Abstract
This work offers researchers the first version of an open-source sperm tracker software (Sperm Motility Tracker, V1.0) containing a novel suit of algorithms to analyze sperm motility using ram and buck sperm as models. The computer-assisted semen analysis is used in several publications with increasing trend worldwide in the last years, showing the importance of objective methodologies to evaluate semen quality. However, commercial systems are costly and versatility is constrained. In the proposed method, segmentation is applied and the tracking stage is performed by using individual Kalman filters and a simplified occlusion handling method. The tracking performance in terms of precision (number of true tracks), the percentage of fragmented paths and percentage of correctly detected particles were manually validated by three experts and compared with the performance of a commercial motility analyzer (Microptic’s SCA). The precision obtained with our sperm motility tracker was higher than the one obtained with a commercial software at the current acquisition frame rate of 25 fps (P < 0.0001), concomitantly with a similar percentage of fragmentized tracks (P = 0.0709) at sperm concentrations ranging 25–37 × 106 cells/mL. Moreover, our tracker was able to detect trajectories that were unseen by SCA. Kinetic values obtained by using both methods were contrasted. The higher values found were explained based on the better performance of our sperm tracker to report speed parameters for very fast motile sperm. To standardize results, acquisition conditions are suggested. This open-source sperm tracker software has a good plasticity allowing researchers to upgrade according requirements and to apply the tool for sperm from a variety of species.
Introduction
Motion analysis on quality assessment of semen samples is of great importance for the positive association with male fertility and because it is in one of the most affected parameter after cryopreservation. However, sperm tracking is quite complex due to cell collision, occlusion and missed detection. Computer-assisted semen analysis (CASA) systems are used in several publications with an increasing trend worldwide in the last years, showing the importance of objective methodologies to evaluate semen quality and predict fertility. It is well known that CASA systems are commonly used for determination of sperm quality from various species (Billard & Cosson 1992, Dietrich et al. 2005), cryopreservation effectiveness (Cueto et al. 2016), toxicity bioassays, prediction of fertility potential or research related to basic sperm biology (Muiño Otero 2008, Buzón Cuevas 2014).
CASA systems provide sequential digital images of each spermatozoa track allowing individual motion analyzing thus facilitating a rapid, precise and accurate assessment of several and meaningful kinetic measurements (Verstegen et al. 2002, Amann & Waberski 2014) that are considered as objective and reproducible, while using identical instrument settings. On the other hand, it has been recognized that among commercial software disadvantages, one can mainly list high cost, need to regularly upgrade and dramatic changes influenced by different settings that are not well documented in publications (Schleh & Leoni 2013). Even when each lab standardizes its own conditions, the setup of the parameters is crucial to allow comparisons between different studies and to obtain reproducibility as well as consistency of internal and external controls (Holt et al. 1994, Fraser 1998). Since there are many factors affecting CASA performance (Broekhuijse et al. 2011), the methodologies and system specificities (equipment, chamber, plate temperature and acquisitions details) have to be fully and clearly described (Verstegen et al. 2002). However, these details are not often given in most publications. Moreover, the accuracy of CASA results is intrinsically dependent on the range of sperm concentrations analyzed (Muiño Otero 2008, Talarczyk-Desole et al. 2017).
Another fact that has to be considered is that motility estimates and concentration using CASA systems are highly influenced by the counting chamber (Hoogewijs et al. 2012, Palacín et al. 2013). Besides spermatozoa speeds vary according to each species, the choice of a particular acquisition velocity is under discussion, since the selected frame rate affects the measure of several kinetic parameters (Davis & Katz 1992, Verstegen et al. 2002). Verstegen et al. (2002) described that trajectories are not well detected when setting of maximum velocity is too low; in these cases, the software generates wrong trajectories since it connects points belonging from different spermatozoa tracks. In most of the cases, a good measure of a high curvilinear velocity (VCL) value is due to a good frame rate setting.
Concerning costs, there are also some open-source systems that are widely useful in sperm motility analysis, e.g. National Institutes of Health has developed a CASA plugin for the ImageJ software (Wilson-Leedy & Ingermann 2011) that has been especially adapted for the kinetic analysis of fish sperm (Verstegen et al. 2002) but also validated for mammalian sperm (Giaretta et al. 2017). Disadvantages of this method include many manual settings, needing to apply different thresholds to each video.
The aim of this work was to develop an automated particle detection tool and a suite of tracking algorithms to analyze motility parameter characteristics using ram and buck sperm as models. Our tool has the clear advantage that is plausible to be extrapolated to other species due to its plasticity to perform changes depending on the researcher’s objectives and the intrinsic characteristics of the samples. Moreover, this prototype is useful to track each spermatozoon since the corresponding trajectory is drawn step by step through the image sequence.
In this way, we developed a sperm tracker software containing a suite of algorithms for sperm motility analysis that includes the stages of detection (frame to frame), tracking and motility analysis for videos of ram and buck sperm cells. A manual validation was performed to compare the tracking performance of our algorithm with that of an available version of the Microptic’s Sperm Class Analyzer-SCA (Microptic, Barcelona, Spain) over the same videos. This work offers open-source software to evaluate semen motility for researchers in the reproductive field.
Materials and methods
Samples collection
Animal handling was performed in accordance with Spanish Animal Protection Regulation, RD 53/2013, which conforms to European Union Regulation 2010/63. Blanca Celtibérica buck and Manchega ram (age >1.5 years) were maintained in a semi-free-ranging regime at El Campillo (Elche de la Sierra, Albacete, Spain) and at experimental farm of University of Castilla-La Mancha respectively. The collection of ejaculates was performed using two different methods: artificial vagina for ram (five males) or electroejaculation for buck (five males), according to the guidelines RD 841/2011 and protocols previously described (Marco-Jimenez et al. 2008, Jimenez-Rabadan et al. 2012). Ram samples were collected and pooled, whereas samples from buck were analyzed individually.
Sperm concentration was calculated by Bürker chamber counting and adjusted to 30 × 106 spermatozoa/mL for ram and 20 × 106 spermatozoa/mL for buck with PBS at 37°C.
Experimental procedure
Objective motility was assessed with a Makler counting chamber (10 µm depth) and samples were observed using a 10× objective (negative phase-contrast field). Each analysis captured several fields with a Basler A302fs digital camera (Basler Vision Technologies, Ahrensburg, Germany) connected to a computer by an IEEE 1394 interface. The image size was 768 × 576 pixels. The acquisition frame rate was set in 25 fps videos, which were simultaneously analyzed by CASA using the Sperm Class Analyzer software (SCA 2002, Microptic, Barcelona, Spain) and by our sperm motility tracker software. Buck sperm-tracking videos produced by our algorithm are available at Vimeo homesite (see references https://vimeo.com). The motility parameters assessed are described in ‘Motility parameters and motility analysis’ section.
Algorithm development
Detection of the cells head
Image processing algorithms were developed in C++ with the Netbeans IDE and using the OpenCV 3.2.0 library. A detection method similar to the one used by Buchelly et al. (2016) for cell segmentation was used but with a highlighting step due to opening Top-Hat (Serra 1982). For the Top-Hat transform, we used a circular structuring element with the sufficient size to enclose one spermatozoon head (11 × 11 pixels). The fixed threshold to obtain the binary image was set to 30. The structuring element used for the binary morphological filter is circle shaped and it has a size of 5 × 5 pixels to remove little noise points, sharp features like the sperm tails and to separate some particles.
