Deep neural network-based interpretable prediction model for survival outcomes in female breast cancer patients: Integrating biomechanical perspectives with clinicopathological features

Background: This study integrates biomechanical perspectives with clinicopathological data to develop a DNN model for survival prediction. By linking tumor size and lymph node status to biomechanical drivers such as solid stress and cell migration forces, we aim to uncover the mechanobiological mechanisms underlying prognosis heterogeneity. Methods: We analyzed data from 37,917 patients in the SEER database, encompassing clinical characteristics, pathological features, and treatment details. The DNN, featuring an attention mechanism, was evaluated using metrics such as accuracy, precision, recall, F1 score, and Area Under Curve (AUC). Interpretability techniques were applied to identify prognostic factors. Results: The DNN model achieved F1 scores of 0.928 and 0.935 for validation and test sets, respectively, with an AUC of 0.96, surpassing traditional models. Key factors identified included regional lymph node positivity, tumor size, and tumor grade, with a notable negative correlation between regional lymph node positivity and survival. Conclusions: DNN models with attention mechanisms demonstrate superior predictive performance and valuable interpretability in identifying critical prognostic factors.