Effect of image sampling frequency on established and smoothing-independent kinematic values of capacitating human spermatozoa

Sharon T. Mortimer1 and M. Anne Swan

Dept of Anatomy & Histology and Institute for Biomedical Research, University of Sydney, Sydney NSW 2006 Australia


    Abstract
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
It is known that image sampling frequency affects sperm kinematic values, although no study has considered the relative effect upon hyperactivated and non-hyperactivated spermatozoa. We determined the relative effect of image sampling frequency on the classification of capacitating human spermatozoa, using the established kinematic measures as well as a series of new, smoothing-independent kinematic measures. Spermatozoa were prepared by direct swim-up from semen and sperm movement was recorded using a video system which gave 201 images/s on freeze-frame playback. Trajectories were reconstructed manually and kinematics were determined using the (x, y) co-ordinates of each track point. Lower image sampling frequencies were approximated by considering every second, third, fourth, sixth and eighth track point for each trajectory. Of the 22 kinematic values investigated, only three were not significantly affected by sampling frequency in both hyperactivated and non-hyperactivated spermatozoa. Also, significant differences were observed between hyperactivated and non-hyperactivated tracks at all image sampling frequencies studied for all but four kinematic measures.

Key words: human/hyperactivation/motility/spermatozoa


    Introduction
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Human sperm movement is analysed routinely as part of sperm function tests in the work-up of an infertile couple, and the results may have an influence upon the couple's choice of treatment. One important aspect of sperm movement which may predict failure of spermatozoa to penetrate the zona pellucida is hyperactivation. While many definitions for human sperm hyperactivation have been published, several of the kinematic values used in these definitions are known to be significantly influenced by the image sampling frequency. Also, curvilinear velocity (VCL; Table IGo) is a track-averaged value, meaning that if the sperm movement pattern switched between hyperactivated and non-hyperactivated during the analysis window, the VCL value may fall below the threshold level required for a trajectory to be defined as hyperactivated (Mortimer and Swan, 1995b).Go


View this table:
[in this window]
[in a new window]
 
Table I. Definitions and abbreviations for `established' kinematic measures
 
Recently, a series of `new' kinematic measures of capacitating human sperm movement has been developed (Mortimer and Swan, 1999Go; Table IIGo). These new values were derived independently of the average path, and so were smoothing-independent. This is important, as one of the major differences between computer-aided sperm analysis (CASA) instruments is the manner in which the average path is smoothed, and hence the way in which the smoothing-dependent kinematic values (such as average path velocity (VAP), straightness (STR), wobble (WOB), amplitude of lateral head displacement (ALH) and beat/cross frequency (BCF); Table IGo) are derived. It has been postulated that the use of smoothing-independent kinematic measures will reduce the confusion which currently exists as to the identification of `hyperactivated' versus `non-hyperactivated' spermatozoa by removing a source of difference between CASA instruments (ESHRE Andrology Special Interest Group, 1998Go). Another point of difference between CASA instruments, and between countries, is the image sampling frequency used for trajectory reconstruction, which may vary between 25–60 images/s. The determination of which kinematic values remained relatively robust over these image sampling frequencies would allow further selection of the values which have the highest potential usefulness in the next generation of CASA instruments.


View this table:
[in this window]
[in a new window]
 
Table II. Definitions and abbreviations for `new' kinematic measures
 
The influence of the image sampling frequency upon kinematic values has been demonstrated previously for human spermatozoa using a first-principles approach (Mortimer et al., 1988Go). However, that study did not include a comparison of the relative effects of sampling frequency upon the kinematic values of hyperactivated and non-hyperactivated trajectories. Zhu et al. (1994) showed by comparison of two CASA instruments that the image sampling frequency (25 versus 30 Hz) influenced the proportion of tracks identified as hyperactivated. The present study was designed to determine the effect of image sampling frequency on a range of both established and new kinematic values for hyperactivated and non-hyperactivated human spermatozoa. It was important to establish the effect of frame rate on all of the kinematic values to determine whether common threshold values for the definition of a trajectory as hyperactivated could be applied across a range of sampling frequencies. The aim of this study was to compare the relative influence of image sampling frequency upon established and `new' kinematic values, using both hyperactivated and non-hyperactivated trajectories.