Concentration measurement
Cell concentration was determined for each sample video as the average number of cells in each frame per square millimeter (cells/cm3), according to Equation (1):
Here, L and d are the setting parameters and depend on the experimental conditions: L is the length of the side of the grid square in pixels and d is the counting chamber depth in micrometers and w and h are the image width and height respectively in pixels. So, the first factor in Eq. (1) refers to the transformation of the lengths from pixels to metric units. On the other hand, the second factor shows the average number of cells in the video sequence determined by the numbers of cells in each frame (n k ) and the total number of frames, N.
Sperm cells tracking
In order to define an object’s model, kinematic variables, shape or geometric descriptors, contours, gray levels or textures can be considered (Lucena López 2003, Azari et al. 2011, Liu et al. 2013, Jeong et al. 2014, Sahbani & Adiprawita 2016). From this set of data, the model is represented by the state x i of the system at the instant i with a given number of degrees of freedom (Lucena López 2003). Our object’s model consisted only on the head centroid or mass center coordinates (Gárate Polar 2015){Gárate, 2015 #1681;Gárate Polar, 2015 #182;Gárate Polar, 2015 #182} and its velocity components. As it does not rely on the geometry of the cell head or on gray levels information, a spermatozoon was treated as a point particle. The dynamics of the system was studied with a first order model, i.e. positions and velocities are measured to predict the future positions. The particle trajectory (j) was defined as the discrete collection of positions at all the instants i. The velocity vector of a particle j between the instants i − 1 and i, was determined using Equation (2):
The model of the dynamics offers an a priori distribution of probabilities about all the possible configurations of the current state of the system p(X i ) taking into account the estimated distributions for the previous instants p(X i −1)–p(X i − n ). On the other hand, the temporal fusion method uses the Bayesian framework to integrate the a priori probabilities with the set of measures Z (coordinates of the centroids of the detected cells in the current frame) to find the a posteriori distribution (Lucena López 2003) given by Equation (3):
The objective was to maximize p(X i |Z) in order to estimate the new state (Lucena et al. 2010) or to give the correct labels to the new detected particles according to the previous known ones. In Eq. (3), the value p(Z|X i ) is the observation model. Despite its limitations, Kalman filter is ideal to use with Gaussian and unimodal distributions (Lucena López 2003), assuming constant or low acceleration rates (Vinaykumar & Jatoth 2014). We associated Kalman filter to each detected particle to predict its future position (Catlin 1989, Azari et al. 2011, Jeong et al. 2014), as follows: let the state of a single particle j at the instant i, and the measurement vector. The state and the measurement are estimated by using Equations (4) and (5):
where A j is the transition matrix for the particle j and has the values shown in Equation (6). On the other hand, H j is the measurement matrix and for this work it corresponds to the identity matrix I ϵ R 4×4; ε j and δ j and are vectors corresponding to the process noise and the measurement noise, respectively. The noise vectors are initialized with a constant value and updated during the execution time.
We used the built-in functions included in the OpenCV library to create and use Kalman predictors. Each of them functions in a cyclic process that consists of three stages and each stage is complemented with a particular routine for our own purposes. First, the system obtains real measures of the state variables and compares them with the measures predicted by the Kalman filters in the previous iteration to do the association by a minimal distance criterion and thus get indirectly the maximization of the a posteriori distribution p(X i , j ,Z j ) for each particle. The result of the first stage gives the cell path over which the motility analysis described in the following section is performed. In the second stage, the system gives to the Kalman predictors the new real data to correct the state X i , j and to update the error vectors ε j and δ j and the covariance matrices that are involved in the inner operations. At the third stage, Kalman filters predict the possible future location of each labeled particle by having its state in the current instant, i.e. finding the values of Z j that maximize p(Z j |X i , j ) and that is used in the first stage of the next iteration.
Motility parameters and motility analysis
Motility of each spermatozoon was defined by its current head velocity descriptors ( Muiño Otero 2008, Buzón Cuevas 2014):
-
VCL: Velocity over the total distance moved in the path length, i.e., including all oscillations that occur in the head track. A ram spermatozoon is considered immotile if it has a VCL less than 10 μm/s, according to Eq. (7).
-
Average path velocity (VAP): Velocity over a calculated, smoothed (low pass filtered) path, i.e., a shorter distance than that used for calculating VCL.
-
Straight-line velocity (VSL): Velocity calculated using the straight-line (Euclidean) distance between the beginning and end of the sperm track.
-
Amplitude of lateral head displacement (ALH): The average value of amplitude of the oscillatory movement of the sperm head in each beat cycle.
-
Beat Cross Frequency (BCF): The frequency with which the actual track crosses the smoothed track (regardless of the oscillation direction).
-
Straightness (STR, %): Measure of the oscillation of the curvilinear path with respect to the average trajectory, calculated as VSL/VCL × 100. Indicates the STR of the middle path.
-
Linearity (LIN, %): Relationship between the VSL and the VCL expressed as VSL/VCL × 100.
-
Oscillation (WOB, %): It is a measure of the oscillation of the curvilinear trajectory with respect to the average trajectory, calculated as VAP/VCL × 100).
-
Total motility (%): Percentage of sperm having a VCL >10 μm/s.
-
Progressive motility (MP, %): Percentage of sperm presenting movement with a STR index ≥80% within the sample.
-
Statics: VCL <10 μm/s.
-
Low progressive: 10 < VCL < 45 μm/s.
-
Mid progressive: 45 < VCL < 75 μm/s.
-
Rapid: VCL >75 μm/s.
-
As described by other state-of-the-art works (Rojas et al. 2012, Liu et al. 2013, Gárate Polar 2015, Hidayatullah et al. 2015), the discrete set of positions for each spermatozoon head (j-th particle), VCL j was obtained as the mean VCL, as described by Equation (7), using the notation defined in the previous sections:
where Fr refers to the frame rate in frames per second, L is the side length of the grid square given in pixels used in Equation (1), n j is the number of points of the j-th particle path, u i , j and v i , j are the components of the velocity defined in Equation (2), for the n j − 1 intervals.
VAP calculation depends on the particular method used for obtaining the smoothed path. Our proposed system uses the method mentioned in Hidayatullah et al. (2015) for smoothing the path and accordingly ALH and BCF parameters.
Manual validation
The variables considered were the number of total trajectories, the precision defined as the number of correct paths over total paths detected (Equation (8)) and the percentage of fragmented trajectories.
where TP represents the number of true positives or good tracks and FP is the number of false positives or wrongly assigned tracks. To classify a track as good or bad, we used the criterion of three independent expert biologists who performed the manual validation for each path considering whether the labels were correctly conserved during occlusion states.
The percentage of correctly detected particles is defined as the number of detected sperm over the total particles labeled by each software according to the criteria of three experts.
