Clinical Computer Vision for Oral Cancer Screening Under Variable Intraoral Camera Exposure: A Domain-Adaptation Evaluation of Artificial Intelligence Referral Specificity Across Primary Care Clinics

Intraoral camera exposure varies across primary care clinics, and this variation may degrade the specificity of artificial intelligence (AI) referral decisions for oral cancer screening. The framework formalises exposure variation as a covariate shift in image intensity and derives a bound on the resulting change in referral specificity under a declared variance structure. A replication-based power analysis of the before-after-control-impact design was conducted across 2000 design cells, with between-cluster SD = 6.0, residual SD = 4.5, and intraclass correlation = 0.64. The analysis indicates that 50 clusters are required to detect a 5 percentage-point effect at 80% power. The framework proposes that exposure normalisation and domain-adversarial training are necessary but not sufficient conditions for preserving referral specificity, and that clinic-level calibration monitoring is required. The contribution is a formal characterisation of the exposure-specificity trade-off and a design template for evaluating domain-adaptation interventions in oral cancer screening.