    Materials and methods
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
The tracks used in this study were 1 s trajectories of the same spermatozoa used in a previous study (Mortimer et al., 1997Go). Briefly, a capacitating sperm population was prepared by swim-up from semen into HTF medium supplemented with 30 mg/ml HSA. Sperm movement in 50 µm-deep MicroCell-HAC chambers (Conception Technologies, La Jolla, CA, USA) was videotaped using a NAC HSV-200 camera and video recorder system, attached to a Reichert Univar microscope (Mortimer et al., 1988Go). A x20 positive-low phase contrast objective was used, with a x2.5 camera ocular and a x1.5 intermediate. A time/date generator was wired in series, and embedded a 0.001 s time code onto the videotape during recording. No motility stimulants were used in this study.

The videotape was replayed in a Panasonic NV-F66A VCR on a Panasonic TC68A61 TV monitor, giving a final magnification of x1900. The trajectories were plotted onto sheets of overhead projector film attached to the monitor's screen. Following normal practice, only trajectories which were in the central portion of the monitor screen were reconstructed to minimize track distortion caused by screen curvature. The (x, y) co-ordinates were determined by placing the sheets over mm graph paper and noting the co-ordinates to within 0.5 mm. Some of the non-hyperactivated tracks had points too close together to be able to differentiate them to >=0.5 mm in each direction, so these were reconstructed at 100 images/s by placing a second overhead projector film sheet over the first and plotting every other point onto the top sheet, and at 66.7 images/s by plotting every third point onto another overhead projector film sheet. The same starting point was used each time.

To obtain the lower image sampling frequencies, the (x, y) co-ordinates entered in the spreadsheets were `culled' (Mortimer et al., 1988Go). For the hyperactivated trajectories, every second point of the 200 Hz trajectory was taken for 100 Hz, every third point for 66.7 Hz, every fourth point for 50 Hz, every sixth point for 33.3 Hz and every eighth point for 25 Hz. A similar procedure was used for the non-hyperactivated trajectories, but the 100 Hz trajectory was used for the 50 and 25 Hz tracks, with every second and fourth point considered, while the 33.3 Hz tracks were derived by considering every second point of the 66.7 Hz trajectories.

There were 23 hyperactivated and 24 non-hyperactivated tracks studied. All of the kinematic values were calculated for each trajectory using the Cartesian methods described previously (Mortimer and Swan, 1995aGo). The average path was estimated by 5-point smoothing for the 25 and 33.3 Hz tracks, 7-point smoothing for the 50 and 66.7 Hz tracks and 11-point smoothing for the 100 Hz tracks. The number of points used for track smoothing was increased with image sampling frequency to reduce the influence of individual track points upon the calculated average path. As the image sampling frequency increases, the number of track points increases, so if a low number of points are used for smoothing, the average path will be pulled towards the track peaks.

The established kinematic values: VCL, VSL, VAP, LIN, STR, WOB, ALHmean, ALHmax and BCF (Table IGo) were calculated for each trajectory, as well as mean angular displacement (MAD; Boyers et al., 1989Go); Dancemean (DNCmean; Robertson et al., 1988Go); fractal dimension (D; Mortimer et al., 1996Go); and a series of new kinematic values (VINmax, VINmean, AVmax, VAM, TPAmax, TPAmean, TPAmxmn, TPAmax(f), TPAmean(f) and TPAmxmn(f); Table IIGo, Mortimer and Swan, 1999).

Statistical analysis
Receiver–operator characteristic (ROC) curve analyses were performed on the data for each image sampling frequency to determine the threshold levels for hyperactivation (Schoonjans et al., 1995Go). The effect of image sampling frequency on each kinematic measure was determined by rank correlation analysis of both the hyperactivated and non-hyperactivated values. Unpaired Wilcoxon analyses were used to compare the kinematic values for hyperactivated and non-hyperactivated tracks at each image sampling frequency. All statistical analyses were performed using MedCalc for Windows (MedCalc Software, Mariakerke, Belgium).


    Results
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
The changes observed in the trajectories of hyperactivated and non-hyperactivated spermatozoa are illustrated in Figure 1Go. The classification of the trajectories was confirmed by applying the 60 Hz hyperactivation thresholds for manually-reconstructed tracks (i.e. VCL >180 µm/s and LIN <=45% and WOB <50% and ALHmean >6.0 µm or ALHmax >10.0 µm) to the 66.7 Hz trajectories (Mortimer and Swan, 1995aGo), and by observation of flagellar beat patterns.



View larger version (20K):
[in this window]
[in a new window]
 
Figure 1. Centroid trajectories of two spermatozoa reconstructed at different image sampling frequencies.