Statistical analysis
Data were analyzed by GLMM (generalized linear mixed effect model) to determine the statistical significance between both software (Zuur et al. 2009). Data associated to cell percentages were analyzed through models with binomial distribution, whereas the number of trajectories was analyzed by models with Poisson distribution. Velocities were analyzed with Gaussian error distribution. Normality of residuals was assessed by plotting theoretical quantiles vs standardized residuals (Q–Q plots). Homogeneity of variance was evaluated by plotting residuals versus fitted values. All analyses were performed using R software version 3.3.3 (RCoreTeam 2017), with the ‘lme4’ package for Poisson and binomial models and the ‘nlme’ package for Gaussian models (Bates et al. 2015, Pinheiro et al. 2017). For all analyses, statistically significant differences were determined at P < 0.05.
Sperm motility tracker software V 1.0
The software is free, and an executable version will be provided upon request (acesari@mdp.edu.ar). The software’s user interface provides a step-by-step guide for users. Important instrumental considerations and settings for users are also included (Table 1). Running times are suitable for standard laptop computers with i3 processor and at least 3 GB of memory. Screen resolution can vary between 1280 × 800 and 1920 × 1080.
Requirements for motility acquisition and analysis.
| Parameter | Characteristics |
|---|---|
| Chamber depth | 10 µm |
| Maximum number of cells per field | ≤120 |
| Optimal sperm concentration | ~35 × 106 cells/mL |
| Acquisition frame velocity with the camera | ≥25 fps/s* |
| Video recording time | 5 s |
| Microscope setting | Phase contrast, 10× |
| Input to the software | Sequence of time-lapse images (MP4, AVI or MOV video) |
| System requirements to run software | I3 processor, 3 GB RAM, 1280 × 800 screen resolution |
*In this work, the kinetic values were compared between methods for 25 fps/s due to camera limitations.
The input to our algorithm software is a sequence of time-lapse images currently encoded either as an MP4, AVI or MOV video file of 5 s acquired at 25 fps. The output of the algorithm is a database (.XLSX) containing the set velocity parameters, population parameters and sperm concentration; a movie (.AVI) with the complete tracks and an image (.BMP) of the tracks.
Sperm tracking videos
Each line indicates the spermatozoa tracked by our sperm tracker software frame to frame. Numbers identify each spermatozoa. Different colors of paths indicate the different sperm velocities (static, low, medium or rapid sperm).
Buck sperm tracking videos can be accessed at: https://vimeo.com/264473468, https://vimeo.com/264473359, https://vimeo.com/264473293, https://vimeo.com/264473195, https://vimeo.com/264473092, https://vimeo.com/264472125, https://vimeo.com/264472024, https://vimeo.com/264471893, https://vimeo.com/264471098, https://vimeo.com/264470218
Ram sperm tracking videos can be accessed at: https://vimeo.com/265053190, https://vimeo.com/264482322, https://vimeo.com/264482904, https://vimeo.com/264483470, https://vimeo.com/264485794, https://vimeo.com/265054082, https://vimeo.com/265055033, https://vimeo.com/265055885, https://vimeo.com/265056710, https://vimeo.com/264487378.
Results
Particles firstly detected and localized with the highest possible accuracy were linked to form particle trajectories. Detected particles had a near elliptical shape although their areas had a low number of pixels (Fig. 1). The small size let us to approximate the spermatozoa heads as point particles and not to consider their shapes as shown within the region in which we could observe centroids detection (Fig. 2). As shown, the intensity degradations avoided the complete detection of the particles’ shape, and thus, this supported the idea of working with the point particle model. It was possible to measure sperm concentration by using (1), having a range of particle concentrations between 12.64 × 106 cells/mL (38.83 cells/frame) and 42.29 × 106 cells/mL (129.92 cells/frame) for the considered samples (Table 2) consistent with sample adjustment (see ‘Materials and methods’ and ‘Sample collection’ sections).
Signal processing for the detection process. Upper panel, a region of interest in a sample frame is selected to explain the detection process. Lower panel, Top-Hat transformation of the selected area (left), binary image obtained by applying a fixed threshold (center) and binary morphological opening to obtain only the heads (right). Bar = 25 µm.
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
Signal processing for the detection process. Upper panel, a region of interest in a sample frame is selected to explain the detection process. Lower panel, Top-Hat transformation of the selected area (left), binary image obtained by applying a fixed threshold (center) and binary morphological opening to obtain only the heads (right). Bar = 25 µm.
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
Signal processing for the detection process. Upper panel, a region of interest in a sample frame is selected to explain the detection process. Lower panel, Top-Hat transformation of the selected area (left), binary image obtained by applying a fixed threshold (center) and binary morphological opening to obtain only the heads (right). Bar = 25 µm.
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
Zoom of a region with detected spermatozoa heads and their centroids. Bar = 10 µm.
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
Zoom of a region with detected spermatozoa heads and their centroids. Bar = 10 µm.
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
Zoom of a region with detected spermatozoa heads and their centroids. Bar = 10 µm.
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
Performance of the proposed tracker software (SMT) compared to SCA system.
