Detect pre-cancerous tongue lesions for early oral cancer diagnosis using deep learning algorithm

Precancerous tongue lesion is a prevalent, complex, and highly perilous kind of cancer. The tumour might be in the salivary glands, tonsils, neck, cheek, and mouth. Oral Cancer (OC) is commonly identified in advanced stages due to the limited accuracy of available screening methods for early detection despite their significant potential to reduce mortality rates. The study exclusively examines lesions that specifically manifest on the tongue. This work demonstrates that one of the deep learning (DL) such as convolutional neural networks (CNN) based models employed are novel in their capacity to effectively identify OC, primarily due to the limited research conducted in this field. The research utilizes a specifically created dataset due to the absence, to the best of our knowledge, of any existing information on tongue lesions occurring in the oral cavity. The research recommends using various methods, such as DenseNet121, DenseNet169, DenseNet201, MobileNet, MobileNetV2, VGG16, VGG19, ResNet50, EfficientNetV2B0, EfficientNetV2B1, EfficientNetV2B2, EfficientNetV2B3, Inception, AlexNet, and transfer learning, to enhance our ability to diagnose OC. The effectiveness of the enhanced technique is assessed based on customized data. The study's input parameters consist of a portrait of the patient's tongue. according to the assessed outcomes of training precision, validation precision, training loss, and validation loss. The outcome indicated that VGG16 achieved the highest performance based on the given parameters. It demonstrated a vital training accuracy of 97.66% and a commendable validation accuracy of 89.06%. The study's Clinical trial number not applicable.