Recognition of Oral Cancer Progression Stages using White Light Imaging

Oral cancer is the sixth most frequent type of cancer worldwide, and it is one of the main causes of mortality. Although early detection during precancerous phases can greatly increase survival rates, late-stage diagnoses are the cause of its high mortality rate. Using image analysis of oral lesions, this study suggests an effective diagnosis technique. The lightweight MobileNetV2 CNN architecture is used to classify the texture and color properties that are extracted from photos by exploring five color spaces. This model enhances accuracy while still being appropriate for resource-constrained, mobile applications by fusing deep learning with handmade features. The approach performs well in binary and multi-class classification tasks, differentiating between benign, malignant, and precancerous lesions. Particularly in underprivileged areas, its scalability and affordable design make it perfect for early oral cancer screening. Keywords: Oral Cancer Detection, Pre-Cancerous Lesions, MobileNetV2, Convolutional Neural Network (CNN), Color and Texture Features, Medical Image Classification, Deep Learning, Lightweight Architecture, Real-Time Diagnosis, Resource-Constrained Devices