Detection and segmentation of oral lesion using Mask R-CNN

Early detection of oral cancer is essential for enhancing patient outcomes and preserving lives. Nevertheless, inaccurate and inappropriate diagnosis may impede the effectiveness of treatment. In recent years, deep learning techniques have assumed greater significance with the advancement of artificial intelligence, establishing themselves as a fundamental technology in the field of medical diagnosis. It automates image classification process and enhances accuracy. The study proposes Mask R-CNN (Region-based Convolutional Neural Network) for identifying the cancerous part or lesion in an image. The Photographic images are annotated and trained using mask R-CNN. With backbone as ResNet-50 extract features and used to obtain the bounding box along with the classification and segmentation mask. ROI (Region of Interest) alignment aids in enhancing the localization of objects by fine-tuning the feature maps and ROI, thereby delivering highly accurate segmentation masks. ROI align also ensures the prevention of data loss. The model provides the mean average precision (mAP) of 75.34%, precicsion of 76.45% and a recall of 73.25%. This technique enables the automated identification of cancerous regions within an image, assisting medical professionals in early detection.