DeepSkinGuard: A Robust Skin Cancer Detection Framework Integrating Modified U-Net Segmentation and QuadraBlendNet Fusion

Background: Global concern over skin cancer prompts the need for early detection to improve patient outcomes. This research addresses the critical issue of skin cancer detection, emphasizing the significance of early diagnosis for improved patient outcomes. The study identifies existing challenges in accuracy and robustness within current detection systems. To overcome these limitations, an advanced methodology is proposed, integrating cutting-edge techniques at every stage of the process. Methods: The research outlines key objectives, including the development of a comprehensive image augmentation strategy, the implementation of sophisticated preprocessing techniques, and the design of a modified U-Net architecture for accurate Region of Interest (ROI) identification. The hybrid optimization model, Bridging Optimization: Eagles’ Cruise Attack Fusion (BOECAF) Algorithm, merges the strengths of the Bald Eagle Search Optimization (BES) and Golden Eagle Optimizer (GEO) to effectively optimize the batch size of the U-Net. Deep learning-based feature extraction is employed using a pre-trained Inception V3, complemented by the extraction of color-based and texture-based features. A weighted feature fusion approach is introduced to effectively combine these diverse features. The integration of the QuadraBlendNet model, incorporating SqueezeNet, Xception, ResNet50, and DenseNet201, culminates in a comprehensive and diverse representation for skin cancer detection. Results: Utilizing Python, the model is implemented and yielded an accuracy of 98%, showcasing its high performance in classification tasks, and the ISIC dataset is used for evaluation. Conclusion: Anticipated outcomes include enhanced accuracy, improved generalization capabilities, and a significant advancement in early skin cancer diagnosis. This research contributes to the field by offering a more reliable and effective approach to address current limitations in skin cancer detection methodologies.