7/21/2014
Modality specific QI:
PET/CT
Robert Jeraj
Associate Professor of Medical Physics, Human Oncology,
Radiology and Biomedical Engineering
Director of Translational Imaging Research Program
University of Wisconsin Carbone Cancer Center
rjeraj@wisc.edu
PET/CT during therapy
BEFORE
DURING
AFTER THERAPY
    …...  
DIAGNOSIS
and STAGING
TARGET
DEFINITION
EARLY
LATE
TREATMENT ASSESSMENT
PET/CT for diagnosis and staging
BEFORE
DURING
AFTER THERAPY
    …...  
DIAGNOSIS
and STAGING
TARGET
DEFINITION
EARLY
LATE
TREATMENT ASSESSMENT
1
7/21/2014
PET/CT for diagnosis and staging
 Qualitative FDG PET/CT has had a tremendous
impact in oncology
Hillner et al 2008, J Clin Oncol 26: 2155
Use of PET/CT
BEFORE
DURING
AFTER THERAPY
    …...  
DIAGNOSIS
and STAGING
TARGET
DEFINITION
EARLY
LATE
TREATMENT ASSESSMENT
Need for quantitative PET/CT
 Where is the tumor?
– SPATIAL accuracy
 What kind of the tumor is there?
– SPECIFIC accuracy
 How much of the tumor is there?
– ABSOLUTE accuracy
 How much has the tumor changed?
– RELATIVE accuracy
2
7/21/2014
WHAT LIMITS PET/CT
QUANTITATIVE ACCURACY?
Limited PET spatial resolution
4 mm
2 mm
1 mm
Limited signal recovery
Partial volume effects
Recovery coefficients
Recovery Coefficient
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0
5
10
15
20
25
30
Sphere Diameter (mm)
3
7/21/2014
Variability of signal recovery
Axial shift
Radial shift
McCall et al 2010, Phys Med Biol 55: 2789
Variability of tracer uptake
Inter-tumor variability
Intra-tumor variability
Pharmacokinetics analysis
1 min
15 min
60 min
FLT PET/CT
4
7/21/2014
PET imaging uncertainties
 Technical factors
– Relative calibration between PET scanner and dose calibrator
– Residual activity in syringe
– Incorrect synchronization of clocks
– Injection vs calibration time
– Quality of administration
 Physical factors
– Scan acquisition parameters
– Image reconstruction parameters
– Use of contrast agents
 Analytical factors
– Region of interest (ROI) definition
– Image processing
 Biological factors
– Patient positioning
– Patient breathing
– Uptake period
Jeraj et al. 2011, in Uncertainties in ext. beam RT
– Blood glucose levels
Boellaard et al. 2009, J Nucl Med 50: 11S
PET imaging uncertainties
 Technical factors
– Relative calibration between PET scanner and dose calibrator (10%)
– Residual activity in syringe (5%)
– Incorrect synchronization of clocks (10%)
– Injection vs calibration time (10%)
– Quality of administration (50%)
 Physical factors
– Scan acquisition parameters (15%)
– Image reconstruction parameters (30%)
– Use of contrast agents (15%)
 Analytical factors
– Region of interest (ROI) definition (50%)
– Image processing (25%)
 Biological factors
– Patient positioning (15%)
– Patient breathing (30%)
– Uptake period (15%)
– Blood glucose levels (15%) Jeraj et al. 2011, in Uncertainties in ext. beam RT
Boellaard et al. 2009, J Nucl Med 50: 11S
IMPACT OF UNCERTAINTIES
(Example of variable
reconstruction parameters)
5
7/21/2014
Reconstruction parameters
Reconstruction type Acquisition Recon Algorithm Grid Size Iterations Post filter
A
256 x 256
B
C
3D
3D OSEM
128 x 128
F
256 x 256
G
6 mm
6 mm
2
E
2D
6 mm
4
D
H
3 mm
2
3 mm
3 mm
2
2D OSEM
5 mm
4
I
128 x 128
5 mm
5 mm
2
J
3 mm
Uncertainty uptake values (ROIs)
20
global
individual
Uncertainty (%)
16
12
8
4
SU
Vm
ax
SU
Vm
ax
SU
Vp
ea
k
SU
Vp
ea
k
SU
Vm
ea
n
SU
Vm
ea
n
SU
Vt
ot
al
SU
Vt
ot
al
0
Reconstruction type
Uncertainty uptake values (within ROI)
25
25
Patient 1
10
5
0
2
4
6
8
Uncertainty (%)
10
12
median = 2.6
IQR = 2.3
20
Frequency (%)
Frequency (%)
15
0
Patient 2
median = 3.1
IQR = 3.9
20
15
10
5
0
0
2
4
6
8
10
12
Uncertainty (%)
6
7/21/2014
APPLICATIONS OF
QUANTITATIVE PET/CT
Target definition and treatment
assessment
PET/CT for target definition
BEFORE
AFTER THERAPY
DURING
    …...  
