Diagnostic performance of artificial intelligence models for predicting glioma recurrence using pre-operative MRI: A systematic review and meta-analysis

Background Glioma recurrence after initial treatment remains a significant clinical challenge. Recent advances in artificial intelligence (AI), including radiomics and deep learning (DL), offer promising tools for improving non-invasive prediction of glioma recurrence. This study aims to evaluate the methodology and diagnostic accuracy of these models in predicting glioma recurrence using preoperative MRI. Methods A comprehensive search of PubMed, Scopus, Web of Science, and Embase was performed up to April 15, 2025, to identify studies utilizing MRI-based AI models for recurrence prediction. Study quality was assessed using the METRICS tool. Pooled estimates of sensitivity, specificity, likelihood ratios, and area under the curve (AUC) were calculated. Between-study heterogeneity was assessed using the I² statistic, and subgroup analysis was conducted to explore potential sources of heterogeneity. Publication bias was evaluated using Deek's funnel plot. Results Eleven studies met inclusion criteria, of which eight contributed to meta-analysis. The pooled sensitivity and specificity were 0.75 (95% CI: 0.67-0.82) and 0.76 (95% CI: 0.66-0.83), respectively, with an AUC of 0.79 (95% CI: 0.75-0.82). Hand-crafted radiomics-based models demonstrated significantly higher pooled specificity than end-to-end DL models (0.78 vs. 0.44, p