Skin Lesions Classification of Dermoscopy Images Using Deep Learning Technique

Skin lesions are a severe disease and the most predominant type of cancer worldwide. It is becoming more prevalent in modern society, with rising cases every year. The World Health Organization (WHO) claims melanoma is from the most severe forms of skin cancer, affecting well over 100,000 people worldwide each year. While early-stage lesions are frequently treatable, the prognosis worsens dramatically as the illness progresses, enforcing the need to identify cancer correctly and as early as possible. The computer-based analysis of dermoscopy images helped in the early detection of melanoma which had a significant impact on increasing the survival rate of the people who are infected from this. However the order of accurately recognizing the types of melanoma is quite challenging because of the small deviations between the lesions and skin, and there are lots of visual similarities between melanoma and non-melanoma lesions. However, images with a lower resolution than the original skin image are produced as a result of segmenting the skin lesions. In this paper, by using a dermoscopic image of a skin tumor, we present a novel deep learning-based method for resolving the issues with skin lesion analysis aimed at enhancing accuracy and efficiency. The models which being proposed are inception ResNet-V2, EfficientNet, InceptionNet, and MobileNet, which are trained and then evaluated on HAM 10000 datasets to compare the performance and accuracy. The suggested strategies produced a robust accuracy in the validation sets. The experimental studies conducted on a clinical dataset reveal that deep learning-based features fare better for classification than conventional machine learning methods.