| Animal | Video no. | Concentration (106 cells/mL) | Software proposed method | SCA system | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| % SCA tracks followed by SMT | Evaluated tracks ± s.e. | % Correctly detected particles (over total labeled particles) | % Fragmented paths ± s.e. | % Precision ± s.e. | Evaluated tracks ± s.e. | % Correctly detected particles (over total labeled particles) | % Fragmented paths ± s.e. | % Precision ± s.e. | |||
| Ram | 1 | 32.49 | 76.4 ± 10.1 | 158 ± 0 | 100 ± 0.0 | 4.2 ± 0.7 | 82.3 ± 3.3 | 102 ± 5 | 84.0 ± 4.2 | 6.5 ± 3.7 | 66.1 ± 3.5 |
| 2 | 28.99 | 81.1 ± 14.5 | 131 ± 8 | 93.1 ± 5.4 | 1.3 ± 1.1 | 85.1 ± 4.1 | 84 ± 14 | 74.3 ± 12.3 | 5.7 ± 2.8 | 76.9 ± 4.8 | |
| 3 | 36.99 | 81.0 ± 11.9 | 165 ± 3 | 89.7 ± 1.6 | 1.8 ± 1.8 | 82.8 ± 8.1 | 93 ± 29 | 73.5 ± 22.8 | 5.0 ± 4.2 | 74.3 ± 8.9 | |
| 4 | 32.60 | 78.2 ± 14.2 | 155 ± 19 | 88.4 ± 10.6 | 3.1 ± 2.8 | 82.5 ± 4.6 | 84 ± 30 | 64.9 ± 23.0 | 5.8 ± 5.6 | 81.8 ± 3.0 | |
| 5 | 29.45 | 83.6 ± 9.8 | 151 ± 11 | 89.9 ± 6.8 | 3.4 ± 3.3 | 86.8 ± 4.6 | 72 ± 14 | 71.6 ± 13.5 | 5.8 ± 2.1 | 85.9 ± 4.8 | |
| 6 | 35.40 | 79.4 ± 1.3 | 179 ± 1 | 99.6 ± 0.3 | 12.7 ± 5.2 | 86.8 ± 2.8 | 108 ± 19 | 82.2 ± 14.4 | 8.9 ± 2.5 | 69.2 ± 9.6 | |
| 7 | 31.80 | 74.3 ± 3.5 | 152 ± 2 | 98.7 ± 1.1 | 9.4 ± 4.4 | 85.1 ± 4.3 | 95 ± 14 | 79.4 ± 11.7 | 6.2 ± 3.6 | 67.8 ± 11.2 | |
| 8 | 32.92 | 82.0 ± 4.9 | 157 ± 1 | 98.1 ± 0.6 | 11.7 ± 4.9 | 79.4 ± 5.0 | 92 ± 6 | 76.0 ± 5.0 | 11.1 ± 6.7 | 69.1 ± 3.7 | |
| 9 | 30.82 | 80.9 ± 2.4 | 150 ± 3 | 98.5 ± 1.7 | 13.1 ± 0.8 | 90.9 ± 4.1 | 99 ± 11 | 82.9 ± 9.3 | 14.9 ± 9.9 | 68.0 ± 12.4 | |
| 10 | 25.70 | 88.4 ± 10.7 | 118 ± 3 | 85.3 ± 1.8 | 6.5 ± 6.0 | 88.2 ± 5.9 | 74 ± 4 | 78.7 ± 4.6 | 8.5 ± 5.6 | 79.6 ± 5.5 | |
| Buck | 1 | 19.82 | 98.2 ± 0.1 | 65 ± 2 | 82.7 ± 2.6 | 6.5 ± 5.7 | 90.8 ± 3.3 | 54 ± 2 | 92.1 ± 3.5 | 0.6 ± 1.0 | 94.4 ± 2.1 |
| 2 | 21.41 | 99.5 ± 0.9 | 83 ± 2 | 96.51 ± 2.0 | 4.4 ± 4.4 | 92.8 ± 2.4 | 60 ± 4 | 90.0 ± 6.0 | 1.1 ± 1.0 | 97.9 ± 2.4 | |
| 3 | 42.29 | 97.1 ± 0.3 | 165 ± 2 | 98.2 ± 1.2 | 9.9 ± 3.9 | 85.1 ± 1.7 | 127 ± 7 | 92.0 ± 5.1 | 1.8 ± 1.9 | 89.4 ± 3.9 | |
| 4 | 33.81 | 97.1 ± 0.9 | 123 ± 6 | 93.2 ± 4.2 | 6.4 ± 1.2 | 85.3 ± 3.4 | 102 ± 2 | 96.8 ± 2.2 | 5.4 ± 2.7 | 93.9 ± 6.3 | |
| 5 | 34.28 | 99.3 ± 1.1 | 114 ± 5 | 93.2 ± 3.9 | 8.3 ± 4.5 | 90.6 ± 4.7 | 97 ± 4 | 92.9 ± 3.9 | 3.8 ± 1.7 | 91.5 ± 4.9 | |
| 6 | 12.64 | 100 ± 0.0 | 44 ± 1 | 98.5 ± 1.3 | 4.5 ± 2.2 | 97.8 ± 0.0 | 40 ± 1 | 96.7 ± 1.4 | 6.7 ± 1.5 | 91.6 ± 3.8 | |
| 7 | 14.7 | 95.2 ± 1.3 | 54 ± 1 | 98.8 ± 1.0 | 7.3 ± 3.1 | 92.1 ± 1.0 | 49 ± 1 | 94.2 ± 1.9 | 2.7 ± 2.3 | 91.9 ± 3.9 | |
| 8 | 14.09 | 95.3 ± 1.1 | 59 ± 1 | 99.4 ± 1.0 | 17.5 ± 9.9 | 89.8 ± 5.1 | 50 ± 2 | 92.0 ± 2.8 | 0.0 ± 0.0 | 85.8 ± 4.0 | |
| 9 | 19.1 | 97.1 ± 2.6 | 88 ± 3 | 93.4 ± 3.4 | 14.2 ± 5.8 | 87.2 ± 3.2 | 69 ± 2 | 95.4 ± 2.1 | 1.4 ± 2.5 | 94.6 ± 3.2 | |
| 10 | 23.59 | 97.5 ± 1.1 | 107 ± 6 | 93.6 ± 5.3 | 20.9 ± 8.1 | 83.1 ± 2.7 | 80 ± 6 | 89.5 ± 6.6 | 3.4 ± 1.7 | 89.1 ± 2.2 | |
In order to evaluate the tracking performance, the trajectories detected by our proposed algorithm were compared to the ones found by Microptic’s – SCA for two kind of sperm samples: buck and ram fresh ejaculates. A high percentage of the paths tracked by SCA were also followed by our algorithm, and moreover, the number of cells followed by the sperm tracker software was higher than the one obtained with the SCA motility software for both kind of samples (Table 2), suggesting that several sperm particles were only tracked by our method (Fig. 3). The percentage of tracked particles that do not correspond to spermatozoa can vary depending on the quality of the sample, on how clean is the media or on the image quality. In this case, the percentage of correctly detected particles of our proposed method was even higher than the percentage for SCA (Table 2, χ 2 = 489.61, Df = 1, P < 0.0001 for ram and χ 2 = 6.19, Df = 1, P = 0.0128 for buck). Even it is indeed an error, it must be considered as possible, and for this reason, provided that the percentage of undesirable particles is low, both software are equipped with a tool allowing manual curation or elimination of these labels.
Visual comparison between trajectories detected by the SCA vs the trajectories detected by the proposed sperm tracker software. The images correspond to the same video acquired at 25 fps (Video 5 of Table 2, https://vimeo.com/264485794), showing the totality of paths detected by the SCA module (upper panel) or our tracker software (lower panel). Bar = 25 µm. Curvilinear velocity (VCL, µm/s) of some sperm that were not detected with SCA software (yellow arrows) are indicated with labels and arrows in the bottom panel corresponding to our software.
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
Visual comparison between trajectories detected by the SCA vs the trajectories detected by the proposed sperm tracker software. The images correspond to the same video acquired at 25 fps (Video 5 of Table 2, https://vimeo.com/264485794), showing the totality of paths detected by the SCA module (upper panel) or our tracker software (lower panel). Bar = 25 µm. Curvilinear velocity (VCL, µm/s) of some sperm that were not detected with SCA software (yellow arrows) are indicated with labels and arrows in the bottom panel corresponding to our software.
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
Visual comparison between trajectories detected by the SCA vs the trajectories detected by the proposed sperm tracker software. The images correspond to the same video acquired at 25 fps (Video 5 of Table 2, https://vimeo.com/264485794), showing the totality of paths detected by the SCA module (upper panel) or our tracker software (lower panel). Bar = 25 µm. Curvilinear velocity (VCL, µm/s) of some sperm that were not detected with SCA software (yellow arrows) are indicated with labels and arrows in the bottom panel corresponding to our software.
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
Regarding the number of evaluated cells, i.e. those automatically detected and also visually tracked by each expert, the proposed method was higher than the SCA module for both species (χ 2 = 450.75, Df = 1, P < 0.0001 for ram samples and χ 2 = 56.66, Df = 1, P < 0.0001 for buck samples analysis, Table 2).
