Dilated Convolutional Neural Network for Skin Cancer Classification Based on Image Data

Abstract

Skin cancer is a disorder of cell growth in the skin. Skin cancer has a big impact, causing physical disabilities that can be seen directly and high treatment costs. In addition, skin cancer also causes death if nor treated properly. Generally, dermatologists diagnose the presence of skin cancer in the human body by using the Biopsy process. In this study, the Dilated Convolutional Neural Network method was used to classify skin cancer image data. Dilated Convolutional Neural Network method is a development method of the Convolutional Neural Network method by modifying the dilation factors. The Dilated Convolutional Neural Network method is divided into two stages, including feature extraction and fully connected layer. The data used in this study is HAM1000 dataset. The data are dermoscopic image datasets which consists of 10015 images data from 7 types of skin cancer. This study conducted several experimental scenarios of changes in the value of d, which are 2,4,6, and 8 to get the optimal results. The parameters used in this study are epoch = 100, minibatch size = 8, learning rate = 0.1, and dropout = 0.5. The best results in this study were obtained with value of d=2 with the value of accuracy is 85.67% and the sensitivity is 65.48%.