Diagnostic Accuracy of Artificial Intelligence in Detecting Oral Potentially Malignant Disorders and Oral Cancer: A Meta-Analysis of Imaging-Based Studies (2015–2024)

Abstract

Aim: Oral cancer is a leading cause of cancer-related morbidity, especially in low-resource settings. Early detection of oral potentially malignant disorders (OPMDs) is key to reducing disease burden. This study was carried out to evaluate the diagnostic accuracy of AI-assisted imaging tools in the detection of OPMDs and oral cancer in community and clinical settings. Materials and Methods: We searched PubMed, IEEE Xplore, Scopus, and Web of Science (2015–2024) for studies using AI to detect OPMDs or oral cancer from images, reporting sensitivity and specificity. A random-effects meta-analysis was conducted. Quality was assessed using QUADAS-2. Summary receiver operating characteristic curves were generated to evaluate the global performance. Results: From 1092 articles screened, 35 studies met inclusion criteria. These studies evaluated >15,000 images using clinical photography, histopathology, optical coherence tomography, and autofluorescence. Pooled sensitivity was 0.919% (95% CI: 0.89–0.94), specificity 0.879 (95% CI: 0.84–0.91), area under curve 0.9758, and diagnostic odds ratio 131.63. Deep learning methods—particularly convolutional neural networks—consistently demonstrated superior performance. Conclusion: AI-assisted diagnostic systems demonstrate high accuracy and potential for scalable, non-invasive screening of OPMDs and oral cancer. Integration into public health programs, particularly in underserved settings, could significantly improve early detection outcomes. Mobile-compatible platforms represent a viable public health tool for oral cancer control.