Machine learning-based models and radiomics: can they be reliable predictors for meningioma recurrence? A systematic review and meta-analysis

Abstract

Background: Predicting recurrence in meningioma patients is vital for improving long-term outcomes and tailoring personalized treatment strategies. While traditional diagnostic methods have advanced, accurately forecasting recurrence remains a persistent and critical challenge. This study explores the cutting-edge application of artificial intelligence (AI)-based models, which seamlessly integrate clinical, radiological, and pathological data, offering a transformative approach to enhancing the reliability and precision of recurrence prediction.

Methods: Eligible studies were identified through a comprehensive search of the Web of Science, Scopus, PubMed, and Embase databases. Extracted and synthesized metrics for analysis included accuracy, sensitivity, specificity, precision, F1 score, and area under the curve (AUC). Out of 2,971 studies screened, six met the inclusion criteria for systematic review, and three were included in the meta-analysis.

Results: The pooled sensitivity and specificity of AI models were 0.86 [95% CI: 0.78-0.92] and 0.86 [95% CI: 0.81-0.90], respectively. The positive diagnostic likelihood ratio (DLR) was 6.33 [95% CI: 4.42-9.08], and the negative DLR was 0.16 [95% CI: 0.09-0.27]. The diagnostic odds ratio (DOR) was estimated at 40.11 [95% CI: 19.30-83.37], with a diagnostic score of 3.69 [95% CI: 2.96-4.42] and a pooled area under the curve (AUC) of 0.93 [95% CI: 0.90-0.95]. Subgroup analysis showed comparable sensitivity (RF: 0.88; LR: 0.84) and specificity (RF: 0.84; LR: 0.84) with no significant heterogeneity (I² = 0%).

Conclusions: These findings highlight the potential of AI-based models to predict meningioma recurrence, offer superior diagnostic accuracy, and aid clinical decision-making. Integrating clinical, radiological, and pathological data through AI-driven models demonstrates substantial promise in enhancing the reliability and efficiency of recurrence forecasting.

Keywords: Artificial intelligence; Deep learning; Diagnostic models, meningioma, recurrent, meta-analysis; Machine learning.