An AI Model Predicts the Efficacy of Neoadjuvant Chemotherapy for Breast Cancer: a Multicenter, Bidirectional Cohort Study

This is a multicenter, bidirectional cohort study. Retrospective training cohort, retrospective validation cohort and prospective test cohort were designed. Data of patients treated in the North Ward of Sun Yat-sen Memorial Hospital of Sun Yat-sen University from January 1, 2002 to August 31, 2023 were retrospectively collected for training cohort, and data of patients treated in the South ward of Sun Yat-sen Memorial Hospital of Sun Yat-sen University for internal validation cohort; Data on patients treated at external centers between January 1, 2002 and August 31, 2023 were retrospectively collected for external validation cohort. Data on patients admitted to Sun Yat-sen Memorial Hospital at Sun Yat-sen University after January 1, 2024 were prospectively collected for the test cohort. Patient data collected included: pathological images and report texts of breast puncture specimens before neoadjuvant chemotherapy, breast magnetic resonance images and report texts before neoadjuvant chemotherapy, postoperative pathological reports and clinical information, etc.. An artificial intelligence large language model based on multi-task and multi-modal data integration was established to accurately predict the efficacy of neoadjuvant chemotherapy for breast cancer, and its predictive efficacy was tested by retrospective validation cohort and prospective double-blind test cohort. The retrospective cohort of this study was followed up to collect clinical data, magnetic resonance imaging and reports, and surgical pathology reports of patients, etc.. When patients had disease recurrence, the DFS time of patients was recorded, and when patients did not have disease recurrence, the last follow-up time was recorded. Baseline data survey was completed during hospitalization of prospective cohort patients. Pathological reports of breast tumors surgically removed after neoadjuvant chemotherapy were obtained during follow-up, as well as the time of disease recurrence and the time of death of patients. Clinical information such as magnetic resonance imaging and reports were collected during follow-up. Follow-up until the end of the 2-year study.