Skin Lesion Image Segmentation using Modified UNet Architecture

Abstract

One of the most dangerous types of skin cancer is malignant melanoma. Early diagnosis, according to modern dermatology, is critical for lowering mortality rates and ensuring that patients receive less invasive therapies. For the early identification of skin lesions, computer-aided diagnostic (CAD) systems are becoming more popular. These systems are made up of various phases that must be selected based on the properties of digital images in order to produce a correct diagnostic. Acquisition, pre-processing, segmentation, feature extraction and selection, and finally classification of dermoscopic images all provide problems that must be met and conquered in order to improve automatic diagnosis of deadly tumors like melanoma. The categorization phase is particularly delicate, and a number of machine learning techniques have been presented over time to address this problem more effectively. The many machine learning approaches that have been proposed and that provide inspiration for the creation of effective frameworks are discussed in this study.