Artificial intelligence for the detection and diagnosis of oral and maxillofacial lesions: evidence, limitations, and future directions
TL;DR
Current data suggest that AI may serve as a valuable adjunct to clinical decision-making, with potential to reduce diagnostic variability and support non-specialist practitioners.
Abstract
Abstract Artificial intelligence (AI) has emerged as a promising tool to support or perform specific oral diagnostic tasks, particularly through advances in machine learning and deep learning. This comprehensive review aimed to synthesize current evidence and explore future directions for AI-driven or AI-supported oral diagnostic workflows across four target pathologies: odontogenic cysts, odontogenic tumors, oral potentially malignant disorders (OPMDs), and oral squamous cell carcinoma (OSCC). A structured search strategy combining MeSH terms and free-text keywords was applied, encompassing target conditions, AI methodologies, diagnostic performance metrics, and clinician comparator groups. Across all pathologies, AI models demonstrated encouraging performance, frequently approaching that of experienced clinicians, particularly in image-based detection and classification tasks. In some contexts, improved sensitivity was observed, suggesting potential value in early disease detection. However, findings were highly variable and often limited by methodological constraints, including retrospective study designs, small or curated datasets, lack of external validation, and heterogeneity in reporting metrics such as sensitivity, specificity, accuracy, and area under the curve. Importantly, no consistent evidence supports the superiority of AI over clinicians across all performance measures. Instead, current data suggest that AI may serve as a valuable adjunct to clinical decision-making, with potential to reduce diagnostic variability and support non-specialist practitioners. Future research should prioritize prospective, multicenter studies with standardized methodologies, robust external validation, and evaluation of real-world and patient-centered outcomes. While AI holds significant promise, its routine clinical implementation for oral diagnostic tasks remains premature, requiring further validation, transparency, and integration into clinical workflows.