Machine learning for virus-like particle segmentation and classification in electron microscope imaging

The development of virus like particle (VLP) in the pharmaceutical industry is used in the preparation of vaccines that mimic the structure of a virus. Depending on their structure, these multi-protein constructs are categorized into three types: single shell, multi-layer, and coiled ribbon. In the present thesis two machine learning models as well as a standard image processing model have been designed in order to automate the discrimination and counting of the three particles types in a VLP suspension, with an aim at replacing the traditional and time-consuming manual counting of the different particles appearing on electron microscopy images. The performances of each model are analysed and their accuracies are compared.