Diagnosing Oral Mucosal Diseases Using Deep Learning

Oral Potentially Malignant Disorders (OPMD) contain a variety of diseases with complex pathological features, most of which are associated with systemic diseases and have a high risk of malignant transformation, making diagnosis challenging. In this regard, we adopted rapidly developing artificial intelligence methods in recent years, using deep learning for feature extraction and multi-task research on images of oral mucosal diseases. Concerning the severity of oral mucosal diseases, we employed a Swin Transformer-based classification model for the classification of benign diseases, OPMD, and oral cancer, achieving an accuracy of 94.2% and a precision of 95.0%. Further research explored the instance segmentation algorithm based on Mask2Former, comparing it with classic segmentation algorithm to achieve automated segmentation of oral mucosal inflammatory lesions. The study shows that deep learning methods can effectively discriminate the severity of oral mucosal disease images and segment lesion areas, providing solid research foundation for intelligent assisted diagnosis.