Artificial Intelligence-Driven MRI for Cervical Nodal Metastasis Detection in Oral Squamous Cell Carcinoma: A Hierarchical Meta-Analysis of Diagnostic Accuracy.

Abstract

Artificial intelligence (AI) applied to magnetic resonance imaging (MRI) may improve detection of cervical lymph node metastases in oral squamous cell carcinoma (OSCC) but is heterogeneous. A systematic review identified observational studies (from 2000) evaluating AI-based MRI in adults with histopathologically confirmed OSCC. Risk of bias was assessed with QUADAS-AI. Diagnostic performance was synthesized using a hierarchical bivariate model. Publication bias and certainty of evidence were assessed using Deeks' test and GRADE. Twelve studies were included; seven datasets (548 participants) were meta-analyzed. Pooled sensitivity was 0.72 (95% CI: 0.62-0.80) and specificity 0.79 (95% CI: 0.73-0.83), with AUC 0.82 and diagnostic odds ratio 9.42. Heterogeneity is mainly related to threshold effects. No significant publication bias was detected (p = 0.536). Evidence certainty was low. AI-assisted MRI shows moderate diagnostic performance. Multicenter validation is required before clinical implementation. AI-supported MRI may serve as an adjunctive tool to improve preoperative risk stratification of cervical lymph node metastasis in OSCC.