Abstract
Oral cancer is a globally prevalent and life-threatening malignancy, where early detection can significantly improve prognosis and reduce mortality. Traditional screening methods are often limited by operator dependence, invasiveness, and high costs, leading to frequent late diagnoses. This systematic review aims to evaluate the current application of artificial intelligence (AI) technology in the early diagnosis and risk prediction of oral cancer, with a focus on diagnostic accuracy, methodological diversity, and clinical translatability. METHODS: We conducted a systematic search across five databases (PubMed, Embase, Cochrane Library, Web of Science, and Scopus), incorporating 63 high-quality studies. The analysis was performed at two levels: data input modalities and the evolution of AI algorithms. Study selection, data extraction, and quality assessment followed standard systematic review protocols. RESULTS: AI models demonstrated high sensitivity and specificity in detecting early oral lesions and differentiating precancerous lesions, showing a trend toward multimodal fusion, lightweight, and high-performance development. However, most studies faced challenges such as insufficient sample sizes, limited external validation, and poor model interpretability. CONCLUSION: AI holds significant potential for improving early oral cancer screening. To fully realize its clinical value, it is essential to establish large-scale multicenter datasets, conduct rigorous prospective validation, enhance model transparency, and address ethical and privacy concerns.