Development and Validation of an Explainable MRI-Based Habitat Radiomics Model for Predicting p53-Abnormal...

Area of Science:

• Radiology
• Oncology
• Medical Imaging

Background:

• Endometrial cancer (EC) molecular subtyping is crucial for treatment.
• p53-abnormal (p53abn) is a key molecular subtype of EC.
• Accurate prediction of p53abn EC aids in personalized treatment strategies.

Purpose of the Study:

• To develop and validate an MRI-based habitat radiomics model (HRM) for predicting p53abn EC.
• To compare the performance of the HRM against whole-region radiomics models (WRM) and clinical models (CM).
• To enhance the interpretability of predictive models using SHAP (SHApley Additive exPlanations) values.

Main Methods:

• Retrospective analysis of EC patients from three hospitals.
• Tumor segmentation into habitat sub-regions using K-means on diffusion-weighted imaging (DWI) and contrast-enhanced (CE) MRI.
• Extraction of radiomics features from T1WI, T2WI, DWI, and CE images.
• Development of predictive models using logistic regression, support vector machines, and random forests.
• Validation using receiver operating characteristic (ROC) curves and DeLong's test.

Main Results:

• The HRM demonstrated superior performance with the highest Area Under the Curve (AUC) values: 0.855 (training), 0.769 (test 1), and 0.766 (test 2).
• The WRM and CM showed significantly lower AUCs across all cohorts.
• The habitat radiomics approach, combined with machine learning and SHAP, provided interpretable insights into risk factor contributions.

Conclusions:

• The MRI-based HRM is a highly effective tool for predicting p53abn EC.
• Habitat radiomics combined with machine learning and SHAP offers a powerful and interpretable method for EC subtyping.
• This approach can assist clinicians in making more informed decisions for EC management.