In my thesis I investigated automatic motility analysis of human semen. The investigation was conducted in three studies.

First, I investigated how to detect and segment sperm cells in bright field microscopy images from the Xcyto 10 image cytometer. I developed a pixel-wise segmentation and detection algorithm based on the use of convolutional neural networks achieving high pixel-wise accuracy, precision and recall.

Second, I studied how to conduct an unbiased estimation of the motility distribution of sperm cells and whether sperm cells can be tracked sufficiently reliably to obtain accurate motility distributions in practice. The study was conducted by analysing a set of semi-automatically annotated sperm cell tracks. Based on the study I recommended a set of guidelines for conducting unbiased motility estimation. I combined our detector from the first study with an existing linker method to obtain an automatic method for tracking of human sperm cells. Using this tracker I obtained motility distributions nearly identical to the theoretical distributions.

Third, I evaluated the automatic system for conducting motility analysis of human sperm by comparing it with manual motility analysis resulting in comparable results. However, more data needs to be collected before finally concluding whether the system can be used during routine analysis