Monkeypox Skin Lesion Classification Using Transfer Learning Approach

Abstract

Monkeypox is classified as a viral zoonotic disease which is transmitted to humans from animals. The recent outbreak of the Monkeypox virus has affected more than 40 countries. With the rapid spread and ever-growing challenges of provisioning PCR (Polymerase Chain Reaction) Tests in areas with less availability, computer aided methods incorporating Deep Learning techniques for automated detection of skin lesions proves to be a feasible solution. The paper proposes a Transfer Learning based approach to classify Monkeypox skin lesions from chickenpox and normal skin images. A total of 5 Transfer Learning models namely- MobileNetv2, ResNet50, Inceptionv3, EfficientNetB5 and Xception have been trained on a skin lesion image dataset sourced from News reports, public health websites and case studies. A comparison of the trained models is provided to select the best performing model which can be further utilized in any application for quick, automated detection of monkeypox skin lesions in remote areas. MobileNetv2 provided the best model accuracy of 98.78% for classification of monkeypox skin lesion images.