CA-LIME: An Algorithm that Explains Classification Results of Cardiomyocyte Beating

Cardiovascular diseases are the leading cause of death and form a growing health concern. For the development and screening of novel drugs in a high-throughput manner, classification systems are developed to classify beating cardiomyocytes as healthy or diseased. With outstanding performance on other medical tasks, the use of machine learning for this application is increasingly explored. However, the predictions of such system are difficult to interpret for humans and provide no insight into the difference between healthy and diseased cardiomyocytes. We propose Contraction Analysis LIME (CA-LIME) to make the decisions made by machine learning algorithms more interpretable. CA-LIME is specifically designed to handle contraction analysis data. The applicability of CA-LIME is shown using samples treated with drugs, showing the explanation are in accordance with their known effects. Finally, explanations are generated for samples derived from a patient to show the applicability to patient specific diseased condition.