Publication

Abstract : This research introduces a comprehensive technique for categorising oral cancer images using deep learning algorithms. The dataset includes a standardized series of grayscale oral cancer images that undergo preprocessing steps, including normalisation, augmentation, and resizing to 299×299 pixels to ensure the model's accuracy and robustness. The pretrained Xception model is primarily based on ImageNet, fine-tuned with the aid of extra dense layers specific to oral cancer classification. Extensive hyperparameter tuning complements the model's performance and achieves high classification accuracy. Additionally, the Siamese network evaluates image pairs and determines their similarity. Performance indicators such as precision, recall, and F1 score were used to validate the model. This highlights the robustness of the model. The dual-model architecture provides a collaborative framework that improves classification accuracy and enhances clinical interpretability through similarity analysis. The proposed approach represents a significant advancement in oral cancer detection and lays the foundation for reliable and scalable diagnostic tools in clinical practice