Radiomics-based models to predict IDH mutation status and prognosis in gliomas using MRI: a multicenter study.

Area of Science:

• Neuro-oncology
• Medical imaging analysis
• Machine learning in medicine

Background:

• Gliomas exhibit significant heterogeneity, impacting prognosis and treatment response.
• Isocitrate dehydrogenase (IDH) mutation is a critical prognostic biomarker in gliomas.
• Radiomics offers quantitative feature extraction from medical images for noninvasive analysis.

Purpose of the Study:

• To develop and validate radiomics-based machine learning models for predicting IDH mutation status in gliomas.
• To assess the prognostic value of radiomics features for overall survival in glioma patients.
• To create a nomogram for personalized risk stratification and survival estimation.

Main Methods:

• Extraction of 1,820 radiomics features from 638 T2-weighted MRI scans (discovery and validation cohorts).
• Development and external validation of machine learning models (logistic regression, random forest) for IDH mutation prediction.
• Computation of a radiomics risk score and assessment of its prognostic value using Cox regression and Kaplan-Meier analysis.

Main Results:

• Radiomics models achieved high accuracy (AUC 0.90/0.68) for IDH mutation prediction.
• A significant association was found between high radiomics risk score and shorter overall survival (P <0.001).
• Age, radiomics risk score, and IDH mutation status were independent prognostic factors; the nomogram demonstrated good predictive performance (C-indices 0.83/0.75).

Conclusions:

• Radiomics from preoperative MRI can noninvasively predict IDH mutation status and overall survival in gliomas.
• The radiomics risk score serves as an independent prognostic factor for personalized risk stratification.
• External validation confirmed the generalizability and clinical utility of the radiomics models.