 
The effect of the image sampling frequency on the kinematic values was determined by rank correlations, as not all of the values were normally distributed (Table IIIGo). The values which were not significantly affected by frame rate for both the hyperactivated and non-hyperactivated tracks were TPAmax(f), TPAmxmn(f) and VSL. This was expected since the TPA(f) values were corrected for the frame rate used, and the VSL is only dependent upon the distance between the first and last track points and the same starting point was always used. The kinematic values which were frame rate-insensitive for only the non-hyperactivated tracks were ALHmean, MADdeg and VAP, and those which were insensitive for only the hyperactivated tracks were fractal dimension and DNCmean, although P = 0.050 for DNCmean so this result was equivocal. While VCL was significantly influenced by the image sampling frequency, the hyperactivated tracks had significantly higher VCL values than the non-hyperactivated tracks at each frequency analysed (all Z > 5.8, P < 0.0001; Figure 2Go). The most commonly-used image sampling frequencies for CASA are within the range 25–60 Hz, so the VCL values for all the 25–66.7 Hz trajectory reconstructions were analysed by ROC curve analysis. A common threshold value for hyperactivation of >142.9 µm/s was derived, with a sensitivity of 94% and a specificity of 97% (Table IVGo).


View this table:
[in this window]
[in a new window]
 
Table III. Rank correlation analysis of the effect of image sampling frequency on kinematic values for hyperactivated and non-hyperactivated tracks (see Table IGo for definitions)
 


View larger version (37K):
[in this window]
[in a new window]
 
Figure 2. Effect of frame rate on kinematics. Values shown are mean ± SD. The circles joined by solid lines are hyperactivated tracks, the squares joined by broken lines are the non-hyperactivated tracks.

 

View this table:
[in this window]
[in a new window]
 
Table IV. Threshold values for hyperactivation determined by receiver–operator curve (ROC) curve analysis, including all tracks reconstructed at 25–66.7 Hz
 
There was no significant difference by unpaired Wilcoxon analysis between the VSL values for hyperactivated and non-hyperactivated tracks for all the image sampling frequencies except 67 Hz (Z = –2.03, P < 0.01) (Figure 2Go). Accordingly, no significant threshold level for hyperactivation could be identified by ROC curve analysis for the range of commonly-used image sampling frequencies (Table IVGo).

There was a marked increase in VAP of hyperactivated tracks between 50 and 66.7 Hz, presumably due to differences in the magnitude of the fixed-point running average used for its calculation. The non-hyperactivated tracks were unaffected by image sampling frequency (Table IIIGo and Figure 2Go). Hyperactivated tracks had significantly higher VAP than non-hyperactivated tracks at 66.7 and 100 Hz only (both Z > 3.5, P < 0.0001). A significant threshold value for hyperactivation independent of image sampling frequency could not be established by ROC curve analysis (Table IVGo).

Corresponding with the increase in VAP at 66.7 Hz, the ALHmax and ALHmean of the hyperactivated tracks dropped between 50 and 66.7 Hz (Figure 2Go). This effect would be expected if the average path was being pulled towards the track peaks thereby decreasing the riser height, leading to a lower ALH value. Both the ALHmax and ALHmean values were significantly higher for the hyperactivated tracks at each image sampling frequency (all Z > 5.86, P < 0.0001) and significant hyperactivation threshold values could be established across the commonly-used image sampling frequencies for each kinematic measure (ALHmax > 8.6 µm and ALHmean > 5.5 µm, both 100% sensitivity and specificity; Table IVGo).

The velocity ratios LIN and WOB declined significantly with increasing image sampling frequency, while STR increased significantly (Figure 2Go and Table IIIGo). All three of the ratio values were significantly lower for hyperactivated tracks than for non-hyperactivated tracks at each image sampling frequency studied (all Z < –5.60, P < 0.0001).

BCF was highly frame rate-dependent, due to it being derived using the average and curvilinear paths, and also to it being a frequency measurement (Table IIIGo). While the BCF of the hyperactivated tracks was significantly higher than that of the non-hyperactivated tracks at 25 and 33.3 Hz (both Z > 3.30, P < 0.0001), it was significantly lower at both 66.7 Hz (Z = –3.11, P < 0.05) and 100 Hz (Z = –3.68, P < 0.0001; Figure 2Go). There was no significant difference observed between the BCF values of the hyperactivated and non-hyperactivated tracks at 50 Hz. Consequently, no common hyperactivation threshold value could be determined by ROC curve analysis (Table IVGo).

