MelaNet50: An Image Enhancement and Explainable Deep Learning Approach for Skin Lesion Classification

Skin is the largest organ that acts as a primary defense system of the human body. Yet, every year, 3.5 million people are affected by skin cancer across the whole world, and the numbers are continuously rising. Only accurate and early diagnosis can help mitigate the complications of this life-threatening global concern. However, the traditional method alone fails to meet such requirements. On top of that, most of the skin of the human body is covered with hair, which hinders the visibility of lesions. As a result, the correctness of diagnosis is reduced. In this study, the skin image clarity has been enhanced using several image enhancement techniques. Furthermore, a deep learning model named MelaNet50, which is inspired by the architecture of ResNet50, has been introduced for early and accurate diagnosis of skin cancer. The proposed MelaNet50 model exhibits good performance metrics when compared to the previous studies conducted on the same topic. Finally, two prominent Explainable AI (XAI) techniques have been integrated with the 94.7% accurate MelaNet50 model to convey the region of interest within the images, based on which the proposed DL (Deep Learning) model is classifying the images. Hence, this study presents an enhanced and trustworthy computer vision-based decision support system for healthcare specialists for early and error-free diagnosis.