Prediction of Early Neoadjuvant Chemotherapy Response of Breast Cancer through Deep Learning-Based Pharmacokinetic Quantification of DCE-MRI

Motivation: Accurate prediction of pCR to neoadjuvant chemotherapy in breast cancer is crucial for personalized treatment. However, the generalizability of prediction algorithms from DCE-MRI is often hindered by imaging protocol discrepancy. Goal(s): This study aims to enhance the generalizability of pCR prediction through deep learning-based pharmacokinetic mapping. Approach: Using a previously developed DL model, pharmacokinetic parameters were retrospectively estimated from clinical multi-phasic DCE-MRI data across four public datasets, which were subsequently applied in radiomic analysis to predict pCR. Results: Our model demonstrated superior and consistent pCR prediction performance across datasets compared to conventional functional tumor volume (FTV) enhancement maps. Impact: This study introduces a quantitative, generalizable approach to early prediction of neoadjuvant chemotherapy (NAC) response in breast cancer, unlocking noninvasive imaging biomarkers with enhanced predictive accuracy and generalizability, thereby facilitating personalized treatment decisions without modifying clinical imaging protocols.