Research on a Conditional Generative Adversarial Network-Based Model for Predicting Neoadjuvant Chemotherapy Response in Breast Cancer

This study proposes a dual-temporal feature fusion model based on a conditional generative adversarial network (cGAN) to achieve early prediction of pathological complete response (pCR) after neoadjuvant chemotherapy (NAC) in breast cancer. By constructing a treatment response simulator, the model utilizes cGAN to generate individualized virtual post-treatment ultrasound imaging features and integrates multi-scale features from pre- and post-treatment phases through spatiotemporal alignment and information complementarity. Experiments involving ultrasound imaging data from 171 breast cancer patients demonstrated that the model achieved an AUC of 0.873 on the test set, with sensitivity and specificity of 70.0% and 90.6%, respectively, outperforming traditional single-time-point models. The method advances the therapeutic response prediction window by 4–6 weeks through feature generation technology and employs a dual-path ResNet architecture to hierarchically fuse tumor heterogeneity features, significantly improving computational efficiency compared to conventional approaches. This research overcomes the limitations of static radiomics by providing a novel decision-support tool for clinical personalized treatment planning, though further validation through multicenter data is required to confirm the model’s generalizability.