A Robust CNN-based Approach for Skin Lesion Detection and Classification

The abnormal cell growth that most frequently develops on skin that has in some way been exposed to the sun is called skin cancer. Yet, areas of our skin that are not commonly exposed to sunlight might potentially acquire this sort of cancer. The three main kinds of skin cancer are Melanoma, Basal cell carcinoma and squamous cell carcinoma. On an estimate around 9,500 people are affected daily by skin cancer in the United States. Early screening and detection of warning signs can help to detect and cure cancer before it becomes serious. Therefore, deep diving in the world of Deep Learning aims to build an easily accessible technology that can detect skin cancer based on the type of dermatoscopic lesions. In this paper, we have explored a similar venture, where we have built a classification model on the type of skin lesions using CNN. We have trained our model on MNIST HAM-10000 dataset. The efficiency of our model has been assessed using accuracy, recall, precision and f1-score. We have achieved maximum accuracy of 99.13% for classification of seven types of skin lesions.