AN EFFECTIVE APPROACH FOR HISTOPATHOLOGIC ORAL CANCER PREDICTION USING BITTERLING FISH OPTIMIZATION TECHNIQUE

Oral cancer poses a substantial global health concern, necessitating immediate action to mitigate its severe consequences. However, existing methods used for oral cancer classification exhibit limitations such as getting trapped in the local optima, high sensitivity to noisy data, and poor convergence speed, resulting in lower prediction reliability and clinical applicability. Hence, this research proposes a Bitterling Fish Optimization with Probability Entropy (BFO-PE) algorithm for feature selection during oral cancer classification. The PE enables the BFO to select optimal feature subsets by simulating the spawning behavior of bitterling fish, updating iteratively toward an optimal feature set, thereby enhancing performance. Initially, the Oral cancer (Tips and Tongue) Image (OCI) dataset and Kvasir Video and Image Repository (KVASIR) dataset are used to validate the model’s performance. The next step, pre-processing involves resizing images to standardized dimensions using interpolation. Then, feature extraction is performed using the ResNet 152 architecture, followed by which feature selection is carried out. Finally, the ResNet layer is deployed for classification of oral cancer into binary classes. The proposed approach attains a superior accuracy of 99.04% on OCI and 99.75% Kvasir datasets, outperforming existing methods like Convolutional Neural Networks - Improved Tunicate Swarm Algorithm (CNN-ITSA). These results demonstrate the effectiveness, generalizability, and potential clinical applicability of the developed system for oral cancer prediction.