Abstract
BACKGROUND: Artificial intelligence (AI) and machine learning (ML) models have recently emerged as promising tools for enhancing diagnostic accuracy. OBJECTIVE: To evaluate the diagnostic accuracy of AI/ML models in detecting TMDs through a systematic review and meta-analysis of existing literature. METHODS: A comprehensive search of electronic databases was conducted to identify studies assessing the diagnostic performance of AI/ML models in TMD diagnosis (PROSPERO-CRD420251035080). Data extraction and quality assessment were conducted independently by two reviewers using the AXIS tool for cross-sectional and Newcastle-Ottawa Scale for cohort studies. Meta-analysis of diagnostic accuracy was performed using pooled sensitivity, specificity, diagnostic odds ratio, and area under the curve. Statistical heterogeneity was assessed with the I(2) statistic. RESULTS: The systematic search identified 368 articles, of which 12 studies met inclusion criteria after screening. Risk of bias assessment showed most observational studies had low to unclear bias, while cross-sectional studies varied from moderate to high quality. Five studies were eligible for meta-analysis and they revealed that AI and machine learning models achieved a pooled sensitivity of 87.1 %(95 %CI:84.9 %-89.2 %) and specificity of 87.0 %(95 %CI:84.8 %-89.2 %) for TMD diagnosis. The diagnostic odds ratio was 45.1(95 %CI:30.5-66.8), with an area under the ROC curve of 0.96, indicating excellent diagnostic accuracy. Moderate heterogeneity I(2) = 38.7 %. CONCLUSION: AI/ML models demonstrate excellent accuracy in differentiating patients with and without TMDs, reinforcing their potential as reliable diagnostic aids in clinical and screening settings. However, variability in input features and lack of standardized model development protocols highlight the need for future research focusing on validation across diverse populations and harmonization of diagnostic criteria.