Prediction of traetment response to neoadjuvant chemotherapy in breast cancer with deep learning for multi-time dynamic contrast-enhanced MRI

Motivation: Accurately predicting pCR after neoadjuvant chemotherapy for patients with breast cancer supports personalized care, allowing for tailored surgical strategies and informed decisions on the extent of tissue resection. Goal(s): To develop a fully automated pipeline that leverages deep learning to predict response to neoadjuvant chemotherapy non-invasively using DCE-MRI images taken before and after NAC. Approach: We designed a contrastive learning loss to extract features reflecting tumor dynamics to predict pCR. Results: The proposed model achieved predictive performance on the external dataset with an AUC ranging from 0.7319 to 0.7411, sensitivity between 0.8000 and 0.8333, and specificity from 0.5472 to 0.6491. Impact: Our study demonstrated that analyzing longitudinal DCE-MRI data before and after NAC, integrated with deep learning, can effectively predict breast cancer response to neoadjuvant chemotherapy. This approach holds promise for guiding personalized treatment planning.