Date of Award
Doctor of Philosophy in Computer Engineering - (Ph.D.)
Electrical and Computer Engineering
Semen analysis is performed routinely in fertility clinics to analyze the quality of semen and sperm cells of male patients. The analysis is typically performed by trained technicians or by Computer-Assisted Semen Analysis (CASA) systems. Manual semen analysis performed by technicians is subjective, time-consuming, and laborious, and yet most fertility clinics perform semen analysis in this manner. CASA systems, which are designed to perform the same tasks automatically, have a considerable market share, yet many studies still express concerns about their accuracy and consistency. In this dissertation, the focus is on detection, tracking, and classification of sperm cells in semen images, key elements of CASA systems. The objective is to improve existing CASA algorithms and systems by applying validated computer vision, tracking, and computational intelligence algorithms.
The first step of the study is the development of simulation models for generating synthetic images of semen samples. The images enable the assessment of CASA systems and their algorithms. Specifically, the simulation models generate time-lapse images of semen samples for various sperm image categories and include ground truth labels. The models exploit standard image processing operations such as point spread functions and 2D convolutions, as well as new models of sperm cell swimming, developed for this study. They embody multiple studies of sperm motility in the form of parameterized motion equations. Use cases are presented to use the swimming models and the simulated images to assess and compare algorithms for sperm cell segmentation, localization, and tracking.
Second, a digital washing algorithm is presented for unwashed semen samples. Digital washing has the potential to replace the chemical washing techniques used by fertility clinics at present, which are costly, time-consuming, and unfriendly to the environment. The digital washing algorithm extracts features from moving sperm cells in an image, and uses these features to identify all sperm cells (moving and stationary) within each studied image (simulated or real). The effectiveness of the digital washing algorithm is demonstrated by comparing the performance of the proposed algorithm to other cell segmentation and detection techniques.
Third, a classification algorithm for sperm cells is developed, based on their swimming patterns. The classification algorithm uses K-means clustering on a subset of motility parameters of sperm cells selected by the Artificial Bee Colony (ABC) algorithm. Results of classification and clustering are shown, using simulated and real semen images. Swimming pattern classification has the potential to increase understanding of the relationship between the distribution of sperm cell swimming modes in a patient’s semen image and the fertility of that patient.
Lastly, a new method is presented to calculate motility parameters from sperm tracks. The movement of sperm cell is modeled as a sinusoidal traveling wave (“traveling sinusoid”). The amplitude and average path of a moving cell are estimated using an extended Kalman filter (EKF). The states estimated by the EKF include position, velocity, amplitude, and frequency of the traveling wave. The motility parameters calculated from this approach are shown to be superior to those calculated by other existing methods in terms of their accuracy and consistency.
CASA developers will find in this study (and in the software made available) new tools to improve the performance of their designs, and to compare and contrast different proposed approaches and algorithms.
Choi, Ji-Won, "Improving the performance and evaluation of computer-assisted semen analysis" (2022). Dissertations. 1598.