Abstract

Skin lesions play a crucial role as the initial clinical symptoms of diseases such as chickenpox and melanoma. By employing digital image processing techniques for skin cancer detection, it becomes feasible to diagnose these conditions without the need for physical contact with the skin. However, the automatic analysis of dermoscopy images, which exhibit characteristics like residue (hair and ruler markers), indistinct borders, varying contrast, and variations in shape and color, poses significant challenges. To overcome these difficulties, effective hair removal through segmentation has been explored extensively in the literature. In this study, we present a skin lesion segmentation system developed using the Convolutional Neural Networks (CNNs) method with the U-Net architecture. The model was constructed and evaluated using the HAM10000 Dataset. The results achieved by the best-performing model were outstanding, with a Pixel Accuracy, Intersection over Union (IoU), and F1 Score of 95.89%, 90.37%, and 92.54%, respectively