Evaluation of the CeliSoft* automated semen analysis system in a routine laboratory settingt

Transcription

1 r FERTILITY AND STERILITY Copyright e 1988 The American Fertility Society Printed in U.S.A. Evaluation of the CeliSoft* automated semen analysis system in a routine laboratory settingt David Mortimer, Ph.D.:\: Nand Goel, B.Sc. Margaret A. Shu, B.Sc.11 The University of Calgary Health Sciences Centre, Calgary, Alberta, Canada The sperm concentration and percentage motility values generated by version 3.2 of the CellSoft (Cryo Resources Ltd., New York, NY) automated semen analyzer on 200 ejaculates were compared with those obtained by standardized traditional methods. Overall, CellSoft gave mean concentrations that were 20.9 X 10 6 /ml lower (95% range of differences = to X 106/mI). However, the difference between methods was not systematic. Below 50 X 10 6 /ml, CellSoft more often gave higher values, and above 100 X 106 /ml, it usually gave lower values. In the middle range, differences were randomly distributed. For motility, the CellS oft values were usually higher than those obtained by visual counting (mean difference = -17.5%, 95% range = -56.0% to +21.0%). Multiple regression analyses revealed a strong concentration dependency such that reliable values will probably be obtained only if all samples are diluted (with homologous seminal plasma) before CellS oft analysis. This upper concentration limit is of the order of 30 to 50 X 10 6 / ml. Without such dilution, this version of CellSoft will not provide sufficiently accurate values for basic semen characteristics and cannot be accepted as a routine diagnostic method. Fertil Steril 50:960, 1988 Careful and thorough semen analyses remain an essential part ofthe initial investigations of all couples presenting with infertility. However, semen analysis characteristics show marked variation due to a number of different factors: (1) interindividual differences or population variance; (2) intraindividual differences such as collection artifacts, abstinence effects, and biologic variation; and (3) methodologic variations or errors.1 While the first two types of factors are biologic in their origin, the last is due to technical inaccuracy in the labora- Received April 16, 1988; revised and accepted August 9, * Cryo Resources Ltd., New York, New York. t Supported by The Alberta Heritage Foundation for Medical Research and The Nat Christie Foundation. :j: Reprint requests: David Mortimer, Ph.D., Endocrine/Infertility Clinic, Department of Obstetrics and Gynaecology, The University of Calgary Health Sciences Centre, 3330 Hospital Drive NW, Calgary, Alberta T2N 4N1, Canada. University of Calgary Medical School. II Diagnostic Semen Laboratory. 960 Mortimer et al. Automated semen analysis tory. The need for standardization of the laboratory procedures for semen analysis has been recognized for many years. 2-4 The publication by the World Health Organization (WHO) of its Laboratory Manual for the Examination of Human Semen and Semen-Cervical Mucus Interaction 5 was an acknowledgement of this need, and it has been used by clinicians and research scientists worldwide. Its success has resulted in the publication of a second, revised and expanded edition. 6 Unfortunately, the widespread use of these standards still remains to be achieved, as revealed, for example, by a survey of laboratories in Australia. 7 A recent development in this area of laboratory medicine has been the appearance of instruments designed to automate various aspects of semen analysis. One such instrument, the CellSoft automated semen analyzer (Cryo Resources Ltd., New York, NY) has received much attention and is now used extensively around the world. Numerous abstracts have been presented at meetings during the

