Machine learning radiomics for prediction of posterior cranial fossa ependymoma PFA and PFB subgroups

Motivation: Ependymoma (EP) is a prevalent intracranial tumor in children, with significant implications of its molecular subtypes on treatment outcomes. However, there is currently a lack of non-invasive methods to distinguish between these molecular subtypes of EP. Goal(s): To develop an non-invasive method to distinguish between these molecular subtypes of EP. Approach: MRI data from 43 ependymoma patients was analyzed in this study, with radiomic features and machine learning used to distinguish molecular subtypes. Results: The AUC for the proposed model was 0.840 (0.723-0.957) in the training group and 0.900 (0.698-1.000) in the test group. Impact: The proposed machine learning model effectively distinguishes between the molecular subtypes of ependymoma, showing strong performance and enhancing diagnostic accuracy, which is expected to provide valuable insights for clinical decision-making.