Mean slope / intercept: 37.9 / -5.4
focused on item 1) above by initiating a multivendor, multicenter, test-retest
o flip angle, 8 averages
Figure 3a:
Ratio
images
Body
coil;
15
2
phantom assessment building upon the previous efforts of the Imaging
Site A,
o
Ratio images
Phased array coil; 15 flip angle, 8 averages
2
Response Assessment Teams (IRAT) DCE-MRI phantom studies [2]. Initial
Week 0
results from this initiative are summarized in this exhibit.
SNR images
15o flip angle; 8 sequential acquisitions
8
Analysis of the signal characteristics in the DCE scans was accomplished by
placing a uniform spherical 2-cm diameter region of interest (ROI) in the center
of each phantom compartment. Mean and median pixel values within each
ROI were calculated, along with SNR and CNR values. Noise in each
compartment was defined as the standard deviation of the differences at each
pixel between one phase and the next divided by √2. Signal was defined as
the mean signal value within each ROI. Contrast was defined as the absolute
difference between the mean signal in an ROI and that of the central 6-cm
sphere. The raw data thus obtained were provided to the QIBA DCE-MRI
Technical Committee for further analysis.
PRELIMINARY RESULTS
C
75
D
50
A'
25
0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
175
150
125
A
100
B
C
75
D
50
A'
25
0
0.5
4.0
1.0
1.5
2.0
IR R1 (s-1)
IR R1
200
Mean correlation coefficient: 0.816
Mean slope / intercept: 38.0 / -2.9
175
150
125
A
100
B
C
75
D
50
A'
25
0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
200
A
100
B
C
75
D
50
A'
0
1.0
1.5
2.0
80
B
C
60
D
40
A'
20
0
2.5
3.0
3.5
160
100
A
80
B
C
60
D
40
A'
0
0.5
1.0
1.5
2.0
IR R1
100
A
80
B
C
60
D
40
A'
20
0
1.5
2.0
2.5
3.0
3.5
3.5
120
100
A
80
B
C
60
D
40
A'
0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
Difference in R1 (VFA vs IR) - Site B
2.0
20090615
y = 0.963x - 0.0985
R² = 0.9855
20090622
20090615
20090622
1.0
0.5
0.0
2.0
2.5
3.0
(s-1)
3.5
Figure 3d:
Site B,
Week 1
20
VFA R1 vs IR R1 - Site B
1.5
4.0
Mean correlation coefficient: 0.994
Mean slope / intercept: 37.6 / -1.8
140
IR R1 (s-1)
2.5
1.5
3.0
IR R1 (s-1)
y = 1.0534x - 0.2113
R² = 0.9903
1.0
160
4.0
3.5
3.0
2.5
(s-1)
Corrected
120
1.0
Figure 3c:
Site B,
Week 0
20
4.0
Mean correlation coefficient: 0.982
Mean slope / intercept: 33.1 / -3.2
0.5
4.0
120
Uncorrected
140
3.5
Mean correlation coefficient: 0.994
Mean slope / intercept: 38.3 / -1.9
140
IR R1 (s-1)
160
3.0
Corrected
A
2.0
2.5
IR R1 (s-1)
100
1.5
Figure 3b:
Site A,
Week 1
25
0.5
120
1.0
4.0
125
4.0
Mean correlation coefficient: 0.985
Mean slope / intercept: 33.0 / -3.0
0.5
3.5
150
Uncorrected
140
3.0
Mean correlation coefficient: 0.974
Mean slope / intercept: 44.6 / -1.5
175
IR R1 (s-1)
160
2.5
(s-1)
Corrected
Signal Intensity (Mean, DCE)
Uncorrected
IR R1
Current Status: Thus far, complete data sets have been obtained from two
sites (two MR vendors) and partial data obtained from one site (third vendor).
DCE Mean Signal Intensity vs. R1: Figure 3 shows the uncorrected and
corrected DCE signal intensity vs. inversion recovery R1 measures for data
obtained at two sites.
IR R1 Measures vs. VFA R1 Measures: Figure 4 shows the VFA-derived R1
measures vs. the inversion recovery R1 measures for data obtained at a single
site, but on two subsequent weeks. The left figure shows the linear regression
while the right figure shows the Bland-Altman plot.
DCE Signal Intensity Variations: The coefficients of variation of the signal
intensity over the duration of the DCE acquisitions for the baseline and week 1
scans were 0.50% and 0.56%, respectively, for Site A, and 0.41% and 0.41%,
respectively, for Site B.
Signal Intensity (Mean, DCE)
B
Signal Intensity (Mean, DCE)
Import the body coil and phased array ratio images
Normalize the range of the two images
Calculate signal intensity ratios (body coil:phased array) for each pixel
Apply 21x21 pixel kernel median filter
Multiply each pixel in the source image by the ratio map pixel data
100
Signal Intensity (Mean, DCE)
1.
