EMViR - Electron Microscopy Image Dataset Of Labelled Respiratory Viruses For Deep Learning Model Development

Diagnostic electron microscopy is a valuable tool for supporting diagnosis and research of pathogenesis of infectious diseases. To advance the application and the development of deep learning in virus detection from electron microscopic images, more publicly available datasets are essential. In this work, we present EMViR, a diverse dataset of 1,084 negative staining electron microscopic images containing virus particles from the virus families Adenoviridae, Coronaviridae, Orthomyxoviridae, and Paramyxoviridae/Pneumoviridae, which are associated with respiratory diseases, along with their corresponding segmentation masks. We also provide a quantitative description of the dataset and propose a data split. Additionally, we present results for semantic multiclass segmentation using different segmentation model architectures. Link to the EMViR data set: https://doi.org/10.5281/zenodo.13949331