High Performance Deep Learning Convolutional Neural Network Model for Automatic Classification of Oral Cancer

Abstract

This research presents a methodology for classifying oral cancer histopathological images using a proposed Convolutional Neural Network (CNN) model. The process involves six stages: dataset collection, image preprocessing, data augmentation, data partitioning, image classification, and performance evaluation. The dataset consists of 1224 histopathological images, divided into cancerous and non-cancerous classes, captured at various magnifications. To address data limitations, a median filter was applied for noise reduction, and images were resized to 128x128 pixels. Data augmentation techniques, such as rotation, shifting, and zooming, expanded the dataset to 4577 images. The dataset was split into 75% training and 25% testing sets. The proposed CNN model was compared to models like VGG19, AlexNet, ResNet50, ResNet101, MobileNet, and InceptionNet. Performance metrics, including precision, recall, specificity, F-measure, and accuracy, were used to evaluate the models. The proposed CNN achieved the highest accuracy of 96%, outperforming the other models. This CNN model offers a promising tool for early oral cancer detection, aiding doctors in diagnosis and treatment planning.