2.
3.
4.
5.
A
200
Difference in R1 (VFA-IR) (s-1)
Scan Protocol: Initial phantom characterization (inversion recovery T1
measurements, phantom cross-comparison scans, initial QIBA protocol scans)
were performed at M.D. Anderson Cancer Center. At each subsequent site,
the phantom was scanned twice, with one week between the scans. During
each scanning session, the phantom was rotated 90o four times and
rescanned at each position. This provides data necessary for a “coffee break”
test–retest analysis as well as a one-week interval test-retest analysis. The
phantom and cuboid were positioned in a phased-array receive coil as shown
in Figure 2. The phantom position at each of the five rotations was identified as
A, B, C, D, and A’. Table 1 summarizes the data obtained at each rotation. All
data were acquired using a 3D fast spoiled gradient echo sequence with all
acquisition parameters matched, vendor-to-vendor, as closely as possible.
The same protocol was used to obtain data one week later. The inversion
recovery (IR) based T1 measurements were only performed once and the
results used as “ground truth” for the subsequent variable flip angle (VFA) T1
measurements. VFA-based T1 measurements are commonly used in DCEMRI applications as they can be obtained in a reasonable time while IR-based
T1 measurements cannot.
Data Analysis: The raw data analysis was carried out using software
developed by VirtualScopics, Inc. From the DCE-MRI acquisition data, signal
intensity, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR)
measures were computed from each of the eight contrast spheres. T1
measures were computed from the variable flip angle data from each sphere.
These measures were obtained both before and after correction of the phased
array coil data for spatial variations in coil sensitivity. This correction was
carried out as follows:
125
Mean VFA R1 (s-1)
Figure 1: Phantom
scanning process.
40 phases for Rotations A & A’, 6 phases for other 6 (40 phases) or
DCE-MRI images
rotations; 30o flip angle; ≤10 s temporal resolution 1 (6 phases)
Table 1: Data acquired at each rotation of the phantom. All data were acquired again
one week later.
150
Signal Intensity (Mean, DCE)
Phantom: Two matched 20-cm internal diameter spherical phantoms were
purchased from The Phantom Laboratory (funded by National Cancer Institute
contracts N01-CO-12400 and 27XS112). For this particular application, the
key component of the phantom design was the inclusion of eight 3-cm
diameter spheres filled with CuSO4-doped H2O to yield T1 relaxation times
ranging from ~300-960 ms. The remainder of the phantom was identical to the
ADNI Magphan phantom [3, 4]. including a 6-cm diameter central sphere filled
with pure water. A 17-cm by 11-cm “cuboid”, also filled with 30 mM NaCl water,
was used to appropriately load the radiofrequency coil. This phantom design
differed from that used by the IRAT MR Committee [2] in the use of 30 mM
NaCl water in the flood section of the phantom and cuboid and no D2O was
used in the 8 contrast spheres. Otherwise, the phantom components and
positioning were identical for the IRAT and QIBA DCE-MRI initiatives.
Scanners and Sites: The phantom studies are initially being performed at five
sites (M.D. Anderson Cancer Center, University of Chicago, University of
Pennsylvania, Duke University Medical Center, and University of California
Davis) utilizing 1.5T scanners from GE, Philips, and Siemens. (Figure 1)
6
175
Signal Intensity (Mean, DCE)
METHODS & MATERIALS
angles; 4 averages
200
Signal Intensity (Mean, DCE)
Variable flip angle 2, 5, 10, 15, 20, 25,
30o flip
Signal Intensity (Mean, DCE)
1University
4.0
0.10
0.05
0.00
1.0
1.5
2.0
2.5
3.0
3.5
-0.05
Figure 4:
-0.10
20090615
-0.15
20090622
-0.20
-0.25
VFA R1 vs IR
R1 Measures
-0.30
-0.35
IR R1 (s-1)
CONCLUSIONS
Results obtained thus far demonstrate, with appropriate choices of pulse
sequences and acquisition parameters across vendors, 1) signal intensity
measures, when corrected for receiver coil sensitivity variations, correlate well
with R1, 2) VFA R1 measures correlate well with IR R1 measures, 3) these
findings are consistent over short times (“coffee break”) and longer times (1
week), 4) such phantom-based assessment of scanner performance is critical
to validate imaging biomarker data from multivendor, multicenter applications.
References
[1] http://qibawiki.rsna.org
[2] http://www.iratnetwork.org
[3] http://www.loni.ucla.eduADNI/
[4] http://www.phantomlab.com/magphan_adni.html