Sperm Recognition and Viability Classification Based on Artificial Intelligence

Semen quality analysis plays an important role in evaluating male reproductive health and is essential before artificial insemination. The computer assisted semen analysis system (CASA), as an auxiliary tool for manual semen analysis, provides physicians with relatively accurate information on sperm motility, but there are more uncertainties in the sperm motility process, and sperm obstruction and occlusion occur frequently, which makes CASA unable to show optimal tracking performance in complex environments. In this paper, we propose a new method based on YOLOv5 for sperm head detection and using joint probabilistic data association (JPDA) for multi-sperm dynamic tracking, which is used to enhance the tracking effect of CASA in the case of intersecting sperm trajectories. The YOLOv5 algorithm is used as a sperm head position detector, and the detection accuracy reaches 97.9%. Using JPDA algorithm for sperm tracking, the ID switching problem of sperm during swimming was better avoided, and the multiple object tracking accuracy (MOTA) reached 91%. Finally, the motion parameters of sperm were calculated, and the viability grading of sperm was achieved by back propagation (BP) neural network.