Instance Segmentation of Oral Cancer Images with Fusion of Swin Transformer and Mask RCNN

Oral cancer is the most preventable cancer if it is diagnosed at an early stage. Artificial intelligence can be a great help in cancer detection. Deep learning architectures are predominantly useful in medical image analysis by identifying patterns and the ability to predict the insights. The study proposes a deep learning methodology using Mask RCNN (Region Based Convolutional Neural Network) for the precise detection and segmentation of oral lesions in photographic images. With the swin transformer as a backbone, it aids the model in extracting features more effectively, thus supporting precise detection. Its ability to identify relationships among different parts of an image is particularly useful in locating the smallest lesions. The precise annotation has helped generate the segmentation mask accurately. The model attains a mean average precision (mAP) of 99.5%, a precision of 92.7% and a recall of 96.6%. This exceptional performance of the model is useful for the medical community to use it as a tool for the early detection of oral cancer.