Development of Skin Cancer Detection Using Deep Spiking Neural Networks

The limitations of conventional techniques necessitate the development of new diagnostic strategies for early detection of skin cancer, the most prevalent form of cancer. This investigation investigates the application of Spiking Neural Networks (SNNs) in creating a dependable and automated skin cancer detection system. The primary goal is to evaluate the feasibility and efficacy of SNNs for this purpose. The methodology entails using the SpikingJelly framework to create and assess an SNN-based model, specifically a modified Spiking VGG-11. Focusing on images classified as melanoma or non-melanoma, the investigation employs the ISIC 2019 dataset. The Spiking VGG-11 model, which features approximately 15 million parameters, achieved a test accuracy of 81.3%, specificity of 81.7%, and precision of 81.5%. This performance is noteworthy in light of its reduced complexity compared to models such as the Spiking VGG-13, which has approximately 134 million parameters. The model's high specificity is especially advantageous in clinical environments, as it minimizes false positives and unnecessary testing. These results emphasize the potential of SNNs to offer diagnostic tools that are both efficient and effective in the detection of skin cancer. These findings underscore the potential of SNNs to enhance the early detection and diagnosis of skin cancer, offering a promising alternative to conventional diagnostic methods.