DIAGNOSIS
and STAGING
TARGET
DEFINITION
EARLY
LATE
TREATMENT ASSESSMENT
Adding FDG PET/CT helps
CT
FDG PET/CT
50% (30%-70%) decrease of the contouring standard
deviation, but it still remains significant!
Steenbakkers et al 2006, Int J Rad Oncol Biol Phys 64: 435
7
7/21/2014
Automatic segmentation needed
CT
128, 2 iter, 5mm
2D OSEM
256, 2 iter, 6mm
3D OSEM
White: SUV70%
Red: SUV40%
256, 2 iter, 3mm
256, 6 iter, 6mm
256, 2 iter, 6mm
128, 2 iter, 6mm
But still large uncertainties
Volume Variations (%)
150
Threshold-based
Impact of Iteration on target volumes
Gradient-based
Region-growing
100
50
0
-50
-100
128x128 grid size
3D Acquisition, 6mm PF
2D Acquisition, 5mm PF
B
C
D
E
F
G
Segmentation Techniques
Standardization is essential
 AAPM TG211 - Classification, Advantages and
Limitations of the Auto-Segmentation Approaches
for PET
– Manual segmentation is NOT the way to go, but
the final manual review required!
– Auto segmentation
•
•
•
•
•
Threshold-based (Erdi 1997, Paulino 2004)
Gradient-based (Geets 2007)
Region-growing (Drever 2006)
Statistical-based (Yu 2009)
…
– Reference benchmark dataset
8
7/21/2014
PET/CT for treatment assessment
BEFORE
AFTER THERAPY
DURING
    …...  
DIAGNOSIS
and STAGING
TARGET
DEFINITION
EARLY
LATE
TREATMENT ASSESSMENT
PET-based response assessment
 EORTC, NCI Recommendations (1999, 2005) 1,2
– SUV-based approach
– SUVmean and SUVmax
– Response categories with thresholds (CR, PR, SD, PD)
– Problems
• SUVmean – collapses information, sensitivity issues
• SUVmax – noise contamination
• fails to use all available functional data
• distribution
• heterogeneity
• no response threshold validation
• few sensitivity studies
• alternative measures
 PET Response Criteria in Solid Tumors (PERCIST) (2009) 3
– SUVpeak
1Young
et al 1999, 2Shankar et al 2006, 3Wahl et al 2009
Multiple measures (radiomics)
 Size measures
– Volume
– 1D size (axial)
SUVpeak
SUVmax SUVmean
 Standardized Uptake
Value (SUV) measures:
– SUVmean
– SUVtotal
– SUVmax
– SUVpeak
1D Size (axial)
250
Volume
Number of Voxels
 Uptake Non-uniformity
measure:
– SUVsd
 …
SUVtotal
SUVsd
200
150
100
50
0
0
5
10
15
20
Standardized Uptake Value
9
7/21/2014
Uncertainty of the imaging features
Galavis et al 2010, Acta Oncol 49: 1012.
Ambiguity of response evaluation
Pre-treatment
Post-treatment
SUV
18
FLT PET/CT
0
Vanderhoek et al 2013, J Nucl Med 54: 1188.
Ambiguity of response evaluation
140
100
Response (%)
ambiguous
response
SUVmean
SUVmax
SUVpeak
SUVtotal
120
80
Progressive
Disease
60
40
20
Stable
Disease
0
-20
-40
Partial
Response
-60
-80
-100
0
2
4
6
8 10 12 14 16 18 20 22 24 26 28 30 32
Tumor
Vanderhoek et al 2013, J Nucl Med 54: 1188.
10
7/21/2014
WHAT CAN WE DO TO MAKE
PET/CT MORE QUANTITATIVE?
