CSMC Artificial Intelligence in Medicine (AIM) Program
Guido Germano, PhD
Quantitative gated nuclear imaging
Disclosure: receipt of software royalties from
Cedars-Sinai Medical Center
Cardiac perfusion SPECT: quantitative analysis
Defect extent, severity
& reversibility
Categorical, summed
& normalized scores
TPD
LVEF
ESV and EDV
Diastolic function
RWM & RWT
Phase analysis
Lung/heart ratio
TID ratio
LV shape
LV mass
PERFUSION
QUANTITATION
FUNCTION
QUANTITATION
OTHER
QUANTITATION
INTEGRATED ANALYSIS
Projections (rest & stress)
Short axis (rest & stress)
Gated short axis
(rest & stress)
Cardiac SPECT perfusion/function quantification
QGS/QPS/AutoQUANT (Cedars-Sinai)
Emory Toolbox(Emory Univ.)
4D-MSPECT(Univ. of Michigan)
Main commercially available software
Germano et al., J Nucl Cardiol 2007;14:433
Garcia et al., J Nucl Cardiol 2007;14:420
Ficaro et al., J Nucl Cardiol 2007;14:455
Validation / normal limits info: www.csaim.com/validation
Quantitative measures of perfusion
Segment-based:
Categorical scores (0-4)
Summed scores (combine extent & severity)
Normalized summed scores (indep. of # segments in model)
Pixel-based:
Extent of defect [%]
Severity of defect
Total perfusion deficit (TPD) [Berman, JNC 2004]
Quantitative measures of perfusion
Gated perfusion SPECT quantitation
LVEF = (EDV-ESV)/EDV * 100
EDV= 3D endocardium at ED
ESV= 3D endocardium at ES
WM = endocardial excursion
WT mostly from partial volume effect
Diastolic function = derivative of T-V curve
J Nucl Med 2009; 50:1418–1426Supported by NIH NHLBI R01 grant: R0HL089765
Visual contour QC Expert agreement
Automatic “bad contour” detectionIncorrect LV Incorrect VP
SQC VPO-VQC VPU-VQC
ROC Area
P value
Sensitivity
Specificity
1.0±0.00
< 0.0001
100%
98%
0.91±0.01
< 0.0001
100%
71%
0.97±0.01
< 0.0001
100%
77%
JNM 2009 J Nucl Med. 2009 Sep;50(9):1418-26 Supported by NHLBI R01 grant: R0HL089765
High accuracy for LV
segmentation detection in
MPS demonstrates that
this algorithm may improve
automated and objective
analysis of MPS.
ROC -bad
contour
Detection
Area =1.0
Abnormality thresholds of SPECT EF and volumes
Gender LVEF EDV ESV
Cedars QGS F 51% 102 ml 46 ml
8 fr, Tc-99m (60 ml/m2) (27 ml/m2)
(Sharir, JNC 2006) M 43% 149 ml 75 ml
(75 ml/m2) (39 ml/m2)
Emory ECT F+M 51% 171 ml 70 ml8 or 16 fr
(Garcia, JNC 2007)
4D-MSPECT F 56-60% 118-122 ml 44-42 ml
8-16 fr, Tc-99m (66-68 ml/m2) (25-24 ml/m2)
(Ficaro, JNC 2007) M 47-52% 183-197 ml 91-82 ml
(91-98 ml/m2) (46-41 ml/m2)
LVEF measurement: 8- vs. 16-frame gating
EDV = 114 ml
ESV = 37 ml
LVEF = 68%
EDV = 111 ml
ESV = 41 ml
LVEF = 63%
Diastolic function: normal limits (QGS)
Akincioglu, JNM 2005
90 normal patients
Mean values:
PFR: 2.62 ± 0.46 EDV/s
TTPF: 164.6 ± 21.7 ms
Abnormality thresholds:
PFR < 1.71 EDV/s
TTPF: > 216.7 ms
PFR
ESV
EDV
Phase analysis in gated perfusion SPECT
The time-volume curve
can be broken down by
segment, wall, vessel, etc.
Example
Phase analysis in gated perfusion SPECT
Van Kriekinge JNM 2008
Phase analysis helps predict response to CRT
Boogers et al, JNM 2009
Integration with CTA
Kaufmann PA, Gaemperli O. J Nucl Cardiol 2009;16: 170-72.
Complementary MPI-CTA ?
