MRI based Clinical-Radiomics Deep Learning Model for Endometrial Cancer Molecular Subtypes Classification: A Multicohort Retrospective Study

Motivation: Accurate classification of molecular subtypes in endometrial cancer (EC) is crucial for prognostic risk assessment and treatment planning. Goal(s): To develop a clinical-radiomics DL model based on MRI for EC molecular subtypes classification. Approach: This retrospective study included 526 EC patients across three institutions. Radiomics features were extracted from multiparametric MRI sequences, and MoCo-v2 was used for self-supervised learning and DL features extracting. Models were built using 12 ML algorithms to select the best-performing model. Results: The clinical-radiomics DL model outperformed others with average AUCs of 0.79 (internal) and 0.74 (external). The highest AUC were 0.86 and 0.81 for p53abn. Impact: The clinical-Radiomics DL Model achieved the optimal performance in classifying molecular subtypes of EC utilizing multiparametric MRI, which demonstrates that preoperative MRI has the potential to help clinicians in accurately assigning patients to their respective molecular subtypes classifications before surgery.