There are in the literature some measures that allow the performance evaluation for tracking systems, but the ideal disparity test should be given by the comparison with a ground truth or the point to point comparison of each instant for all the tracks, as done by Philip et al. (2014) Fang et al. (2017), or Chau et al. (2004). Unfortunately, the SCA module does not offer that information to compare the differences between each path with the one obtained by our proposed method. According to the criterion of the experts, trajectories were classified as good or bad considering whether the labels were correctly conserved during occlusion states. In this way, our system allows to identify and draw each sperm trajectory frame to frame, representing an advantage over other commercial systems (see ‘Sperm tracking videos’). In terms of performance, a 5-s video showing 120 cells/field acquired at 25 fps is enough to produce a complete data sheet by this algorithm in 30 ms using a 1.3 GHz Intel Core i3 processor with 3 GB 1600 MHz DDR3 RAM. Precision and percentages of fragmented paths were evaluated to compare each system through a manual tracking by three independent experts (Fig. 4 and Table 2). We showed that the performance of our method is similar to the one measured for the Microptic’s SCA Motility module, with a better occlusion handling evidenced by the higher precision (χ 2 = 151.03, Df = 1, P < 0.0001, Fig. 4A) and a similar percentage of fragmented ram sperm tracks (χ 2 = 3.26, Df = 1, P = 0.0709, Fig. 4C). On the contrary, for buck sperm samples, Microptic’s SCA Motility module showed better precision and lower fragmented tracks than our method (χ 2 = 16.99, Df = 1, P < 0.0001 and χ 2 = 95.95, Df = 1, P < 0.0001 respectively, Fig. 4B and E). When the precision of each system or the percentage of fragmented trajectories is plotted depending on the particle concentration, the better precision of our algorithm can be observed at higher concentrations, while SCA was more successful for low concentrations ranges (Fig. 4C and F).
Performance of the sperm tracker software. Percentage of fragmented paths (A, B and C) or precision (D, E and F) for our sperm tracker system (SMT software) compared to SCA (SCA software) at 25 fps. Ram (A and D) or buck (B and E) sperm were analyzed. C and F panels gather results of all the analysis considering cell concentration as a variable. Data assemble the reports of three independent experts that analyzed ten videos for each species. *Value significantly different with respect to SCA software (P < 0.05).
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
Performance of the sperm tracker software. Percentage of fragmented paths (A, B and C) or precision (D, E and F) for our sperm tracker system (SMT software) compared to SCA (SCA software) at 25 fps. Ram (A and D) or buck (B and E) sperm were analyzed. C and F panels gather results of all the analysis considering cell concentration as a variable. Data assemble the reports of three independent experts that analyzed ten videos for each species. *Value significantly different with respect to SCA software (P < 0.05).
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
Performance of the sperm tracker software. Percentage of fragmented paths (A, B and C) or precision (D, E and F) for our sperm tracker system (SMT software) compared to SCA (SCA software) at 25 fps. Ram (A and D) or buck (B and E) sperm were analyzed. C and F panels gather results of all the analysis considering cell concentration as a variable. Data assemble the reports of three independent experts that analyzed ten videos for each species. *Value significantly different with respect to SCA software (P < 0.05).
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
The dataset of kinetic values obtained by using both methods over the same ram recorded samples showed that our method reported higher average speed values (χ 2 = 592.53, Df = 1, P < 0.0001 for VCL, χ 2 = 118.19, Df = 1, P < 0.0001 for VSL and χ 2 = 112.33, Df = 1, P < 0.0001 for VAP, Fig. 5A), average AHL (amplitude of the lateral displacement of the head, χ 2 = 102.55, Df = 1, P < 0.0001, Fig. 5B), BCF (wavelengths of the flagellar beat, χ 2 = 157.09, Df = 1, P < 0.0001, Fig. 5B) and also a higher motile population (χ 2 = 85.17, Df = 1, P < 0.0001, Fig. 5C) compared to SCA reports. Similar results were observed for these kinetic values when buck samples were analyzed (Fig. 5D, E and F, χ 2 = 23.49, Df = 1, P = 0.0004 for VCL, χ 2 = 12.81, Df = 1, P = 0.0003 for VSL, χ 2 = 17.55, Df = 1, P < 0.0001 for VAP, χ 2 = 76.20, Df = 1, P < 0.0001 for BCF, χ 2 = 6.39, Df = 1, P = 0.0115 for ALH and χ 2 = 18.52, Df = 1, P < 0.0001 for total motility). Manual validation showed that SCA failed in tracking and reporting sperm with very high speeds often rendering in non-detected cells (e.g., sperm indicated with yellow arrows in Fig. 3) or otherwise fragmenting the trajectories with low speed assigned to a little stretch (e.g., sperm indicated with yellow arrows in Fig. 6). Accordingly, the report of kinetic values obtained by using both methods over the same cells (i.e. cells tracked by both software) showed comparable values (Fig. 6, green arrows). Consequently, differences can be explained on the basis of the better performance of our sperm tracker to report speed parameters for very fast motile sperm and to the increment in the number of tracks (Table 2) mainly corresponding to motile spermatozoa (Fig. 5C and F).
Report of kinetic parameters analyzed by the commercial CASA system (SCA) and the sperm motility tracker software (SMT) proposed by us. Sperm velocities: VAP, VSL and VCL (A and D), ALH and BCF (B and E) as well as total motility (C and F) were analyzed by both methods (SCA and SMT software) over the same samples. Analysis of ram samples are shown in A–C, whereas analysis of buck samples are shown in D and E. *Value significantly different with respect to SCA software (P < 0.05). ALH, amplitude of lateral head displacement; BCF, beat cross frequency; VAP, average path velocity; VCL, curvilinear velocity; VSL, straight-line velocity.
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
Report of kinetic parameters analyzed by the commercial CASA system (SCA) and the sperm motility tracker software (SMT) proposed by us. Sperm velocities: VAP, VSL and VCL (A and D), ALH and BCF (B and E) as well as total motility (C and F) were analyzed by both methods (SCA and SMT software) over the same samples. Analysis of ram samples are shown in A–C, whereas analysis of buck samples are shown in D and E. *Value significantly different with respect to SCA software (P < 0.05). ALH, amplitude of lateral head displacement; BCF, beat cross frequency; VAP, average path velocity; VCL, curvilinear velocity; VSL, straight-line velocity.
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
Report of kinetic parameters analyzed by the commercial CASA system (SCA) and the sperm motility tracker software (SMT) proposed by us. Sperm velocities: VAP, VSL and VCL (A and D), ALH and BCF (B and E) as well as total motility (C and F) were analyzed by both methods (SCA and SMT software) over the same samples. Analysis of ram samples are shown in A–C, whereas analysis of buck samples are shown in D and E. *Value significantly different with respect to SCA software (P < 0.05). ALH, amplitude of lateral head displacement; BCF, beat cross frequency; VAP, average path velocity; VCL, curvilinear velocity; VSL, straight-line velocity.
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
Visual comparison between trajectories detected by the SCA (upper panel) vs the trajectories detected by the purposed sperm tracker software (bottom panel). The images correspond to the same video (Video 2 of Table 2, https://vimeo.com/264482322) acquired with 25 fps, showing the totality of paths detected with the SCA module (upper panel) and our tracker software (bottom panel). Bar = 25 µm. Curvilinear velocities (VCLs, µm/s) of some paths fragmented by SCA software but tracked correctly by SMT are indicated with labels (1–9) and yellow arrows in the corresponding panels. Tracks with similar VCL (µm/s) comparing SMT to SCA are indicated with labels (10–13) and green arrows in the corresponding panels.
