Exploring Data Modalities and Advances in Related AI Technologies for Oral Cancer Detection

Oral cancer diagnosis represents a significant public health burden; late-stage detection of oral cancer is a major issue for ineffective treatment. Multimodal approaches from artificial intelligence have emerged as a pretty promising approach to address this challenge. In this paper, a comprehensive review of recent studies of oral cancer detection across varied data modalities that utilise technologies such as computer vision, natural language processing, acoustics analysis, Internet of Things, and machine learning and Deep Learning (DL) is presented. Across the reviewed literature, unique datasets spanning imaging, histopathology, spectroscopy, and clinical text are identified and represented. Reported performance metrics vary by modality, such as image-based DL methods, which achieved accuracies between 91% and 99% and area under the curve values up to 0.95, spectroscopy-based approaches reported accuracies above 92%. These results highlight the diagnostic potential of varied data modalities for future research direction, and small, imbalanced datasets, lack of external validation, and personalisation are major concerns to be addressed.