DNCmean increased significantly with image sampling frequency for the non-hyperactivated tracks (P < 0.0001 by rank correlation analysis) but not for the hyperactivated tracks (P = 0.050 by rank correlation analysis; Table IIIGo, Figure 2Go). The DNCmean of hyperactivated tracks was significantly higher at each image sampling frequency studied (Z = 5.87, P < 0.0001 for all).

MAD was also frame rate-dependent, with the hyperactivated and non-hyperactivated values converging with increasing image sampling frequency (Figure 2Go and Table IIIGo). The hyperactivated tracks had significantly higher MADdeg values than the non-hyperactivated tracks at image sampling frequencies of 25 to 50 Hz (all Z > 4.45, P < 0.0001), with no significant difference between the 66.7 and 100 Hz values.

The fractal dimension values for the hyperactivated trajectories were not significantly affected by the image sampling frequency, but increased with increasing image sampling frequency for the non-hyperactivated tracks (Table IIIGo and Figure 2Go). The fractal dimension of the hyperactivated tracks was significantly higher than for the non-hyperactivated tracks at each frame rate studied (all Z > 5.86, P < 0.0001), and a threshold value of fractal dimension >1.22 across the commonly-used frequencies was established by ROC curve analysis (Table IVGo).

VINmax, AVmax and VINmean increased significantly with increasing frame rate (Figure 3Go and Table IIIGo), and all were significantly higher for the hyperactivated tracks at all image sampling frequencies (all Z > 5.70, P < 0.0001). VAM increased significantly with image sampling frequency (Figure 3Go and Table IIIGo). The VAM of the hyperactivated tracks was significantly greater than that of the non-hyperactivated tracks at each image sampling frequency studied, although the difference was not as marked at 25 Hz (Z = 3.79, P < 0.05) as it was at the other frame rates (all Z > 5.74, P < 0.0001).



View larger version (32K):
[in this window]
[in a new window]
 
Figure 3. Effect of frame rate on `new' kinematics. Values shown are mean±SD. The circles joined by solid lines are hyperactivated tracks, the squares joined by broken lines are the non-hyperactivated tracks.

 
The TPA values were highly frame rate-dependent, decreasing significantly with increasing image sampling frequency (Figure 3Go and Table IIIGo), due to a relative decrease in the area bounded by the three consecutive track points (Figure 1Go). The TPAmax(f) and TPAmxmn(f) values were not significantly influenced by the image sampling frequency (Figure 3Go and Table IIIGo). All of the TPA values were significantly higher for the hyperactivated tracks than for the non-hyperactivated tracks at each frame rate studied (all Z > 5.74, P < 0.0001).

All of the kinematic values for each track obtained at each image sampling frequency (except 100 Hz) were included in ROC curve analyses to determine whether a threshold value which would be consistent across all of the commonly-used image sampling frequencies could be obtained for each kinematic measure. The only kinematic values for which a consistent hyperactivation threshold could be determined with 100% sensitivity and specificity, irrespective of frame rate (from 25 to 66.7 Hz) were ALHmax, ALHmean and DNCmean (Table IVGo). All of the other kinematic measures had threshold values with >90% sensitivity and specificity for the range of image sampling frequencies, except for VAM (89.1 and 85.9%), MADdeg (82.2 and 68.5%), VSL (66.0 and 43.5%), VAP (65.5 and 41.3%) and BCF (51.4 and 83.7%). Even though the results from a range of image sampling frequencies were included, it was interesting to note that the threshold values of the established kinematic measures determined by ROC curve analysis were similar to those obtained previously for the determination of hyperactivated trajectories at 60 Hz (Mortimer and Swan, 1995aGo).


    Discussion
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
While the effect of image sampling frequency upon the perceived kinematic values of a trajectory has already been reported (Mortimer et al., 1988Go), no specific attempt was made in that study to determine the relative effects of image sampling frequency upon hyperactivated and non-hyperactivated tracks. Also, the influence of image sampling frequency upon the `new' kinematic values had to be established, to indicate whether any of these variables were sufficiently robust to be applied independently of the image sampling frequency (Mortimer and Swan, 1999Go). This is an important consideration in the development of any new kinematic values to be measured by CASA, since ideally they should be able to be calculated in the same way irrespective of the CASA instrument or video frame rate.

