ANN-Based Classification of Skin Cancer Lesions in Dermatological Imaging

Early detection of skin cancer, especially melanoma, is one of the major steps toward lowering the mortality rate. However, the existing method often employs subjective human judgment to diagnose; therefore, it contains its own set of limitations. This research presents a novel technique to auto-identify lesions in the images of dermatology skin cancer using the ANN system. The proposed solution will address the problem of data imbalance by resizing, normalizing, and augmenting images as part of its comprehensive preparation infrastructure. Optimization techniques, such as dropout and L2 regularization, are applied to prevent overfitting in a custom-designed ANN architecture with multiple concealed layers. The Adam optimizer is used for training and cross-validation purposes on the ISIC 2020 Skin Cancer Dataset. This classifier, following the ANN architecture, was able to surpass the traditional machine learning methods and CNN about accuracy of 94.5%, recall, precision, and F1-score by consuming fewer CPU resources, as shown in the experimental results. The authors hope that this technology can be a boost to the current methods of diagnosing skin cancer, which is quite crucial to clinical practice.