Slomka et al, Expert Rev Cardiovascular Therapy 2008 Jan;6(1):27-41
Estimate: 10-15% of patients need both
J Nucl Med 2009;
50:1621–1630
Automatically Match Cardiac Phase
•Evaluate cost functional across all phases
where k denotes phase number and n is total number of phase and E is the cost functional
*
1 2 3{1,..., }
arg min ( , , , ),kk n
k E c c c
Phase 9 Phase 16
where
Woo et al. Med Phy 2009 (accepted)
Automatically Match Cardiac
Phase
•Evaluate cost functional across all phases
where k denotes phase number and n is total number of phase and E is the cost functional
*
1 2 3{1,..., }
arg min ( , , , ),kk n
k E c c c
Phase 9 Phase 16
where
Woo et al. Med Phy 2009 (accepted)Woo et al. Med Phys. 2009;36:5467-79.
Automated MPS/CTA image fusion
Translations
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
8.0
9.0
10.0
X Y Z
[mm
]
Rest
Stress
Rotations
0
1
2
3
4
5
6
7
Angle X Angle Y Angle Z
Deg
ree
Rest
Stress
Accuracy of automated CTA-MPI fusion
Rigid
body
1-2 sec
Computing
time
Slomka
et al
JNM2009
QPS analysis: original contours
TPD =1%
SSS =1
#3
QPS analysis with CTA-guided contours
RCA TPD
7 %
SPECT/CTA volume/surface fusionCATH: proximal RCA 100%, no significant LAD, LCX disease
#3
RCARCA
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.2 0.4 0.6 0.8 1
1-Specificity
Se
ns
itiv
ity
LAD-TPD
CTA guided
LAD TPD
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.2 0.4 0.6 0.8 1
1 - Specificity
Se
ns
itiv
ity
LCX-TPD
CTA guided
LCX TPD
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.2 0.4 0.6 0.8 1
1 - SpecificityS
en
sit
ivit
y
RCA-TPD
CTA guided
RCA TPD
**
CTA-guided MPS quantification
LCX RCA
N=35
patients
J Nucl Med 2009; 50:1621–1630
CTA-MPS fusion: possible applications
comparison of CTA and MPS perfusion imaging
Cedars-Sinai Medical Center
Normal ROI, plaque lesion limits, fast luminal centerline within limits
Automatic CTA measurements/annotation
NCPNCP volume 112 mm3
Dey et al, ACC 2010
Quantitative PET
Motion-frozen perfusion imaging
Images obtained at Cedars-Sinai Project PI: Dan Berman
Lantheus clinical trial 18F- flurpiridaz (BMS747158)
EF stress QPET/QGS Rb-82 vs. post-stress CTA
EF QPET vs. CTA
10
20
30
40
50
60
70
80
90
10 20 30 40 50 60 70 80 90
EF-CT
QP
ET
EF
-str
Identity
Difference Plot
-20
-15
-10
-5
0
5
10
15
10 20 30 40 50 60 70 80 90
Mean of All
Dif
fere
nce (
QP
ET
EF
-str
- E
F-C
T)
Identity
Bias (-3.1)
95% Limits of agreement
(-17.2 to 11.1)
Bias = -3%
r =0.90
EF QGS vs CTA
10
20
30
40
50
60
70
80
90
10 20 30 40 50 60 70 80 90
EF-CT
QG
S E
F-s
tr
Identity
Difference Plot
-30
-25
-20
-15
-10
-5
0
5
10 20 30 40 50 60 70 80 90
Mean of All
Dif
fere
nce (
QG
S E
F-s
tr -
EF
-CT
)
Identity
Bias (-13.6)
95% Limits of agreement
(-28.0 to 0.9)
Bias= -14%
r =0.87
Slomka, Germano, Bengel, J Nucl Med. 2009; 50 (Supplement 2):1167
PET/CT fusion ED
Slomka, Germano, Bengel et al, SNM 2009
PET/CT fusion ES
Gated PET vs. Gated CTA
GATED SPECT FOR DIASTOLIC
PERFUSION ASSESSMENT
“Motion frozen” gated myocardial perfusion SPECT
Summed images (8 frames)
Conventional summation Summation after morphing
to ED template
Slomka et al., JNM 2004
“Motion frozen” gated myocardial perfusion SPECT
Slomka et al., JNM 2004
Displacement
vectors are used for
warping ES onto
ED
MF accuracy in obese population
0102030405060708090
100
Sensitivity Specificity Accuracy0
102030405060708090
100
Sensitivity Specificity Accuracy
S-TPD
MF-TPD
P = NS P = NS
92% 92%
59%
82% 82%88%
P < 0.05P < 0.05
P < 0.05
P < 0.05
93% 95%
55%
77%80%
89%
A B≥50% Stenosis ≥70% Stenosis
N=90
Suzuki Y et al. J Nucl Med. 2008 Jul;49(7):1075-9
QUANTITATIVE RVEF
FROM PERFUSION SPECT
Case example
Top Related