- Preprint Article
- 10.17504/protocols.io.14egn6p4ql5d/v2
Optimizing CASA Data: Transforming Sperm Coordinates into Long Format for Enhanced Machine Learning Analysis v2
- Jan 27, 2026
- Cindy Rivas Arzaluz + 3 more +3
Some Computer-Assisted Sperm Analysis (CASA) systems allow the retrieval of detailed data for each sperm analyzed during a capture routine. By "capture routine," we refer to the process of recording a video sequence for a defined period, typically one or two seconds. The data obtained include traditional kinematic or motility parameters, such as VCL (Curvilinear Velocity), VAP (Average Path Velocity), VSL (Straight-Line Velocity), LIN (Linearity), STR (Straightness), BCF (Beat Cross Frequency), and ALH (Amplitude of Lateral Head Displacement). Additional parameters may also be available, depending on the CASA system's manufacturer and software version. Since motility parameters are derived from coordinate data, they serve as condensed representations of sperm kinematic behavior. Consequently, the coordinate data contains a wealth of additional information, enabling not only the reconstruction of motility parameters but also the trajectories followed by individual sperm. It is important to note that these trajectories cannot be reconstructed solely from motility parameters. Thus, we emphasize that each trajectory has an associated set of motility parameters (Rodríguez-Martínez et al., 2023). Despite the inherent richness of coordinate data, current methodologies for identifying kinematic subpopulations in datasets have relied exclusively on motility parameters (Ramón and Martínez-Pastor, 2018). Coordinates can also be used to reconstruct the trajectories of individual sperm analyzed in a CASA system. These trajectory images can subsequently serve as input for machine learning algorithms, which can cluster the images into groups (subpopulations). These subpopulations can then be statistically characterized based on their associated motility parameters. In this protocol, we describe how we constructed a coordinate dataset to serve as input for machine learning algorithms, specifically the one implemented by Rodríguez-Martínez et al. (2023). The data corresponds to coordinates of hamster sperm analyzed using a CASA system (SMAS, Version 3.18), with a capture speed set at 50 fps for one second (Fujinoki M, personal communication). Each capture routine in the SMAS system generates two files: the first contains motility parameter data, while the second contains the coordinates of the detected sperm. The procedure comprises three stages: (1) acquisition and initial adjustments, (2) adding identifiers, and (3) constructing the final dataset. Files are saved with the “.ods” extension (compatible with LibreOffice Calc), and the readODS library is used to import them into the analysis workflow.
Read more