Publication

Abstract : Oral cancer causes high morbidity and mortality in India. Early detection is critical but hindered by the varied presentation of lesions. In this study, we developed a deep convolution neural network (CNN) model called 'OralNet' for oral cancer classification from lesion images. The model has multiple convolutional and pooling layers for robust feature extraction. We used a dataset of benign and malignant oral lesion images from various hospitals in Karnataka, India. After training on 80% of the data, the OralNet model achieved 98.4% training accuracy, outperforming pretrained models like Xception (88.7%) and InceptionResNetV2 (86-69%). The high accuracy demonstrates the model's potential for reliable early detection of oral cancer. Expanding the training data size can further improve generalization capability. Overall, deep learning holds promise for oral cancer diagnosis, especially in low-resource settings, by enabling automated and accurate image analysis