Skin Cancer Classification Using Wine Transformer

Abstract

Skin lesions, which may indicate benign or malignant conditions, require accurate classification for the early detection and treatment of skin cancer. Traditional methods often struggle with extracting relevant features and handling imbalanced datasets, leading to reduced performance. This study presents a hybrid deep learning model for improved skin cancer classification using dermoscopic images. The proposed system combines the SwinTransformer with a custom CNN-based module called the Dense Group Shuffle Non-Local Attention (DGSNLA) network, which incorporates DenseNet169, Group Shuffle Depth-wise blocks, and enhanced non-local attention mechanisms. By fusing global and local features, the model enhances feature representation and improves classification accuracy. Evaluated on the HAM10000 dataset, the proposed approach achieves an accuracy of 94.21%, an F1-score of 96.64%, a precision of 96.62%,and a recall of 96.25%, demonstrating its effectiveness in skin lesion classification.