AI-assisted image analysis of in vitro oral cell morphology for early disease detection

Premalignant and malignant oral lesions pose a significant global health burden, with delayed diagnosis adversely affecting patient outcomes. Therefore, it is of interest to evaluate the diagnostic performance and reliability of an artificial intelligence (AI)-based image analysis system in identifying early morphological changes in cultured oral epithelial cells under simulated pathological conditions. Human oral keratinocytes were exposed to oxidative stress and inflammatory cytokines and phase-contrast images were analyzed using a convolutional neural network trained to classify normal, stressed and dysplastic cells. The AI system demonstrated high accuracy (94.2%), sensitivity (93.8%), specificity (95.1%) and strong agreement with expert cytopathologists (κ = 0.91), with reduced analysis time compared to manual evaluation. AI-based image analysis shows promise as a reliable adjunctive tool for early detection and screening of oral lesions.