Skin Lesion Classification using Deep Learning Architectures

Abstract

Skin cancer is one of the major types of cancers that can arise from various dermatological disorders and can be classified into various types according to texture, structure, color and other morphological features. Identifying the lesions from skin images can be an important step in pre-diagnosis to aid the doctors and infer the medical condition of the patient. Recent work has focused on classifying only melanoma from a given set of skin lesion images. However, some types of skin lesions (Acctinic Keratosis and basal cell carcinoma) can become malignant over a period of time. So by detecting these classes we can say we are cutting down the risk of malignancy and doing the task of early detection. We are able to classify different types of skin lesions (basal cell carcinoma, benign keratosis, dermatofibroma, vascular lesions, melanoma, and melanocytic nevi) with an accuracy of above 80% with Mobilenet, VGG-16 and our custom model which we have designed. With the help of this models, which will be embedded in skin lesion analyzer machines. This can give the patients as well as doctors a good idea of whether or not there is a need for medical attention and can avoid unnecessary panic/false alarms. We are using different deep learning architectures to classify skin lesions with good accuracy relative to existing work.