Monkeypox Skin Lesion Detection Using Deep Learning Models
Selen Gürbüz, Galip Aydın
Abstract
In the last two years, monkeypox virus has started to appear as the biggest epidemic threat after the COVID19 epidemic. Cases have now been reported in more than forty countries outside of Africa. In cases where Confirmatory Polymerase Chain Reaction (PCR) testing is not possible, studies for detection by image analysis are extremely important. Within the scope of this study, Kaggle Monkeypox Image dataset, which is available as open source, was used. In order to increase the sample dataset, firstly, data replication methods were applied to the images. The results of 5 pre-trained deep learning models (DesNet121, ResNet50, Xception, EfficientNetB3, EfficientNetB7) for the detection of Monkeypox virus are presented comparatively. The success of the methods is demonstrated by accuracy, recall, precision, F1 score and confusion matrix. The detection accuracy rates of DesNet121, ResNet50, Xception, EfficientNetB3, EfficientNetB7 methods are 72%, 75%, 73%, 82% and 90%, respectively. The detection success rates obtained have shown that it is a supportive practice for physicians for rapid screening.