A Metaheuristic Optimization Based Deep Feature Selection for Oral Cancer Classification

Oral cancer is a serious hazard to world health, with many new cases recorded each year. Researchers have been concentrating on developing medical image analysis-based computer-aided diagnostic (CAD) systems for oral cancer. To this end, we propose a novel model that we name GrayWolf Optimization (GWO) based deep Feature Selection Network (GFS-Net). Initially, we use an attention-aided NASNet Mobile, a convolutional neural network (CNN) architecture, to extract features from the input images. Next, we use a metaheuristic-based optimization algorithm, called GWO, to get rid of the extraneous features obtained from the CNN model. For the final classification task, the K-Nearest-Neighbours (KNN) classifier is applied with this optimal feature set. Our model is evaluated on two publicly accessible oral cancer datasets, histopathologic oral cancer identification dataset and oral cancer (lips and tongue) dataset, that yields classification accuracies of 92.86% and 93.94%, respectively. The code and additional results are available at https://github.com/stellarsb7/GFS-Net.