2 past 2 years on the use of CellS oft, but few refereed papers have as yet been published on its clinical application.b- ll While the CellSoft system's precision, i.e., consistency between duplicate determinations on the same sample or between repeat analyses of the same sequence of videotape, appears to be very good,b,9 until very recently there had not been any rigorous evaluation of the system for the accuracy of its results. 1O - 13 Quite obviously, before any such system can be accepted in routine use, it must be validated by systematic comparison with other accepted reliable quantitative techniques, such as those recommended by the WHO. 5,6 The purpose of this study was to undertake just such an evaluation by comparing the values generated by CellSoft for the concentration of spermatozoa and the percentage of motile spermatozoa in semen samples with those obtained by standardized, quality-controlled, traditional manual methods based upon the recommendations of the WH014 and which have previously been shown to be of high precision, accuracy, and reliability.15 Because CellSoft is being marketed as a computer-assisted semen analyzer, and it is in this role that most of its users employ the system, we considered this level of evaluation to be of highest priority. Clearly, one of the great advantages of systems such as CellS oft is the automated determination of sperm movement characteristics, validation of which is the subject of other studies to be reported at a later date. MATERIALS AND METHODS Two hundred semen analyses carried out at the University of Calgary Diagnostic Semen Laboratory from a mixed population of infertility patients and semen donors were performed according to our routine procedures.14 Semen analyses were usually commenced at 30 minutes after ejaculation and initial evaluations of sperm motility and vitality completed by 60 minutes. Standard semen characteristics of interest to the present study were sperm concentration; the percentages of motile, vital, and morphologically normal spermatozoa; the proportion of spermatozoa involved in clumping; visual ratings of the incidences of "round cells" and debris; and the concentration of peroxidase-positive leukocytes. Clumping was assessed on a subjectively rated scale with 5% increments based upon observation of at least 10 randomly selected fields of the wet preparation.14 All groups of cells adherent to each other were considered as clumps, and the overall proportion of the spermatozoa that were involved in such clumps was estimated to the nearest 5%. In addition, videotape recordings (National Television Systems Committee [NTSC] standard, Beta-I format) were made within a few minutes of the visual counts, and analyzed subsequently using version 3.2 of the CellSoft system. The standard CellSoft parameter settings used throughout this study were: 30 frames analyzed at an image-sampling frequency of 30 Hz; 4 frames minimum sampling for motility; 15 frames minimum sampling for both velocity and ALH (amplitude of lateral head displacement) measurements; 10 11m/sec threshold velocity for both velocity and ALH measurements; 200 J.Lm/sec maximum velocity; minimum linearity of 2.5 for ALH measurement; and a cell size range of 5 to 25 pixels with a magnification calibration of J.Lm/pixel. These settings were established at the outset of the study in mid-1986 after discussions with Cryo Resources staff and other CellSoft users at that time. For the sake of data comparability, system parameter settings were not altered during the study (see Discussion). Determination of Sperm After ensuring thorough mixing, a 50-J.LI aliquot was taken with a positive displacement pipette and diluted in 950 J.Llof a bicarbonate-formaldehyde diluent (50 g NaHC03, 10 m136% to 40% formaldehyde and 0.25 g trypan blue C made up to 1000 ml with reverse osmosis water, filtered and stored at +4 C). Dilutions were stored in tightlycapped autoanalyser cups at +4 C until counted (usually within 24 hours) using double-chamber Improved Neubauer hemocytometers. Each diluted semen aliquot was vortex-mixed for at least 10 seconds before 10 J.LI were transferred to each of the hemocytometer chambers so that each dilution was counted in duplicate. Hemocytometers were left in the humid chamber (a desiccator with gauze pads in the bottom soaked with water) for 10 to 15 minutes to allow all the spermatozoa time to settle down on to the counting grid. Counts were performed using a 20X objective and phase contrast optics according to the number of spermatozoa found in the first (top left-hand corner) large square of the grid. If <10 spermatozoa were found then the whole grid (25 large squares) was counted on each side of the hemocytometer. If 10 to 40 spermatozoa were found then two horizontal rows were counted in each side (2 X 10 large squares counted Mortimer et al. Automated semen analysis 961. i:...

3 Table 1 Semen Analysis Characteristics of the 200 Samples Used in the Present Study as Determined Using Either Traditional Manual Methods or by CellSoft 95% range Characteristic Units Mean SD (min-max) Semen analysis Sperm concentration 106/ml N.A." Cube root concentration to Motile % to 73.6 Progressively motile % to 69.2 Vital ("live") % to Morphologically normal % to 69.2 CellSoft Sperm concentration 106/ml N.A." Cube root concentration to Motile % to b "NA, not applicable. b Distribution of this characteristic was not improved by any of the transformations tested. in total). With >40 spermatozoa per large square, only the four corner squares plus the center square were counted in each side (2 X 5 large squares counted). Only recognizable spermatozoa, including loose heads, were counted; other germinal line cells and free tails were ignored. If a spermatozoon lay on the dividing line between two adjacent squares, then it was counted only if it was on the upper or left-hand sides of the square being counted. The total number of spermatozoa counted per hemocytometer was divided by an appropriate correction factor (10, 4, or 2 for 2 X 25, 2 X 10, or 2 X 5 large squares counted) to give the sperm concentration in the original semen samples in millions/ml. Determination of Sperm Motility For each sample a 10 }LI aliquot of liquefied thoroughly mixed semen was mounted between a warmed slide and 22 X 22 mm coverslip. These preparations were then scored at 400X magnification using phase contrast optics by counting the numbers of motile spermatozoa in several randomly selected fields away from the coverslip edge. Sufficient fields were scored so that at least 200 spermatozoa were counted. In each field the number of progressively motile spermatozoa was counted first, counting only those spermatozoa that were present in the field at a given moment. If the number of spermatozoa in an entire field was too great for rapid visual counting then a small area of the field was delineated using an eyepiece graticule. Progressive spermatozoa were defined as those motile spermatozoa which showed definite space gain (at least two sperm head lengths per second, or an approximate straight line velocity of 10 }Lm/sec). After the progressive spermatozoa have been counted then the numbers of nonprogressively motile and immotile spermatozoa present in the same area were counted. Nonprogressive spermatozoa were defined as those motile spermatozoa that did not show definite space gain (straight line velocity < 10 }Lm/sec), and included those spermatozoa showing only feeble flagellar beating. Immotile spermatozoa were those which showed no flagellar movement at all. These three figures were expressed as percentages adding up to 100 with rounding off, if necessary, being applied to the largest category or categories. Statistical Analysis All analysis of the data was performed using the Statistical Package for the Social Sciences. I6 Plots showing the agreement between the two methods of measurement were constructed according to the method described by Bland and Altman. I7 Data for sperm concentration values were transformed to optimize the normality of their distributions.is Several transformations were applied: logarithmic (Loge), square root, and cube root. Cube root transformation gave the smallest skewness value and was therefore used in further analyses. Transformation of the differences between manual and CellSoft values was not possible due to the negative sign attributed to the cases where CellSoft's value was greater than that obtained manually. RESULTS The characteristics of the 200 semen samples (Table 1) confirmed that a wide range of semen 962 Mortimer et al. Automated semen analysis