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
Visual comparison between trajectories detected by the SCA (upper panel) vs the trajectories detected by the purposed sperm tracker software (bottom panel). The images correspond to the same video (Video 2 of Table 2, https://vimeo.com/264482322) acquired with 25 fps, showing the totality of paths detected with the SCA module (upper panel) and our tracker software (bottom panel). Bar = 25 µm. Curvilinear velocities (VCLs, µm/s) of some paths fragmented by SCA software but tracked correctly by SMT are indicated with labels (1–9) and yellow arrows in the corresponding panels. Tracks with similar VCL (µm/s) comparing SMT to SCA are indicated with labels (10–13) and green arrows in the corresponding panels.
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
Visual comparison between trajectories detected by the SCA (upper panel) vs the trajectories detected by the purposed sperm tracker software (bottom panel). The images correspond to the same video (Video 2 of Table 2, https://vimeo.com/264482322) acquired with 25 fps, showing the totality of paths detected with the SCA module (upper panel) and our tracker software (bottom panel). Bar = 25 µm. Curvilinear velocities (VCLs, µm/s) of some paths fragmented by SCA software but tracked correctly by SMT are indicated with labels (1–9) and yellow arrows in the corresponding panels. Tracks with similar VCL (µm/s) comparing SMT to SCA are indicated with labels (10–13) and green arrows in the corresponding panels.
Citation: Reproduction 156, 1; 10.1530/REP-17-0755
Discussion
In this work, we presented a new detection and tracking algorithm that can effectively identify immotile as well as motile and progressive sperm heads from two different species, with different concentration ranges and bearing different proportions of motile sperm. We demonstrated that the proposed approach can successfully handle challenges such as cell collision and occlusion, succeeding in multiple sperm tracking, when the spermatozoa concentration up to 42.29 × 106 cells/mL. Our free access tool was validated against CASA SCA, providing similar values of sperm parameters but was more efficient in the number and precision of detected tracks at high concentration ranges, as well as in relation to the lower number of fragmented trajectories.
Some single-particle tracking algorithms have been developed; however, they mostly failed in following them simultaneously when more than ten cells co-exist (Imani et al. 2014, Tinevez et al. 2017). Recently, an automated multi-sperm tracking algorithm capable to detect and track simultaneously hundreds of human sperm cells from two samples was presented with the limitations of long time required to process each video at low acquisition speed and lack of validation against a standardized method (Urbano et al. 2017).
Many cell segmentation methods have been proposed in literature for microscopy image sequences. Some works first binarize the images and others use a matching template. Rojas et al. (2012), Gárate Polar (2015) and Hidayatullah et al. (2015) proposed a fixed threshold and then a binary morphological filter; Buchelly et al. (2016) applied a mathematical morphology gray filter to highlight sperm heads and a later threshold; Liu et al. (2013), Vinaykumar and Jatoth (2014) applied temporal frame differencing, a fixed threshold and a binary morphological filter and others use background subtraction, thresholding and binary filtering (Azari et al. 2011, Jeong et al. 2014). Other approaches also exist that use simultaneous detection and tracking with their own considerations (Karthikeyan et al. 2012, Boryshpolets et al. 2013).
It is consensus that standardization is needed to avoid variations in semen analysis (Palacín et al. 2013). One of the most important settings of the assay is cell concentration. In this sense, Wilson-Leedy and Ingermann (2007) have studied the effect of the cell concentration upon motility measurements, as well as we do, finding that the main limitation is particle density. The widest dynamic range allowed the higher plasticity of the tool, which is critical when considering working with sperm from different species For example, the VCL range for ram motile sperm is between 189.8 ± 40.7 and 39.8 ± 21.0 µm/s (Ledesma et al. 2017), while for fish, the VCL range is between 330 ± 70 and 20 ± 15.0 µm/s for a variety of species (Fauvel et al. 2010, Fabbrocini et al. 2016). This is an advantage when compared to commercial systems that are specie specific. On the other hand, there should be some other adjustments needed for correct sperm identification in other species associated to differences in sperm morphology and size.
According to Lucena Lopez (2003), a typical tracking scheme has four basic essential elements: image features, model of the objects, model of the dynamics and a temporal fusion method. The most of the works in the state-of-the-art use intensity distribution (Karthikeyan et al. 2012, Rojas et al. 2012, Jeong et al. 2014, Gárate Polar 2015, Hidayatullah et al. 2015); however, there also exist more features that can be used like color (Lucena et al. 2010, Fang et al. 2017), motion history (Liu et al. 2013), optical flow (Lucena et al. 2015), frequency descriptors (Pei et al. 2006), and others. As mentioned before, we used the intensity distribution as the image feature required to detect the spatial distribution of cells at each instant, because of the high contrast obtained between foreground and background in the scene.
A common trouble in tracking systems is the occlusions handling. In this situation, two or more objects in the 2D scene get very close to each other and the detection module often considers them as a single object, giving to the system the ambiguity of which label corresponds to this new object and how to treat the absence of the missing others. The tracking scheme must lead with this situation and take a proper decision. The method (Azari et al. 2011) applied template matching in the region of occlusion and used correlation to identify the parts corresponding to each merging individual object. In Jeong et al. (2014), the aspect ratio or width/height is considered to detect when objects are merging or splitting. In Lucena et al. (2010), authors use a combined model of the mean-shift and the CAM-shift algorithms to improve robustness to occlusion. The occlusion handler of Sahbani and Adiprawita (2016) uses the statistics of the blob size (standard deviation) to find an occlusion situation by means of an occlusion threshold. When an occlusion condition occurs due to merging objects, the label of the new particle corresponds to the label of the previous object that presented the closer prediction point to the measured mass center. Meanwhile, the position of the hidden object is predicted during a test interval of six frames with an increasing search radius. Then, if the particle appears during the test interval and inside the search region, it will recover its original label and its path will be completed with the previous estimated locations.
Many similar works have developed solutions to make an automatic motility analysis, both for human and other animals’ sperm. In this work, we take the known methods to determine the motility parameters and own considerations, but we put our major interest in the system performance evaluation. An important fact to consider is the objectivity of the VCL measurement and the subjectivity of the VAP. The VAP parameter depends on the smoothness degree of the spermatozoon trajectory, and there is no information about a unified criterion to perform that low pass filtering operation and, in consequence, each system performs it in a different way giving probably different results for the same sample video. The subjectivity in the method to measure the VAP parameter also carries subjectivity in measuring the other ones that depends on it: ALH and BCF.
On the other hand, the VAP calculation depends on the particular method used for obtaining the smoothed path. In Hidayatullah et al. (2015), authors showed the implementation of a moving average filter with a fixed size of five elements. Rojas et al. (2012) used an approximation based on the Bezier Plane method. Wilson-Leedy and Ingermann (2007) used a moving average filter, size of which depends on the frame rate. The Microptic’s – SCA establishes VAP as one of the modifiable parameters by the user and thus makes it more inter subjective.
