Authors: C. Wang∗, J. Choi∗, L.F. Urbano, P. Masson‡, M. VerMilyea §, M. Kam.∗ ∗New Jersey Institute of Technology, Newark ; ‡Penn Fertility Care, Philadelphia; §Ovation Fertility, Austin.
Summary abstract: Human sperm motility analysis is a key method in assessing male fertility. It was suggested that the performance of automatic sperm motility analysis systems can be enhanced by adopting multi-target tracking algorithms developed originally for radar technology. We reviewed and appraised several target tracking algorithms operating on synthetic and actual sperm images, and compared their performance. Simulations and observations of images of real sperm cells suggest that the joint probability data association filter with track-coalescence-avoiding (JPDA*) outperforms other evaluated algorithms. This is also the result obtained on images of swimming tadpoles. Results demonstrated that the JPDA* outperforms the NN, the PDA and the JPDA algorithms in these applications.
What is known already: Measures of sperm motility have been used for decades to evaluate male fertility in clinical andrology. Several techniques are being used for estimation of sperm motility. The most common method uses microscopes or cameras operated by technicians who count sperm cells and assess their characteristics using manual measurements. In this process, sperm motion quality is appraised visually according to standard protocols. Another class of widely used tools is computer-assisted sperm analysis (CASA) systems. CASA systems have the potential to provide fast, automatic and more objective sperm analysis than analysis relying on human operation and have been the subject of several studies. Some authors have opined that some commercial CASA machines are deficient in their ability to dispose track coalescences, which diminish their ability to perform sperm trajectory reconstruction. If the wrong track reconstruction data are used in the analysis, values of key kinematic parameters may not be calculated correctly.
Participants/materials, setting and methods: For this study, we created video simulations of human sperm motion and used them to compare four algorithms applied to automatic sperm analysis. The mean optimal sub-pattern assignment (OSPA) distance was used as performance criterion. Four multi-target tracking algorithms were compared: the nearest-neighborhood (NN), probabilistic data association (PDA), joint probabilistic data association (JPDA) and joint probabilistic data association with track-coalescence-avoiding (JPDA*). Performance of different algorithms was also compared using images of real human sperms and swimming tadpoles video clips.
Main results: The JPDA* algorithm met or exceeded the performance of the NN, the PDA and the JPDA algorithms in all studied scenarios.
Wider implications of the findings: Radar tracking algorithms such as the probabilistic data association (PDA) filter and the joint probabilistic data association (JPDA) filter have potential to assist in automatic sperm analysis systems.