Mpox Skin Lension Detection using Deep Learning Approaches

The Mpox contagion is an orthopoxvirus that causes Mpox( Mpox), a complaint with symptoms analogous to smallpox, although less severe. Mpox( Mpox) is a viral complaint that affects humans and creatures causing a rash and skin lesions in humans and creatures. The identification and opinion of Mpox lesions are important for the proper treatment and operation of the complaint. Due to the effect of Covid outbreak which effected human life, rearmost start of Mpox has now become a cause of concern for healthcare providers all over the world. So we need to have prior thinking or early analysis to slow down the progression and spread of Mpox. In this study, we used the Convolutional Neural Network( CNN) models to classify Mpox skin lesion images into different orders. The dataset used for detecting Mpox has been given by the CSE Department, Islamic University, Kushtia- 7003, Bangladesh. We employed the Mpox skin lension dataset intimately available from Kaggle( 770 images). This skin lension dataset consists of four classes similar as Mpox, Chickenpox, Measles, and Normal skin images. In the coming step, several pre-trained deep literacy models, videlicet, VGG- 16, VGG- 19, ResNet50V2, ResNet101V2, Xception, DenseNet121, and MobileNetV2 to identify the most accurate model for classifying the images. We also performed data addition ways, including gyration, zooming, and flipping, to increase the size of the training dataset and ameliorate the model’s conception capability. Xception achieves the best overall accuracy of 89.24%, while DenseNet121 achieved Test accuracy : 87.34%, ResNet101V2 achieved Test accuracy of 84.17%, ResNet50V2 achieved Test accuracy of 87.34%, MobileNetV2 achieved Test accuracy of 86.07%, VGG-16 achieved Test accuracy of 82.27%, VGG-19 achieved Test accuracy of 81.01% respectively. Although the preliminary results obtained from the limited dataset are encouraging, a more diverse dataset covering a wider demographic is needed to improve the models generalizability.