Integrative Approaches for Skin Cancer Detection and Classification : A Dual modal Analysis

The early detection of skin cancer is of critical importance, as it can lead to in fatal outcomes if left unaddressed. Given the limited accessibility of dermatological expertise to the general population, it becomes imperative to devise an economical, efficacious, and precise methodology capable of efficient & reliable diagnosis of melanoma and other forms of skin cancer. A data-driven paradigm emerges as the optimal solution for the early and accurate identification of skin malignancies. In this study, we explored an array of Machine Learning (ML) and Deep Learning (DL) algorithms, identifying the most effective model for this context.To conduct this study, we utilized the ISIC 2024 challenge dataset, which contains images of cropped skin lesions, along with metadata associated with each lesion.The developed algorithm is operable without the necessity for specialized clinical intervention and demonstrates exceptional accuracy in differentiating between malignant and benign skin cancer cases, utilizing both cropped skin lesion images from Total Body Photography (TBP) and corresponding tabular data/metadata. For CSV format metadata, Random Forest performed best with an accuracy of 99.928% and an F1-Score of 99.92%, and for cropped skin lesion images, ConViT Tiny performed best with an accuracy of 96.99% and an F1-Score of 96.9%. The output of the proposed work would help in creating a benchmark for preliminary diagnosis of the cancerous tissues and timely prognosis.