Higher Diagnostic Accuracy for Melanoma in Dermoscopy Images using Convolution Neural Network

Abstract

Dermoscopy is an effective method in the early detection of melanoma, increasing the diagnostic accuracy of clinical visual analysis in the hands of experienced physicians. A pigment network whose structure varies in size and shape is called an uncommon or typical pigment network (APN). The median split pixel clustering algorithm is based on the characteristics of an image colour histogram after the lesion border has been used to segment the lesion from the remainder of the dermoscopic image; the median split algorithm has been applied to pixels in the lesion area.  The classification is further carried out by convolution neural network (CNN) in the proposed system segmentation model in order to increase the efficiency of classification. Finally, the input images are compared to the database images and then indicate whether they are normal or abnormal.