AI-Powered Oral Cancer Detection Using Machine Learning and Deep Learning

Abstract

Globally, oral cancer remains a major problem. Late diagnosis is often associated with lower chance of survival. Although established techniques such as a biopsy are essential, they have limitations, especially in communities with no specialists. This study investigates how deep learning can improve the diagnostic process by developing and accessing convolutional neural network (CNN) models on images of oral lesions. Three architectures were trained and evaluated which are MobileNet, a modified VGG16, and a VGG16 baseline. Through thoughtful image processing and accepting greater dataset of 655 instances, they were assessed. The modified VGG16 achieved particularly strong performance under fresh test data with accuracy up to 98%. In addition, it had a great F1-score, recall, and precision. It reduced missed cancers and false alarms considerably because it achieves perfect recall of healthy tissue and perfect precision of cancerous lesions. The findings suggest that AI models can quickly and accurately diagnose oral cancer, particularly in resource-limited areas or places with less expertise.