Convolutional Siamese network-based few-shot learning for monkeypox detection under data scarcity

Abstract

Monkeypox is a rare zoonosis and has been recognized as a public health emergency of international concern by the World Health Organization (WHO). Clinically, monkeypox shares many similarities with smallpox and varicella. How to accurately diagnose monkeypox is a problem faced by medical staff. Since the discovery of monkeypox, related studies have been few and ignored for a long time. Research on monkeypox is limited by the lack of available data. In order to overcome the problem of sample scarcity, we propose a few-shot learning method based on the VGG16 Siamese network model, which aims to classify monkeypox and similar skin disease images with a small number of samples. The model contains two identical VGG16 subnetworks with shared weights for extracting feature vectors of input image pairs. The distance between feature vectors is measured to determine whether the input pair belongs to the same category. We validate our model on the Monkeypox Skin Lesions dataset (MSLD). The accuracy of monkeypox image classification reaches 90.32%, and the AUC reaches 90.60%. Our proposed model has the best performance among similar studies. This confirms the effectiveness of the few-shot learning method based on convolutional Siamese networks for image classification tasks when data is scarce. This method can be used to solve the problem of monkeypox detection and is expected to be extended to the diagnosis of other rare diseases.