The method used in this study differed from that of Mortimer et al. (1988) in that all the analyses were performed by Cartesian methods, using each track's (x, y) co-ordinates, rather than by a combination of manual and semi-automated methods. However, the concept of `plucking' or `culling' track points to give trajectories equivalent to different image sampling frequencies had been introduced previously (Mortimer et al., 1988Go).

The shape of the trajectories changed with image sampling frequency, with much more detail observable in the 50–100 Hz than in the 25 and 33.3 Hz track reconstructions (Figure 1Go). This observation provides further justification for the recommendation that trajectory analysis of capacitating human sperm populations be performed at image sampling frequencies of at least 50 Hz (ESHRE Andrology Special Interest Group, 1998Go). As would be expected, the distance between consecutive track points was inversely proportional to the image sampling frequency, with less distance between points with increasing image sampling frequency.

Image sampling frequency exerted a significant effect on the values of most kinematic values, and this effect was not always the same for hyperactivated and non-hyperactivated tracks (Table IIGo). The only kinematic measures which were not significantly affected by frame rate or motility pattern were TPAmax(f), TPAmxmn(f) and VSL. VSL was not affected by the changing image sampling frequency, as it is only the distance between the first and last track points, and the same starting point was used for each image sampling frequency. The observation of no effect of image sampling frequency is in contrast to a study comparing VSL measured by different CASA instruments (Morris et al., 1996Go). However, in that study different sampling times, as well as frequencies, were used and this meant that different track portions were analysed. Here, the same track portion was re-analysed, so there was no effect of sampling time. The relative insensitivity of the TPA(f) values to image sampling frequency demonstrated the success of multiplication of the three-point area value by the image sampling frequency to correct for the reduction in distance between consecutive track points with increasing frame rates, as had been postulated (Mortimer and Swan, 1999Go).

For other kinematic measures, i.e. ALHmean, VAP and MAD, only the non-hyperactivated tracks were not significantly affected by image sampling frequency (Table IIIGo). There was a marked increase in the VAP of hyperactivated tracks between 50 Hz and 66.7 Hz, presumably because the degree of smoothing was not sufficient for the 66.7 Hz trajectories, i.e. they were undersmoothed. Undersmoothing occurs when the number of points used for the fixed-point running average is too low, and the average path contains deviations towards the track apices. This can also result in decreased ALH values, as the riser distance, the distance between a track point and its smoothed point on the average path, is reduced. The alternative possibility was that the 50 Hz path was oversmoothed, with the smoothed average path being shorter than the `true' average path. Oversmoothing is the opposite to undersmoothing, with so many points included in the smoothing algorithm that an apex's smoothing will be influenced by the points comprising an apex on the opposite side of the track. This results in a generally straight average path, with very minor deviations to mark the presence of an apex (Davis et al., 1992Go). Correspondingly, the ALH values from an oversmoothed average path are higher than the `true' ALH, as there is a greater distance between a track point and its smoothed point on the average path.

As predicted by this observation, the ALH values of the hyperactivated tracks dipped 50–66.7 Hz, indicating a probable smoothing error (Figure 2Go). The relative insensitivity of the non-hyperactivated paths to frame rate indicated that they were probably smoothed sufficiently at each image sampling frequency. The difference between the degree of smoothing required for hyperactivated and non-hyperactivated tracks was a further indication of the differences in movement patterns between hyperactivated and non-hyperactivated spermatozoa. This observation also illustrated the difficulties encountered with the use of smoothed values, since even if the correct degree of smoothing is used for one motility type, it is not necessarily appropriate for all motility types. In any given population of capacitating spermatozoa at any given time, there would be expected to be both hyperactivated and non-hyperactivated spermatozoa, as well as some switching between motility patterns. If the kinematic values used to classify spermatozoa were influenced by the average path calculation then, depending upon the magnitude of the fixed-point running average used, the same track could be classified differently depending upon the ALH value obtained.

