EPCO-09. INTEGRATIVE MODELING OF EPIGENETIC AND MRI FEATURES IN MENINGIOMA PROGNOSTICATION

Abstract

Abstract Meningiomas are the most common CNS tumors and are often benign, but a subset recurs or behaves aggressively. Traditional grading systems and imaging features have limited accuracy in predicting recurrence. Prior work, including ours, has shown that DNA methylation signatures can reliably classify meningiomas based on recurrence risk. We hypothesized that combining methylation-based markers with imaging features would enhance recurrence prediction beyond either method alone. In this study, we analyzed 20 patients with primary meningioma, encompassing both sexes, diverse grades, and histological subtypes, all with at least five years of follow-up. Patients were categorized as confirmed recurrence (CR, n = 16) or no recurrence (CNR, n = 4) based on longitudinal imaging. Fourteen imaging features linked to recurrence were extracted from preoperative MRI. We applied our Meningioma-epigenetic Liquid Biopsy (MeLB) classifier, developed from 38 methylation markers profiled using the Illumina EPICv1 array, to assign each sample a high- or low-risk recurrence score. We built three random forest models: one using imaging features alone, one using methylation data alone, and one combining both. Model performance was evaluated using 1,000 training/testing cycles with area under the curve (AUC) comparisons. Predictors were ranked based on association with CR/CNR and MeLB scores using the Kruskal-Wallis test and correlation analysis. Performance comparisons were made using the Wilcoxon rank-sum test. The integrated model significantly outperformed imaging alone in predicting recurrence (p < 0.001), and both individual approaches in predicting MeLB scores (p < 0.001). Key predictors included specific methylation signatures and standard imaging features such as tumor size, hyperostosis, multifocality, calcification, irregular margins, and parenchymal invasion. This proof-of-concept study demonstrates that integrating epigenetic classifiers with radiologic features improves prediction of meningioma recurrence and supports further validation in larger, independent cohorts.

Ähnliche Arbeiten