Fig. 1. Overall workflow of data collection, evaluation of the standalone performance of deep learning (DL) model, DL-assisted pathologists’ revision, and the neoadjuvant chemotherapy response prediction.

DL standalone performance was evaluated in cases that were concordant among pathologists (N = 210). In cases where there was disagreement between pathologists and DL (N = 226), pathologists performed the DL-assisted revision. Then the results were applied to neoadjuvant therapeutic response prediction.