Artificial intelligence-assisted detection of oral cancer and oral potentially malignant disorders on smartphone images: a systematic review and meta-analysis.

Abstract

Background: Accurate early diagnosis of oral cancerous lesions is critical for effective treatment and reliable prognosis prediction. With technological advancements, artificial intelligence (AI) has been proposed as a transformative tool for cancer detection, utilizing the analysis of digitized dental photographs to improve diagnostic accuracy.

Objective: This study aims to review and synthesize the application of AI in diagnosing oral cancer and oral potentially malignant disorders (OPMDs), with a specific focus on dental images captured using smartphones.

Methods: A comprehensive literature search was performed across Cochrane Library, PubMed, Scopus, and Web of Science databases up to August 2026 to identify articles on the application of AI in the diagnosis of OPMDs and oral cancers using smartphone photography. Included studies were evaluated for risk of bias using Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2). Diagnostic performance metrics including sensitivity, specificity, positive likelihood ratio (LR+), negative likelihood ratio (LR-), diagnostic score, and diagnostic odds ratio (DOR) were synthesized.

Results: 22 studies underwent qualitative synthesis, with 13 included in meta-analysis. Selected studies showed generally positive results for AI in diagnosing oral cancer and OPMDs. Most models, mainly convolutional neural networks (CNNs) based with some Transformers, demonstrated medium‑to‑high sensitivity and specificity. Pooled sensitivity was 0.89 (95% confidence interval [CI] 0.85-0.92) and pooled specificity was 0.87 (95% CI 0.81-0.92). Some advanced models reached F1 scores close to 90%, indicating a strong balance between detection and accuracy and highlighting the promising clinical potential of AI-assisted oral lesion diagnosis.

Conclusion: AI detection models based on smartphone images have demonstrated promising performance in diagnostic analyses, showing significant potential for screening OPMDs and oral cancers, especially in low-resource settings. Future research should focus on enhancing model explainability and conducting external validations to ensure generalizability, thereby enabling safe and effective integration into broader applications.

Clinical significance: The mobile AI screening paradigm can supplement routine oral examinations by frontline medical workers and reduce missed diagnoses of subtle premalignant lesions in remote areas. It also helps rationalize biopsy referrals and cut the economic burden of late-stage oral cancer treatment for populations with limited medical resources.