Mobile AI-Assisted Oral Lesion Triage for Community Oral Cancer Screening: Prospective Field Evaluation Study

Artificial intelligence (AI)-based tools for oral cancer screening have shown promising performance in curated or retrospective image datasets, although such evidence may not fully reflect performance in real-world mobile community screening workflows. In remote or resource-limited settings, intraoral images are often acquired by trained non-specialist personnel under variable field conditions, making workflow-integrated image quality assurance, risk stratification, and expert oversight essential for human-supervised AI-assisted screening.

This study aimed to evaluate the prospective field performance of a parameter-locked, mobile AI-assisted oral lesion triage system embedded within a human-supervised routine community oral cancer screening workflow.

We conducted a prospective community-based field evaluation in eastern Taiwan from June 17 to December 31, 2025. Trained non-specialist personnel acquired standard white-light intraoral images using handheld mobile devices during routine oral cancer screening activities. The AI system incorporated on-site image quality assessment and lesion-level risk stratification into green, yellow, and red triage categories. Images that remained technically inadequate after repeated acquisition attempts were classified as ungradable and excluded from performance analysis. Three board-certified oral and maxillofacial specialists established an operational clinical reference standard through structured image review and consensus adjudication. The primary outcome was lesion-level identification of high-risk lesions requiring specialist referral, defined as red versus non-red triage. Diagnostic performance was estimated with Wilson 95% confidence intervals; precision-recall performance and pre-consensus inter-rater agreement among specialists were also assessed.

Among 602 screened participants, 4283 interpretable intraoral images were included in expert adjudication and lesion-level analysis. The AI system flagged 68 red lesions; 12 were confirmed as high risk by expert adjudication, and 4 additional high-risk lesions were identified during expert review, yielding 16 high-risk events for the primary red-triage analysis. For high-risk lesion identification, sensitivity was 75.0% (95% CI 50.5%-89.8%) and positive predictive value was 17.6% (95% CI 10.4%-28.4%); specificity, negative predictive value, and accuracy were 98.6%, 99.9%, and 98.5%, respectively. ROC-AUC values were 0.957 for red, 0.931 for yellow, and 0.964 for green triage; precision-recall analysis showed PR-AUC values of 0.8235 for red, 0.9052 for yellow, and 0.9476 for green triage. Pre-consensus specialist agreement was high, with a Fleiss' kappa of 0.921 and a Gwet's AC1 of 0.981.

A parameter-locked mobile AI-assisted oral lesion triage system was feasibly integrated into a real-world community oral cancer screening workflow operated by trained non-specialist personnel. In this low-prevalence field setting, red triage showed moderate sensitivity and modest positive predictive value, supporting its potential role as human-supervised referral-prioritization support rather than autonomous diagnosis or standalone clinical decision-making. Further multicenter studies with blinded assessment, participant-level outcomes, workflow-efficiency measures, and longitudinal follow-up are needed before clinical impact and scalability can be determined.