Optimizing Hyperparameters with Pelican Algorithm for Deep Learning–based Oral Cancer Detection

This paper presents a comprehensive study on oral cancer segmentation using deep learning techniques, focusing on the U-Net and DenseNet architectures. U-Net, designed for biomedical image segmentation, leverages its encoder-decoder framework to produce high-resolution outputs critical for identifying tumor boundaries in oral cancer cases. DenseNet, recognized for its densely connected layers, improves feature reuse and mitigates the vanishing gradient issue, thereby enhancing overall model performance. The combined use of these models demonstrates the effectiveness of deep learning in medical image analysis, supporting early detection and treatment planning. The study highlights the significance of annotated datasets specific to oral cancer segmentation, which serve as the ground truth for training and validation. Given the limited availability of such datasets, data augmentation techniques are applied to expand the training samples and improve the generalization capability of the models. The optimization of hyperparameters using the Pelican Optimization Algorithm further refines model accuracy and performance. Results indicate improved segmentation outcomes compared to traditional approaches. This research underscores the value of combining robust architecture with intelligent optimization strategies in the context of oral cancer detection.