All of the remaining kinematic values were highly influenced by the image sampling frequency, regardless of the motility classification of the trajectory, although many still gave significantly different values for hyperactivated and non-hyperactivated tracks at all image sampling frequencies. The relationship between hyperactivated and non-hyperactivated tracks was inconsistent for both MAD and BCF (Figure 2Go). The convergence of the MAD values for hyperactivated and non-hyperactivated trajectories with increasing image sampling frequency reduced the potential value of this kinematic measure, as the trend of modern CASA instruments is towards increasing image sampling frequencies for kinematic analysis. Also, a theoretical study of MAD has predicted that as image sampling frequency increases, MAD would decrease, reaching zero for an image sampling frequency of infinity (Owen and Katz, 1993). It was presumed that the crossover effect observed for BCF was probably due to aliasing at the lower image sampling frequencies. Aliasing occurs when the frequency of the event being measured exceeds the Nyquist number, i.e. half the frequency of image sampling frequency (Owen and Katz, 1993; Davis and Siemers, 1995). The effect of calculation method and image sampling frequency upon the BCF of a trajectory will be explored further in another study.

In conclusion, while frame rate affected both the `established' and `new' kinematic measures, discrimination between hyperactivated and non-hyperactivated trajectories was possible at the image sampling frequencies commonly used by CASA instruments. Further, independent evaluation of the applicability of the smoothing-independent kinematic measures by application of the values in different CASA instruments is now required.


    Notes
 
1 To whom correspondence should be addressed at: Dept of Animal Science, University of Sydney, Sydney NSW 2006, Australia Back


    References
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Boyers, S.P., Davis, R.O. and Katz, D.F. (1989) Automated semen analysis. Curr. Probl. Obstet. Gynecol. Fertil., XII, 167–200.

Davis, R.O. and Siemers, R.J. (1995) Derivation and reliability of kinematic measures of sperm motion. Reprod. Fertil. Dev., 7, 857–869.[ISI][Medline]

Davis, R.O., Niswander, P.W. and Katz, D.F. (1992) New measures of sperm motion. I. Adaptive smoothing and harmonic analysis. J. Androl., 13, 139–152.

ESHRE Andrology Special Interest Group (1998) Guidelines on the application of CASA technology in the analysis of human spermatozoa. Hum. Reprod., 13, 142–145.[Free Full Text]

Morris, A.R., Coutts, J.R.T. and Robertson, L. (1996) A detailed study of the effect of videoframe rates of 25, 30 and 60 Hertz on human sperm movement characteristics. Hum. Reprod., 11, 304–311.[Abstract]

Mortimer, S.T. and Mortimer, D. (1990) Kinematics of human spermatozoa incubated under capacitating conditions. J. Androl., 11, 195–203.[Abstract/Free Full Text]

Mortimer, S.T. and Swan, M.A. (1995a) Kinematics of capacitating human spermatozoa analysed at 60 Hz. Hum. Reprod., 10, 873–879.[Abstract]

Mortimer, S.T. and Swan, M.A. (1995b) Variable kinematics of capacitating human spermatozoa. Hum. Reprod., 10, 3178–3182.[Abstract]

Mortimer, S.T. and Swan, M.A. (1999) The development of smoothing-independent kinematic measures of capacitating human sperm movement. Hum. Reprod., 14, in press.

Mortimer, D., Serres, C., Mortimer, S.T. and Jouannet, P. (1988) Influence of image sampling frequency on the perceived movement characteristics of progressively motile human spermatozoa. Gamete Res., 20, 313–327.[ISI][Medline]

Mortimer, S.T., Swan, M.A. and Mortimer, D. (1996) Fractal analysis of capacitating human spermatozoa. Hum. Reprod., 11, 1049–1054.[Abstract]

Mortimer, S.T., Schoëvaërt, D., Swan, M.A. and Mortimer, D. (1997) Quantitative observations of flagellar motility of capacitating human spermatozoa. Hum. Reprod., 12, 1006–1012.[ISI][Medline]

Owen, D.H. and Katz, D.F. (1993) Sampling factors influencing accuracy of sperm kinematic analysis. J. Androl., 14, 210–221.[Abstract]

Robertson, L., Wolf, D.P. and Tash, J.S. (1988) Temporal changes in motility parameters related to acrosomal status: identification and characterization of populations of hyperactivated human sperm. Biol. Reprod., 39, 797–805.[Abstract]

Schoonjans, F., Zalata, A., Depuydt, C.E. and Comhaire, F.H. (1995) MedCalc: a new computer program for medical statistics. Computer Methods Progr. Biomed., 48, 257–262.

Zhu, J.J., Pacey, A.A., Barratt, C.L.R. and Cooke, I.D. (1994) Computer-assisted measurement of hyperactivation in human spermatozoa: differences between European and American versions of the Hamilton-Thorn motility analyser. Hum. Reprod., 9, 456–462.[Abstract]

Submitted on April 24, 1998; accepted on February 1, 1999.