Multimodal MRI-based radiomics model for molecular subtypes prediction and prognosis evaluation of posterior fossa ependymoma

Motivation: The WHO classifies posterior fossa ependymomas (PF-EPNs) into Groups A (PFA) and B (PFB) based on DNA methylation patterns, resulting in distinct clinical outcomes but posing challenges for molecular diagnosis. Goal(s): We aimed to develop a radiomics model based on multimodal MRI to predict PF-EPN subtypes and prognosis. Approach: Using a large cohort, we developed a support vector machine (SVM) classifier that utilizes T1WI, T2WI, CET1WI and age to differentiate PFA from PFB. Results: The model achieved AUCs of 0.937, 0.926, with accuracies of 0.927, 0.875 in internal and prospective test sets, respectively, and successfully stratified PF-EPNs into high- and low-risk groups. Impact: The multimodal MRI-based radiomics model predicts molecular subtypes of PF-EPNs and enables risk stratification, provides non-invasive insights for clinical treatment decisions. This approach facilitates patient selection for targeted genetic analysis, enhances treatment precision, and improves monitoring and family counseling.