An Enhanced Skin Cancer Detection Method Utilizing TriBlendNet and Deepdilated Focus U-Net

Skin cancer remains a major global health concern, demanding advanced methods for accurate and early identification. This work presents an integrated framework that employs a Generative Adversarial Network (GAN) for effective data augmentation and a novel Deep Dilated-Focus U-Net enhanced with attention mechanisms for precise lesion segmentation. For classification, a hybrid model named TriBlendNet is proposed, combining the advantages of the SqueezeNet and DenseNet121 architectures. Using the SIIM-ISIC 2019 dataset, the proposed system outperforms existing models such as SqueezeNet, DenseNet121, ResNet50, and VGG16. The TriBlendNet model achieved an outstanding accuracy of 98.59%, along with high precision, recall, and specificity, showcasing its strong capability for reliable and efficient automated skin cancer detection.