Oral cancer classification using deep learning based frameworks

Abstract

Oral cancer remains a significant challenge in medical diagnostics, necessitating accurate and efficient classification methods to aid in diagnosis and treatment planning. In this study, we present a comprehensive approach to oral cancer classification utilizing deep learning techniques applied to two distinct datasets, both of which are publicly available on Kaggle.com: oral cancer images (lips and tongue) and histopathological images of oral cancer. For the first dataset consisting of histopathological images of oral cancer, a similar methodology was employed. Three state-of-the-art transfer learning CNN models, MobileNetV2, EfficientNetB3, and NASNetMobile, were trained on the dataset. Individual model testing accuracies of 89.68%, 92.06%, and 92.86% were attained, respectively. Subsequently, a fuzzy rank-based ensemble approach was applied, utilizing the Gompertz function to integrate predictions and enhance classification accuracy. The ensemble method yielded a significant improvement, achieving a testing accuracy of 94.44%. For the second dataset comprising oral cancer images, we employed two transfer learning backbone models, MobileNetV2 and DenseNet121. These models were enhanced through the application of the Convolutional Block Attention Module (CBAM) and subsequently merged. That results an impressive accuracy of 96.97% on the test set, demonstrating the effectiveness of the proposed methodology for oral cancer classification from image data. These findings underscore the potential of deep learning techniques for accurate oral cancer classification across diverse datasets. By leveraging both image-based and histopathological data, our research contributes to the development of reliable computational methods for oral cancer diagnosis, with implications for personalized treatment strategies and improved patient outcomes.