Enhancing Monkeypox diagnosis: a machine learning approach for skin lession classification

Abstract

Virus which causes monkeypox is capable of infecting both humans and nonhuman primates. In order to effectively treat monkeypox and prevent the disease’s further spread, it is necessary to diagnose the skin lesions caused by the disease in their earliest stages and accurately classify them. In this dissertation, we study the possibility of a system which is Machine Learning (ML) based for the classification and detection of monkeypox, a skin illness caused by the varicella-zoster virus. We acquired photos of monkeypox lesions from kaggle, augmented them in order for us to develop and test our own machine learning models. We built a basic mobile app that enables users to take images with their smartphones and then send those pictures to our machine learning models so that the pictures can be analyzed and categorized. We will update it in the future according to our users needs. Our primary objective is to examine the utility and effectiveness of applying machine learning models for the purpose of categorizing and identifying monkeypox. The possible effects of the proposed system on current healthcare systems and the usefulness of machine learning models based on a number of factors will be looked into. The goal of the study is to shed light on how machine learning models could be used in the medical field, especially in disease classification and identification. We compared ResNet50, InceptionV3, Xception model, Denesenet121 and Mobilenet. Our research results in improved accuracy, precession call and f-1 score in MobileNet and Xception Model. In order to analyse the models’ output, we also discovered a confusion matrix. In Mobile net, we discovered a mean accuracy of 0.97 and a precision of 0.96. The F-1 score was 0.968, and the mean recall was 0.968. The mean precision is 0.989, the mean recall is 0.989, the mean f-1 score is 0.98 and the accuracy is 0.986 in the Xception model.