Initial results of combining DWI with DCE MRI to predict pathologic complete response in I-SPY 2

1. The role of diffusion-weighted MRI in the
prediction of response in I-SPY 2 TRIAL
9/29/2018
Wen Li, Lisa J Wilmes, David C Newitt, Ella F Jones, Savannah Partridge, John
Kornak, Jessica Gibbs, Bo La Yun, Matthew Tanaka, Laura J Esserman, Nola Hylton
ISMRM 2018

2. Lisa Wilmes:
I have no financial interests or relationships to disclose with regard to the
subject matter of this presentation.
Declaration of
Financial Interests or Relationships

3. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL3
Content
 Background and Purpose
 Methods
 Results
 Conclusions
 Discussions

4. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL4
I-SPY 2 TRIAL
Investigation of Serial Studies to Predict Your Therapeutic Response through
Imaging and Molecular Analysis
 A response-adaptive phase II trial testing
novel agents for breast cancer
 I-SPY 2 opened in March 2010; 22 sites
(>1500 patients screened)
 Drugs ”graduate” within subtypes defined by
hormone receptor (HR) status, HER2
status, and Mammaprint score
 4 longitudinal MRIs are acquired during
treatment
 The primary endpoint is pathologic complete
response (pCR) at surgery
I-SPY 2 schema
MRI1
(DCE & DWI)
MRI0
(DCE & DWI)
MRI2
(DCE & DWI)
MRI3
(DCE & DWI)

5. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL5
Quantitative breast MRI
 Functional tumor volume (FTV) measured from
DCE-MRI was found to predict pathologic
complete response (pCR) and recurrence free
survival (RFS) in I-SPY 1
 In I-SPY 2, FTV is used to adjust patient
randomization ratios and to evaluate drug efficacy
 DW-MRI was included in I-SPY 2 to investigate the
predictive value of apparent diffusion coefficient
(ADC) by subtype
DCE DWI ADC
DCE FTV

6. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL6
Purpose
 To investigate the additive predictive value of DW-MRI to
DCE-MRI by subtype for prediction of response in the I-SPY 2
TRIAL

7. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL7
Content
 Background and Purpose
 Methods
 Results
 Conclusions
 Discussions

8. 8
DW-MRI acquisition
The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL
Good quality
b=800 ADC map
b=800 ADC map
 Data acquired from 21 sites
- 1.5 and 3 T
- 3 scanner vendors
 DW-MRI was acquired prior to DCE-MRI
 SE-EPI sequence with fat suppression with
total scan time ≤ 5 min
 Quality ranking: exclude, poor, good
Poor quality

9. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL9
ADC measurement
 ADC maps were generated using
2 b-values (0 and 800)
 Multi-slice ROI was manually
delineated on the ADC map for the
whole tumor
0.8
1.29
2.26 2.19
Mean tumor ADC (x10-3mm2/sec)
T0 T1 T2 T3
DW-MRI ADC map DCE-MRI
T0
Pre-treatment
T1
Early treatment
T2
Inter-regimen
T3
Pre-surgery

10. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL10
Imaging predictors
DCE & DWI DCE & DWI DCE & DWI DCE & DWI

11. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL11
Imaging predictors
FTV0, ADC0 FTV1, ADC1 FTV2, ADC2 FTV3, ADC3
DCE & DWI DCE & DWI DCE & DWI DCE & DWI
%∆FTV0_1, %∆ADC0_1
%∆FTV0_2, %∆ADC0_2
%∆FTV0_3, %∆ADC0_3

12. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL12
Imaging predictors
FTV0, ADC0 FTV1, ADC1 FTV2, ADC2 FTV3, ADC3
pCR is the primary
endpoint of I-SPY 2
DCE & DWI DCE & DWI DCE & DWI DCE & DWI
%∆FTV0_1, %∆ADC0_1
%∆FTV0_2, %∆ADC0_2
%∆FTV0_3, %∆ADC0_3

13. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL13
Statistical analysis
 Binary outcome: pCR vs. non-pCR
 Single predictor analysis
- Wilcoxon rank sum test: difference in MR metrics for
pCR vs. non-pCR
- Area under the ROC curve (AUC) for predicting pCR
 Multiple predictor analysis
- Logistic regression model was evaluated
- AUC was calculated with 10-fold cross validation

14. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL14
Content
 Background and Purpose
 Methods
 Results
 Conclusions
 Discussions

15. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL15
Patient characteristics
 415 patients from 4 completed I-SPY 2 drug arms
 354 patients with measurable ADC
36%
17%10%
37%
Subtype distribution
HR+/HER2-
HR+/HER2+
HR-/HER2+
HR-/HER2- 0%
20%
40%
60%
80%
100%
pCR rate by subtype
pCR non-pCR

16. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL16
Single predictor analysis
 In general, ADC increases more in pCR than in non-pCR and the differences increase over treatment
 In the full cohort, %∆ADC shows statistically significant prediction of pCR at all time points
 %∆ADC appears to predict pCR at inter-regimen and pre-surgery in all subtypes, though HR+/HER2+ does
not reach statistical significance
-50
0
50
100
Full
H
R
+/H
E
R
2-
H
R
+/H
E
R
2+
H
R
-/H
E
R
2+H
R
-/H
E
R
2-
%DADC0_3
outcomes
non_pCR
pCR
*** ***
* p<0.05
**p<0.0001
Percent change (%∆) in ADC from baseline
Early treatment Inter-regimen Pre-surgery
-50
0
50
100
Full
H
R
+/H
E
R
2-
H
R
+/H
E
R
2+
H
R
-/H
E
R
2+H
R
-/H
E
R
2-
%DADC0_1
outcomes
non_pCR
pCR
**
-50
0
50
100
Full
H
R
+/H
E
R
2-
H
R
+/H
E
R
2+
H
R
-/H
E
R
2+H
R
-/H
E
R
2-
%DADC0_2
outcomes
non_pCR
pCR
** * * *

17. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL17
Multiple predictor model
Combining predictors with logistic regression model
+
FTV0
+
%∆FTV0_2
subtype
+
%∆ADC0_3
+
ADC0
pCR
1. Select individual predictors for inclusion using likelihood ratio test (p<0.05)
2. Include predictors that increase AUC
3. Fitted separate models using only predictors up to T0, T1, T2, T3, respectively
FTV0
+
subtype
+
ADC0
%∆FTV0_1
FTV0
+
subtype
+
+
%∆ADC0_1
%∆FTV0_2
+
FTV0
+
subtype
+
%∆ADC0_2

18. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL18
Full cohort AUCs by time point
Pre-treatment T0 n=346 Early treatment T1 n=323
Inter-regimen T2 n=282 Pre-surgery T3 n=257
AUCs for ROC curves on the left
AUC (95% CI) p
T0
FTV+subtype 0.71 (0.68, 0.75)
<0.0001
FTV+subtype+ADC 0.70 (0.66, 0.73)
T1
FTV+subtype 0.73 (0.70, 0.77)
0.63
FTV+subtype+ADC 0.74 (0.70, 0.77)
T2
FTV+subtype 0.74 (0.71, 0.78)
0.015
FTV+subtype+ADC 0.76 (0.73, 0.80)
T3
FTV+subtype 0.73 (0.70, 0.77)
<0.0001
FTV+subtype+ADC 0.80 (0.76, 0.82)
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
False positive rate
Truepositiverate
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
FTV0 + subtype
FTV0 + ADC0 + subtype
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
False positive rate
Truepositiverate
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
%DFTV0_1 + FTV0 + subtype
%DFTV0_1 + FTV0 + %DADC0_1 + subtype
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
False positive rate
Truepositiverate
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
%DFTV0_2 + FTV0 + subtype
%DFTV0_2 + FTV0 + %DADC0_2 + subtype
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
False positive rate
Truepositiverate
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
%DFTV0_2 + FTV0 + subtype
%DFTV0_2 + FTV0 + %DADC0_3 + ADC0 + subtype

19. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL19
Full cohort AUCs by time point
Pre-treatment T0 n=346 Early treatment T1 n=323
Inter-regimen T2 n=282 Pre-surgery T3 n=257
AUCs for ROC curves on the left
AUC (95% CI) p
T0
FTV+subtype 0.71 (0.68, 0.75)
<0.0001
FTV+subtype+ADC 0.70 (0.66, 0.73)
T1
FTV+subtype 0.73 (0.70, 0.77)
0.63
FTV+subtype+ADC 0.74 (0.70, 0.77)
T2
FTV+subtype 0.74 (0.71, 0.78)
0.015
FTV+subtype+ADC 0.76 (0.73, 0.80)
T3
FTV+subtype 0.73 (0.70, 0.77)
<0.0001
FTV+subtype+ADC 0.80 (0.76, 0.82)
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
False positive rate
Truepositiverate
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
FTV0 + subtype
FTV0 + ADC0 + subtype
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
False positive rate
Truepositiverate
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
%DFTV0_1 + FTV0 + subtype
%DFTV0_1 + FTV0 + %DADC0_1 + subtype
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
False positive rate
Truepositiverate
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
%DFTV0_2 + FTV0 + subtype
%DFTV0_2 + FTV0 + %DADC0_2 + subtype
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
False positive rate
Truepositiverate
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
%DFTV0_2 + FTV0 + subtype
%DFTV0_2 + FTV0 + %DADC0_3 + ADC0 + subtype

20. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL20
Subtype AUCs at inter-regimen
Full cohort (no subtype adj.)
n=282 pCR rate: 33%
HR+/HER2-
n = 101
pCR rate: 13%
HR+/HER2+
n = 46
pCR rate: 28%
HR-/HER2+
n = 25
pCR rate: 56%
HR-/HER2-
n = 110
pCR rate: 47%
—%∆ADC0_2
—%∆FTV0_2 + FTV0
—%∆FTV0_2 + FTV0 + %∆ADC0_2
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
False positive rate
Truepositiverate
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
AUC: 0.65
AUC: 0.66
AUC: 0.67
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
False positive rate
Truepositiverate
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
AUC: 0.71
AUC: 0.57
AUC: 0.58
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
False positive rate
Truepositiverate
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
AUC: 0.66
AUC: 0.74
AUC: 0.72
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
False positive rate
Truepositiverate
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
AUC: 0.62
AUC: 0.64
AUC: 0.58
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
False positive rate
Truepositiverate
0.0 0.2 0.4 0.6 0.8 1.0
0.00.20.40.60.81.0
AUC: 0.77
AUC: 0.59
AUC: 0.73

21. Presentation Title21
Content
 Background and Purpose
 Methods
 Results
 Conclusions
 Discussions

22. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL22
Conclusions
 As single predictors, both FTV and ADC can predict pCR as
early as after 3 cycles of neoadjuvant chemotherapy (T1) in
the full cohort
 Models combining FTV and ADC yielded the highest cross-
validated AUCs at inter-regimen and pre-surgery time points
in the full cohort
 The association between percent change of FTV and ADC
and pCR varies among breast cancer subtypes

23. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL23
Content
 Background
 Methods
 Results
 Conclusions
 Discussion

24. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL24
Discussion
 I-SPY 2 is an on-going multi-center multi-treatment clinical trial
- Differences in system manufacturers and field strength
- DW-MRI image quality highly variable across clinical sites in I-SPY 2
- Patients from multiple treatment arms were combined
 Analysis by subtype, the goal of the study, is continuing
- Sample size is limited in certain breast cancer subtype, such as HR-/HER2+
- 11 drugs have graduated and available for analysis
 Improvements in breast DW-MRI acquisition and analysis may
lead to higher predictive performance

25. The role of diffusion-weighted MRI in the prediction of response in I-SPY 2 TRIAL25
Acknowledgements
 Patients who participated in I-SPY 2
 All members of I-SPY TRIAL Investigators Network
 ACRIN Core Laboratory
 Funding: U01 CA151235; R01 CA132870