Classification of dermoscopic images using soft computing techniques

Abstract

Medical diagnosis using machine learning techniques has great attention over the last two decades. The detection of skin cancer based on visual information requires highly skilled dermatologists, and also it is a time-consuming process. To analyse the condition of patients and diagnose the diseases at the earliest, an automated classification system is needed that may help to enhance the clinical decision. There are many clinical trials available to classifying melanoma skin cancer. In this study, soft computing techniques are enabled such as image processing, Genetic Algorithm (GA) and Deep Learning Neural Network (DLNN) to get accurate result of classification. To achieve this, three major modules are developed. The first part is dermoscopic image preprocessing from which the dermoscopic images can be prepared by removing noises and skin hairs. The second one is feature extraction and selection. The former one utilizes 3-dimensional Discrete Wavelet Transform (3DWT), and later one uses GA. The last module is the knowledge discover step where the dermoscopic images are classified using an appropriate DLNN classifier. In the result of comparative analysis, a maximum accuracy of 98.67% is obtained using the proposed system on PH2 database and 94.50% on International Skin Imaging Collaboration (ISIC) 2017 database.