Program Information
MRI-Based Prediction of Recurrence-Free Survival in Breast Cancer Patients Early On in Neoadjuvant Chemotherapy
K Drukker*, H Li , N Antropova , A Edwards , J Papaioannou , M Giger , Univ Chicago, Chicago, IL
Presentations
SU-F-605-7 (Sunday, July 30, 2017) 2:05 PM - 3:00 PM Room: 605
Purpose: To investigate whether MRI-based breast cancer radiomics has the ability to predict recurrence-free survival early in patient treatment with neoadjuvant chemotherapy
Methods: The independent test set included the ACRIN 6657 dynamic contrast-enhanced MRI set with the selection criterion that patients were imaged at baseline (pre-treatment) and after completion of the first cycle of chemotherapy (127 patients, 37 with a recurrence and 90 without). Our method was completely automated apart from manual indication of the approximate tumor center. After automatic tumor segmentation, the volume of the most-enhancing voxels within the tumor at the pre-treatment exam and its change with respect to baseline were calculated. A linear discriminant classifier involving these 2 features was trained on a separate training set of 48 cases to distinguish between patients with and without recurrence of disease. Performance on the independent test set was evaluated in the task of predicting the event of a recurrence using ROC analysis and in the task of associating recurrence-free survival (in days) using a Cox regression model and C-statistics controlling for patient age, race, and hormone receptor status.
Results: The area under the ROC curve in the prediction whether or not a recurrence would occur was 0.71 (standard error 0.05). The C-statistic for the association with recurrence-free survival was 0.72 (0.06, 95% confidence interval [0.62; 0.84]).
Conclusion: The performance of our method in terms of C-statistic rivaled that reported in the ACRIN study (0.72 (95% confidence interval [0.60; 0.84]) which involved the same dataset, manual delineation of the functional tumor volume, and knowledge of the pre-surgical residual cancer burden. Advantages of our method are its reliance only on MRIs early in treatment, and ease of use with the only manual input being the tumor center due to use of completely-automated tumor segmentation, feature-extraction, and merging into a response signature.
Funding Support, Disclosures, and Conflict of Interest: U01CA195564 KD royalties from Hologic. MLG stockholder in Hologic Inc. and Quantitative Insights Inc., co-founder Quantitative Insights Inc., royalties from Hologic Inc., General Electric Company, MEDIAN Technologies, Riverain Technologies LLC, Mitsubishi Corporation and Toshiba Corporation
Contact Email: