A Shearlet-Based Second Order System for Classifying Oral Cancer: An Analysis of Histopathological Images

Abstract

In this study, we propose a Second Order Shearlets (SOS) based system for Oral Cancer Classification (OCC) that leverages histopathological images. The fundamental premise of the system is the observable variations in texture patterns between normal and abnormal cells within these images, which can be exploited for differentiation. The images undergo a transformation from the Red-Green-Blue (RGB) color space to the Hue-Saturation-Value (HSV) color space, followed by the extraction of co-occurrence texture features via the SOS system. Further enhancement of feature extraction is achieved by applying a median filter for de-noising the histopathological images. The proposed SOS-OCC system, equipped with a probabilistic classifier at the final stage, was presented with an assortment of 1224 images for evaluation. The results indicated a noteworthy classification accuracy of 98.6% when employing stratified k-fold cross-validation, thereby underlining the system's efficacy in identifying oral cancer-related abnormalities. Moreover, a comparative analysis was conducted with Wavelet, Curvelet, and Contourlet-based representation systems to underscore the superior performance of the SOS-OCC system. This study provides valuable insights into the application of the SOS approach to oral cancer classification and sets a promising precedent for future research.