Sperm kinetic parameters determined by our software compared with the values offered by the reference software (Microptic’s – SCA) over the same samples were able to get comparable output data when measuring the same sperm particles. However, due to the better performance of our software to correctly track high-speed sperm, a higher percentage of rapid sperm and consequently average higher speed values were reported by our algorithm. It is important to consider that the measure of VSL depends only on the final and initial points of each path, so it could be directly validated by the tracking performance. Moreover, for different samples and laboratories, comparisons between available CASA systems should be carefully done since several factors inherent to motility acquisition settings affect the standardization. The other parameters (VAP, ALH and BCF) depend on which smooth filter was applied to the original path and currently there is no standardized criterion to select one as the best choice, as mentioned before.
Finally, although most of the studies conducted nowadays to boost standardization of sperm motility assessment systems are focused on the software capacities, in this study, we also analyze the equipment requirements. It is known that commercial systems have been improving their software and also associated cameras according to users’ demand. However, in the research field, labs acquire commercial CASA systems that cannot be often modernized and furthermore, publications are based on available equipment. In this sense, the choice of the velocity parameter describing the motility also depends on the video camera used. According to Wilson-Leedy and Ingermann (2007), low speed recording will hide the modifications of tracks during large time intervals (1/25 s for example) so that VCL and VAP would be quite similar. Our tool can be adapted to a range of acquisition speed (fps), suggesting that the tracking system could manage different number of frames in the same time-lapse. This is particularly useful for species with high-speed sperm, complex trajectories or unusual head/flagella movements.
In conclusion, this work presented a new open-source sperm tracker software to analyze sperm motility at a range of different cell concentrations, taking ram and buck sperm as models. The tool has the possibility to be adapted by the creators to any other sperm species.
Declaration of interest
The authors declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research reported.
Funding
This research was supported by National Agency for Scientific and Technological Promotion (ANPCyT, PICT-2015-3682) and National Scientific and Technical Research Council (CONICET, Argentina, PIP 2014-0273) grants.
References
Amann RP & Waberski D 2014 Computer-assisted sperm analysis (CASA): capabilities and potential developments. Theriogenology 81 5.e11–17.e13. (https://doi.org/10.1016/j.theriogenology.2013.09.004)
Azari M, Seyfi A & Rezaie AH 2011 Real time multiple object tracking and occlusion reasoning using adaptive kalman filters. In 7th Iranian Machine Vision and Image Processing (MVIP), pp 1–5.
Bates D, Mächler M, Bolker B & Walker S 2015 Fitting linear mixed-effects models using lme4. Statistical Software 67 1–48. (doi:10.18637/jss.v067.bib1)
Billard R & Cosson MP 1992 Some problems related to the assessment of sperm motility in fresh-water fish. Journal of Experimental Zoology 261 122–131. (https://doi.org/10.1002/jez.1402610203)
Boryshpolets S, Kowalski RK, Dietrich GJ, Dzyuba B & Ciereszko A 2013 Different computer-assisted sperm analysis (CASA) systems highly influence sperm motility parameters. Theriogenology 80 758–765. (https://doi.org/10.1016/j.theriogenology.2013.06.019)
Broekhuijse MLWJ, Šoštarić E, Feitsma H & Gadella BM 2011 Additional value of computer assisted semen analysis (CASA) compared to conventional motility assessments in pig artificial insemination. Theriogenology 76 1473–1486. (https://doi.org/10.1016/j.theriogenology.2011.05.040)
Buchelly F, Pedetta A, Pastore J, Herrera K & Ballarin V 2016 Automatic tracking of flagellar rotation of bacteria. In VII Congreso Latinoamericano de Ingeniería Biomédica (CLAIB), Bucaramanga, Colombia, p 46. (available at: http://www.abioin.com/pdf/cartilla.pdf)
Buzón Cuevas A 2014 Análisis Cinético y Morfométrico del Espermatozoide del Caballo Empleando el Sistema Sperm Class Analyzer, p 172. University of Cordoba, Spain.
Catlin DE 1989 Estimation, control, and the discrete kalman filter. Applied Mathematical Sciences (New York: Springer-Verlag) 71 285.
Chau DP, Chau D-p, Bremond F, Thonnat M. 2004. Online evaluation of tracking algorithm. In 3rd International Conference on Crime Detection and Prevention (ICDP 2009). London, United Kingdom. pp.ISBN: 978-1-84919-207-1, P38, 2009.
Cueto M, Gibbons A, Bruno M & Fernández J 2016 Manual de Obtención, Procesamiento y Conservación del Semen Ovino, p 22. Instituto Nacional de Tecnologia Agropecuaria, Gobierno de Argentina, Buenos Aires, Argentina.
Davis RO & Katz DF 1992 Standardization and comparability of CASA instruments. Journal of Andrology 13 81–86.
Dietrich GJ, Kowalski R, Wojtczak M, Dobosz S, Goryczko K & Ciereszko A 2005 Motility parameters of rainbow trout (Oncorhynchus mykiss) spermatozoa in relation to sequential collection of milt, time of post-mortem storage and anesthesia. Fish Physiology and Biochemistry 31 1–9. (https://doi.org/10.1007/s10695-005-3527-4)
Fabbrocini A, D’Adamo R, Del Prete F, Maurizio D, Specchiulli A, Oliveira LF, Silvestri F & Sansone G 2016 The sperm motility pattern in ecotoxicological tests. The CRYO-Ecotest as a case study. Ecotoxicology and Environmental Safety 123 53–59. (https://doi.org/10.1016/j.ecoenv.2015.08.018)
Fang Y, Yuan Y & Li L 2017 Performance evaluation of visual tracking algorithms on video sequences with quality degradation. IEEE Access 5 2430–2441. (https://doi.org/10.1109/ACCESS.2017.2666218)
Fauvel C, Suquet M & Cosson J 2010 Evaluation of fish sperm quality. Journal of Applied Ichthyology 26 636–643. (https://doi.org/10.1111/j.1439-0426.2010.01529.x)
Fraser L 1998 Guidelines on the application of CASA technology in the analysis of spermatozoa. ESHRE Andrology Special Interest Group. European Society for Human Reproduction and Embryology. Human Reproduction 13 142–145. (https://doi.org/10.1093/humrep/13.1.142)
Gárate Polar DA 2015 Modelo Heurístico Para la Determinación de la Motilidad en Células Espermáticas Mediante el Análisis Automático de Tracking en Video, p 40. Pontificia Universidad Católica del Perú.
Giaretta E, Munerato M, Yeste M, Galeati G, Spinaci M, Tamanini C, Mari G & Bucci D 2017 Implementing an open-access CASA software for the assessment of stallion sperm motility: relationship with other sperm quality parameters. Animal Reproduction Science 176 11–19. (https://doi.org/10.1016/j.anireprosci.2016.11.003)
Hidayatullah P, Awaludin I, Kusumo RD & Nuriyadi M 2015 Automatic sperm motility measurement. In Information Technology Systems and Innovation (ICITSI), 2015 International Conference on Information Technology Systems and Innovation (ICITSI), Bandung - Bali, Indonesia, pp 1–5.
Holt W, Watson P, Curry M & Holt C 1994 Reproducibility of computer-aided semen analysis: comparison of five different systems used in a practical workshop. Fertility and sterility 62 1277–1282. (https://doi.org/10.1016/S0015-0282(16)57201-X)
Hoogewijs M, De Vliegher S, Govaere J, De Schauwer C, de Kruif A & Van Soom A 2012 Influence of counting chamber type on CASA outcomes of equine semen analysis. Equine Veterinary Journal 44 542–549. (https://doi.org/10.1111/j.2042-3306.2011.00523.x)
Imani Y, Teyfouri N, Ahmadzadeh MR & Golabbakhsh M 2014 A new method for multiple sperm cells tracking. Journal of Medical Signals and Sensors 4 35.
