Skin Cancer classification: An Advanced Deep Learning Model Architecture for Classifying Dermoscopic Images

Abstract

Skin cancer is a widely spread type of cancer that impacts a significant number of individuals globally. As per the data presented by the World Health Organization, almost 2-3 million cases of non-melanoma skin cancers and about 132,000 cases of melanoma skin cancers are identified every year. The survival rate decreases as the skin cancer stage progresses. The timely identification of skin cancer can significantly improve the survival rate of patients by up to 74%. However, it is an expensive and challenging procedure to identify the cancer type in the early stages. Machine learning and deep learning have the potential to make a significant impact on the classification of skin cancer. A robust medical support system for classifying skin cancer from dermoscopic images is required for the prognosis of skin cancer. Therefore the primary objective of our research is to develop a model which automatically classifies skin cancer into distinct types. We conducted several experiments on skin cancer dataset utilizing machine learning and deep learning techniques, including fine-tuning and transfer learning. We observed that the deep learning fine-tuned model, which utilized the Densenet201 architecture to implement fine-tuning, performed exceptionally well across all experiments. Furthermore, we evaluated the performance of all models to classify seven types and further three categories of skin cancer. The accuracy levels achieved by all classification models were notably commendable, meeting the established benchmark for classification.