Melanoma Skin Cancer Detection Using InvolutionNet

Early and precise detection of melanoma skin cancer is very critical in improving outcomes for patients’ well-being. Deep learning techniques, especially convolutional neural networks (CNNs), have shown remarkable success towards achieving this goal. They are a corner stone in medical image processing. Despite the success, they have some limitations such as high computational complexity, lack of efficient long range dependency modeling, and the need for extensive data augmentation. This study proposes InvolutionNet model for melanoma classification, aiming to address these shortcomings. InvolutionNet incorporates the involution operation which allows the model to have better long-range dependencies and relationships across channels. We introduce an involution-based architecture tailored for melanoma detection. The InvolutionNet model achieves an overall accuracy of 87.0%. The model achieves a weighted F1-score of 0.87. The precision for the benign and malignant class comes out as 0.85 and 0.90 respectively. The recall for the benign and malignant class comes out as 0.91 and 0.83 respectively. These finding demonstrate the potential of InvolutionNet for precise melanoma classification. The model showcases competitive or superior performance compared to existing CNN architectures proposed in the literature. The proposed methodology contributes to the advancement of computer aided melanoma detection and accurate diagnosis. It also promotes the adoption of involution in medical image analysis.