Detection of oral potentially malignant disorders and oral cancer with white light through deep learning: a systematic review and meta-analysis

Abstract

OBJECTIVES: The research question was: How accurate is artificial intelligence (AI) in diagnosing Oral Potentially Malignant Disorders (OPMD)/oral cancer in patients of any age, using digital photographic images under white light? MATERIALS AND METHODS: A systematic review was performed in the PubMed, Web of Science and Scopus databases. The inclusion criteria were: detection of OPMD or Oral Squamous Cell Carcinoma (OSCC). The risk of bias was assessed with the QUADAS-C tool. Forest plots were generated with the specificity and sensitivity of oral cancer and OPMD. A bivariate random effects meta-analysis was performed, by combining sensitivity and specificity. RESULTS: The bivariate model showed a pooled sensitivity of 0.81 (95% CI, 0.728-0872) and a grouped specificity of 0.161 (95% CI, 0.113-0.223) for OPMD detection, and a combined sensitivity of 0.836 (95% CI, 9.730-0.906) and combined specificity of 0.138 (95% CI, 0.081-0.226) for oral cancer. CONCLUSIONS: Despite substantial heterogeneity among the various Deep Learning algorithms evaluated, most models demonstrated acceptable diagnostic capability for identifying OPMD and OSCC from digital images. These findings support the promising role of artificial intelligence in enhancing diagnostic processes within dentistry. CLINICAL RELEVANCE: Being able to diagnose OPMD through the use of digital photography allows for an early diagnosis, especially in places with limited resources. The possibility of performing assessments via smartphone devices increases accessibility and reduces dependence on costly specialized equipment, broadening its applicability in clinical practice.