Machine Learning to Automate the Visual Interpretation of Chemical Agglutination Tests

Many point-of-care tests rely on visual changes in color, shape, and size to convey results that can be read by the naked eye. One category of such tests is an agglutination test (AT), which relies on the clumping of micro-particles or cells in the presence of a target analyte. Although visual inspection is convenient and fast, it is subjective, prone to errors, and limits decision-making to coarse-grained results. We present an open-source software framework designed to facilitate the development and interpretation of ATs. This framework includes a web-based annotation interface for curating new image datasets, a computer vision pipeline that extracts informative AT features, and a machine learning module that allows AT developers to study how an AT agglutinates over time during future experiments. We present two case studies of our framework being used to develop and interpret tests.