Clustering of Human Sperm Swimming Patterns in Time-Lapse Images

In observations of sperm movements in 2-dimensional time-lapse images, the presence or absence of episodic rolling of the sperm cells creates different types of progressive swimming patterns. Development of automatic tracking of sperm trajectories in time-lapse images provides an opportunity to investigate these patterns. In this study, we cluster sperm cells by swimming types, using motility parameters calculated from sperm swimming tracks obtained by the joint probability density association filter (JPDAF). We apply k-means clustering and artificial bee colony (ABC) algorithm search on synthetic and real sperm swim data to identify the different swimming types. The result is clusters with interpretable distinctive features, demonstrating the potential to provide a clustering tool for automated sperm subpopulation analysis.