Harmonization of PET/CT imaging
 Harmonization of acquisition
– Minimize limitations due to different scanner hardware
and software
 Harmonization of scanning protocols
– Creating harmonized imaging protocols, which need to
be tuned to specific scanners
 Harmonization of image analysis
– Unifying image analysis protocols, which often means
centralized analysis
 Harmonization of reporting
– Standardized reporting, otherwise not comparable data
Harmonization of acquisition
1 min/bp
5 min/bp
NOISE
10:1
CONTRAST
50:1
20:1
RESOLUTION
11
7/21/2014
SNR harmonization
Gemini TF
DVCT 10 SS
DVCT 14 SS
DVCT 20 SS
DVCT 28 SS
DVCT 35 SS
DVCT 10 SS
SNR2
Gemini TF
DVCT 14 SS
DVCT 20 SS
DVCT 28 SS
DVCT 35 SS
count density
SNR variability: 64%  6%
RC harmonization
After harmonization
1.4
1.4
1.2
1.2
1.0
1.0
0.8
0.8
RC
RC
Before harmonization
0.6
0.4
sphere size
28 mm
17 mm
10 mm
DVCT
Gemini TF
0.6
0.4
0.2
0.2
0.0
0.0
1
2
3
4
5
1
2
mins/bp
3
4
5
mins/bp
DVCT: 256 x 256, 35 SS, 2 IT, 3 PF
Gemini TF: 144 x 144, 33 SS, 3 IT
DVCT: 256 x 256, 14 SS, 2 IT, 4 PF
Gemini TF: 144 x 144, 33 SS, 3 IT
RC variability: 9%  4%
Harmonization of scanning protocols
35
14SS_2IT_4PF
Site1
30
frequency (%)
Site2
25
20
15
10
5
0
50
100
150
200
250
SUVmax
12
7/21/2014
Harmonization of image analysis
• Repeatability results of
double baseline 18F-FDG
PET scans were similar for
all SUV parameters
assessed
• Centralized QA and
centralized image analysis
improved intra-subject CV
from 15.9% to 10.7% for
averaged SUVmax
Velasquez et al. 2009, J Nucl Med 50: 1646
Test/retest reproducibility
Site1
SUVmax
SUVmean
SUVtotal
0.02
0.99
9.1
0.00
0.99
5.1
-0.01
0.998
20.8
Mean diff
ICC
CPD
τ ... inter-patient variation
σ ... intra-patient variation
Z0.975 … the 97.5th percentile of standard normal distribution
σ
… standard deviation for variation due the repeated scans
Test/retest reproducibility
Site1
Site2
Site3
Kurdziel 2012
SUVmax
SUVmean
SUVtotal
Mean diff
ICC
CPD
0.02
0.99
9.1
0.00
0.99
5.1
-0.01
0.998
20.8
Mean diff
ICC
CPD
0.01
0.91
19.5
0.01
0.87
5.8
0.08
0.92
70.1
Mean diff
ICC
CPD
0.03
0.97
2.9
0.01
0.92
2.1
0.12
0.99
18.0
Mean diff
0.09
0.06
0.20
ICC
0.93
0.91
0.95
CPD
52.7
33.2
135.3
13
7/21/2014
Test/retest reproducibility
Site1
Site2
Site3
Kurdziel 2012
SUVmax
SUVmean
SUVtotal
Mean diff
ICC
CPD
0.02
0.99
9.1
0.00
0.99
5.1
-0.01
0.998
20.8
Mean diff
ICC
CPD
0.01
0.91
19.5
0.01
0.87
5.8
0.08
0.92
70.1
Mean diff
ICC
CPD
0.03
0.97
2.9
0.01
0.92
2.1
0.12
0.99
18.0
Mean diff
0.09
0.06
0.20
ICC
0.93
0.91
0.95
CPD
52.7
33.2
135.3
Test/retest reproducibility
Site1
Site2
Site3
Kurdziel 2012
SUVmax
SUVmean
SUVtotal
Mean diff
ICC
CPD
0.02
0.99
9.1
0.00
0.99
5.1
-0.01
0.998
20.8
Mean diff
ICC
CPD
0.01
0.91
19.5
0.01
0.87
5.8
0.08
0.92
70.1
Mean diff
ICC
CPD
0.03
0.97
2.9
0.01
0.92
2.1
0.12
0.99
18.0
Mean diff
0.09
0.06
0.20
ICC
0.93
0.91
0.95
CPD
52.7
33.2
135.3
SUVmax
SUVmean
SUVtotal
0.02
0.99
9.1
0.00
0.99
5.1
-0.01
0.998
20.8
Test/retest reproducibility
Patient reproducibility
Site1
Mean diff
ICC
CPD
Individual lesion reproducibility
Site1
SUVmax
SUVmean
SUVtotal
min, max
min, max
min, max
Mean diff
-0.06, 0.06
-0.03, 0.01
-0.21, 0.17
ICC
0.86, 0.99
0.84, 0.99
0.94, 0.99
CPD
7.8, 23.8
2.9, 8.0
12.5, 151.0
14
7/21/2014
Conclusions
 Qualitative PET/CT sufficient for diagnosis and staging
 Quantitative PET/CT essential for target definition and
treatment response assessment
 Harmonization of PET/CT imaging necessary to
increase quantitative accuracy:
– Harmonization of acquisition
– Harmonization of imaging protocols
– Harmonization of image analysis
– Harmonization of reporting
 Test/retest reproducibility essential to establish
confidence intervals
15