Multiparametric MRI Habitat Models Improve Non-Invasive Risk Stratification and Prognosis Prediction in Endometrial Carcinoma

Motivation: Endometrial carcinoma (EC) risk stratification and prognosis is essential for treatment planning, yet conventional radiomics often lacks biological interpretability. Goal(s): To assess if MRI-based habitat analysis can be a promising non-invasive tool for risk stratification and prognosis prediction in EC. Approach: We developed structural and functional habitat models using T2WI, ADC, and CE-T1WI MRI sequences and used pathological examination results as the gold standard to evaluate habitat models' performance in predicting EC risk. Results: Habitat models, particularly GMM- and K-means-based models, achieved comparable or improved predictive performance relative to the clinical model, furthermore offered non-invasive cellularity and vascularity information, enhancing EC risk assessment. Impact: MRI-based habitat analysis and imaging provides a non-invasive method for predicting EC risk levels and prognosis at the celluar level, aiding treatment decision-making.