Early prediction of breast cancer response to neoadjuvant chemotherapy using multitimepoint and multiregion MRI with graph convolutional network

Neoadjuvant chemotherapy (NAC) is a standard treatment for locally advanced breast cancer, yet patient responses remain highly variable. Early noninvasive prediction of pathologic complete response (pCR) is critical for optimizing treatment strategies. In this study, we propose a graph convolutional network (GCN) framework that integrates multitimepoint and multi-regional information, including tumor core, peritumoral tissue, and intratumoral habitats from baseline and two-cycle dynamic contrast-enhanced MRI (DCE-MRI), to predict pCR. By modeling spatiotemporal heterogeneity, the approach captures predictive signals overlooked by conventional radiomics. The model was trained on a large cohort and externally validated on an independent cohort, achieving superior performance compared with clinical model. These findings highlight the potential of GCN-based models and support precision treatment adaptation in breast cancer patients.