Performance analysis of melanoma classifiers with CNN-based segmentation framework

Abstract

Skin cancer is one of the most frequent cancer, accounting for about half of all cancer diagnoses globally. Melanoma is a skin cancer that arises from melanocytes, which are pigment-producing cells. Melanoma occurrence and fatality rates have risen dramatically in recent years. Early detection can assist the medical experts in treating the disease that attracted diverse researchers. This article focuses on different segmentation technique and machine learning-based classification approach for melanoma detection. The classification process is simplified, and the segmentation approaches recognize the melanoma regions easily. The segmentation uses a convolutional neural network (CNN) with U-Net and ResNet. The segmented image is passed into the classification phase, where the classification process is attained by a support vector machine (SVM) with diverse kernels and k-nearest neighbor (KNN). The investigation is achieved using Acquired values from the machine learning classifier, and the performance indicates that the SVM is highly effective, which is enhanced by the segmentation process.