Jeong J-M, Yoon T-S & Park J-B 2014 Kalman filter based multiple objects detection-tracking algorithm robust to occlusion. In SICE Annual Conference 2014, Sapporo, Japan, pp 941–946.
Jimenez-Rabadan P, Ramon M, Garcia-Alvarez O, Maroto-Morales A, del Olmo E, Perez-Guzman MD, Bisbal A, Fernandez-Santos MR, Garde JJ & Soler AJ 2012 Effect of semen collection method (artificial vagina vs electroejaculation), extender and centrifugation on post-thaw sperm quality of Blanca-Celtiberica buck ejaculates. Animal Reproduction Science 132 88–95. (https://doi.org/10.1016/j.anireprosci.2012.04.005)
Karthikeyan S, Delibaltov D & Gaur U 2012 Unified probabilistic framework for simultaneous detection and tracking of multiple objects with application to bio-image sequences. In 19th IEEE International Conference on Image Processing (ICIP), Orlando, FL, USA, pp 1349–1352.
Ledesma A, Zalazar L, Fernández-Alegre E, Hozbor F, Cesari A & Martínez-Pastor F 2017 Seminal plasma proteins modify the distribution of sperm subpopulations in cryopreserved semen of rams with lesser fertility. Animal Reproduction Science 184 44–50. (https://doi.org/10.1016/j.anireprosci.2017.06.015)
Liu J, Leung C, Lu Z & Sun Y 2013 Quantitative analysis of locomotive behavior of human sperm head and tail. IEEE Transactions on Biomedical Engineering 60 390–396. (https://doi.org/10.1109/TBME.2012.2227319)
Lucena M, Fuertes JM, De La Blanca NP & Marín-Jiménez MJ 2010 Tracking people in video sequences using multiple models. Multimedia Tools and Applications 49 371–403. (https://doi.org/10.1007/s11042-009-0376-7)
Lucena M, Fuertes JM & de la Blanca NP 2015 Optical flow-based observation models for particle filter tracking. Pattern Analysis and Applications 18 135–143. (https://doi.org/10.1007/s10044-014-0409-3)
Lucena López MJ 2003 Uso del Flujo Óptico en Algoritmos Probabilísticos de Seguimiento. Tesis Doctoral. University of Jaén, Informatic Department. (file:///C:/Users/andre/Downloads/66.pdf)
Marco-Jimenez F, Vicente JS & Viudes-de-Castro MP 2008 Seminal plasma composition from ejaculates collected by artificial vagina and electroejaculation in Guirra ram. Reproduction in Domestic Animals 43 403–408. (https://doi.org/10.1111/j.1439-0531.2007.00923.x)
Muiño Otero R 2008 Evaluación de la motilidad y viabilidad del semen bovino mediante el uso de sistemas casa y citometría de flujo: identificación de subpoblaciones espermáticas. In Patoloxía Animal, p 354. Universidade de Santiago de Compostela, Servizo de Publicacións e Intercambio Científico. Santiago de Compostela, Spain.
Palacín I, Vicente-Fiel S, Santolaria P & Yániz J 2013 Standardization of CASA sperm motility assessment in the ram. Small Ruminant Research 112 128–135.
Pei S-C, Kuo W-Y & Huang W-T 2006 Tracking moving objects in image sequences using 1-D trajectory filter. IEEE Signal Processing Letters 13 13–16. (https://doi.org/10.1109/LSP.2005.860550)
Philip RC, Ram S, Gao X & Rodr JJ 2014 A comparison of tracking algorithm performance for objects in wide area imagery. In Image Analysis and Interpretation (SSIAI), 2014 IEEE Southwest Symposium, San Diego, CA, USA, pp 109–112.
Pinheiro J, Bates D, DebRoy S & Sarkar D 2017 nlme: Linear and Nonlinear Mixed Effects Models. R package version 3.1-131.
RCoreTeam 2017 R: A Language and Environment for Statistical Computing. R Core Team. Vienna, Austria: R Foundation for Statistical Computing. (available at: http://www.R-project.org/)
Rojas HA, Rojas JA, Zuleta GA & Madrigal CA 2012 Extraction of new features for classification of porcine sperm motility. In 17th Symposium of Image, Signal Processing, and Artificial Vision, Medellin, Antioquia, Colombia, pp 323–327.
Sahbani B & Adiprawita W 2016 Kalman filter and iterative-hungarian algorithm implementation for low complexity point tracking as part of fastmultiple object tracking system. In 6th International Conference on System Engineering and Technology (ICSET), pp 109–115.
Schleh C & Leoni AL 2013 How to optimize the benefits of computer assisted sperm analysis in experimental toxicology. Journal of Occupational Medicine and Toxicology 8 6. (https://doi.org/10.1186/1745-6673-8-6)
Serra J 1982 Image Analysis and Mathematical Morphology, p 630. Academic Press, London. Orlando, Florida,USA.
Talarczyk-Desole J, Berger A, Taszarek-Hauke G, Hauke J, Pawelczyk L & Jedrzejczak P 2017 Manual vs. computer-assisted sperm analysis: can CASA replace manual assessment of human semen in clinical practice? Ginekologia Polska 88 56–60. (https://doi.org/10.5603/GP.a2017.0012)
Tinevez J-Y, Perry N, Schindelin J, Hoopes GM, Reynolds GD, Laplantine E, Bednarek SY, Shorte SL & Eliceiri KW 2017 TrackMate: an open and extensible platform for single-particle tracking. Methods 115 80–90. (https://doi.org/10.1016/j.ymeth.2016.09.016)
Urbano LF, Masson P, VerMilyea M & Kam M 2017 Automatic tracking and motility analysis of human sperm in time-lapse images. IEEE Transactions on Medical Imaging 36 792–801. (https://doi.org/10.1109/TMI.2016.2630720)
Verstegen J, Iguer-Ouada M & Onclin K 2002 Computer assisted semen analyzers in andrology research and veterinary practice. Theriogenology 57 149–179. (https://doi.org/10.1016/S0093-691X(01)00664-1)
Vinaykumar M & Jatoth RK 2014 Performance evaluation of alpha-beta and kalman filter for object tracking. In International Conference on Advanced Communication Control and Computing Technologies, Bandung, Indonesia, pp 1369–1373.
Wilson-Leedy JG & Ingermann RL 2007 Development of a novel CASA system based on open source software for characterization of zebrafish sperm motility parameters. Theriogenology 67 661–672. (https://doi.org/10.1016/j.theriogenology.2006.10.003)
Wilson-Leedy J & Ingermann R 2011 Computer assisted sperm analysis using ImageJ; description of necessary components and use of free software. (available at: https://imagej.nih.gov/ij/plugins/docs/CASAinstructions.pdf)
Zuur AF, Leno EN, Walker N, Saveliev AA & Smith GM 2009 Mixed Effects Models and Extensions in Ecology with R, p 574. New York: Springer, Verlag.
