Detection and Classification of Skin Cancer Using Custom-Built CNN

Skin cancer, an extensively prevalent manifestation of malignancy, exerts a substantial toll on millions of lives each year. The failure to promptly identify and address it during its initial stages can result in its dissemination to various anatomical regions. A principal catalyst for this ailment is the anomalous proliferation of skin cells, often incited by exposure to sunlight. As the malady progresses, there is a marked decline in the likelihood of survival. Early detection of skin cancer poses a formidable challenge and is frequently accompanied by considerable financial burdens. Within this investigation, the HAM10000 dataset assumes the role of the training set, facilitating an assessment of the efficacy of deep learning-based models. The focus centers on pivotal metrics such as accuracy and precision. The current paper advocates for the adoption of a deep learning paradigm, specifically a Convolution Neural Network (CNN), to discern and categorize skin cancer. A meticulously designed deep CNN model is implemented for training on our dataset. In the evaluation phase, the model achieves a commendable accuracy of 99.42% on test data, surpassing established Computer-based vision Models – Mask R-CNN by 17.68% and YOLO v8 by 6.43%. Notably, the proposed deep CNN model outperforms contemporaneous models while maintaining computational parity. Future work could focus on integrating our CNN model into clinical settings, with a user-friendly interface for dermatologists, and expanding its evaluation to include diverse skin types and conditions for broader applicability.