COVID-19-related Skin Disease Classification using Convolutional Neural Network

Abstract

COVID-19 is affecting the skin by the virus in a wide range according to the global reports. The main purpose of this research is to develop a classification model for the four COVID-19- related skin diseases. In this thesis, an ImageJ tool is used to produce segmented images and a CNN method is used to develop the COVID-19-related skin disease classification model. The COVID-19 skin related images are collected from DermNet and Bing images. Initially ImageJ tool is used to segment the lesion region from the nearby healthy skin manually then CNN algorithm is used to classify skin diseases as Livedo-like pattern, Maculopapular rash, Purpuric or Urticaria. Classification results are found with and without segmented images. The proposed classification model result with segmented-image dataset produced training accuracy of 97% and validation accuracy of 99% with epoch 50, and an average testing accuracy of 99%. According to these values the CNN based model with segmented-images yields better result and it also improves classification performance. The research is mainly used for keeping documents of the skin symptoms related to COVID-19 and it is also helpful for the frontline physicians to identify the patients early.