4 y'... ~ , oj. ~...-I.f.. ' 100 :.,: ::.".:. t e 1.- e.. ~ Manual 400 (eusoft = Manual) Figure 1 Scatterplot of CeliSoft sperm concentration values against manual semen analysis sperm concentration values for the 200 samples used in the present study. The broken line represents the line of correspondence (i.e., slope = 1.0) while the dotted line is that fitted using linear regression. quality was covered by the study. The mean semen characteristics provided by CellSoft for these samples are also included in Table 1. Even these summary statistics revealed clear discrepancies between the traditional manual and CellSoft methods of measurement. Sperm Figure 1 is a scatterplot ofthe paired determinations of sperm concentration in the 200 samples. Clearly more samples lie below the line of correspondence (i.e., slope = 1) than above it. Linear regression (r = 0.672, P < 0.001) demonstrated a significant deviation from a direct correspondence between the two measurement methods. Because of the limitations of the scatterplot/linear regression method of analysis for this type of data, the method proposed by Bland and Altman 17 was applied (Fig. 2). This analysis revealed very large differences between the paired manual and CellSoft determinations of sperm concentration. Overall, CellSoft gave a sperm concentration value lower than manual analysis. The mean difference was (standard deviation [SD] = 68.13) giving a 95% range of to X 106/ml. Only 38/200 samples (19.0%) had CellSoft values within 10% of the manual sperm concentration counts. However, because there were often large discrepancies between CellSoft and manual analysis (with the latter method being known to have high accuracy and precision 15 ), the data were further examined by plotting the differences between the two methods against the manual analysis data (Fig. 3). The relationship between the magnitude of the difference and the manually derived sperm concentrations demonstrated that at low sperm concen if 300 '" ~ :;;> ~ ~ ; SD :.. -.~=*~~~~::.~~::-;'--:.:=--~~------~ mean SD, 400, o 200 (Manual + [eilsoft) / 1 Figure 2 Plot of the differences between the CeliSoft and manually derived values for sperm concentration for the 200 samples used in the present study against the average sperm concentration obtained using the two methods of measurement. The dashed line represents the mean difference between the two methods and the broken lines the 95% confidence limits (±2 SD) of the differences. trations (1 to 50 X 106/ml) CellSoft more often gave higher values (i.e., a negative difference, 29/43 samples), was randomly biased over the range 51 to 100 X 106/ml (34/65 samples with negative differences), and biased toward low values when the sperm concentration was above 100 X 106/ml (7/42 samples with negative differences for the 101 to 150 X 106/ml range and 6/50 samples with negative differences for the >150 X 106/ml range [with 30/ 44 of the samples in the latter range showing differences of> +50 X 106/ml]). This distribution of differences between manual and CellSoft analyses was significantly skewed (4 X 4 x 2 test with Yates' correction = , P < 0.001). Multiple regression analysis with use of the Cell Soft sperm concentration values as the dependent _ _... --_ _- --'-_ SO...., o, 0 100, 0... hi., 1. o :~~~:lf.~:';~~------~---. mean,.,.,..1,., i i I i o Manual -2 SO Figure 3 Plot of the differences between the CellS oft and manually derived values for sperm concentration for the 200 samples used in the present study against the sperm concentration obtained by manual semen analysis_ The dashed line represents the mean difference between the two methods and the broken lines the 95% confidence limits (±2 SD) of the differences. Mortimer et al. Automated semen analysis 963

5 _... ## Table 2 Multiple Regression Analyses Using CellSoft Sperm and Percent Motile Characteristics as Dependent Variables and Semen Analysis Characteristics as the Predictor Variables CellSoft dependent variable % motile Semen analysis predictor variable Clumping % motile Constant % motile % normal forms Clumping Constant Multiple B R D.R 2 (nonstandardized) variable (Table 2) revealed that the semen analysis sperm concentration was the most significant predictor and accounted for 45.2% of all variance. Clumping and percent motile spermatozoa accounted for a further 1.6% and 1.4 % of all variance, respectively. Because only a total of 48.2% of all the variance could be accounted for, and since the data for sperm concentration were known to be non-normally distributed, another multiple regression was performed using the cube root-transformed (CUBRT) concentration data. In this case, only the transformed manual sperm concentration determinations were a significant predictor variable (multipler = , F1198 = , P < 0.001) with the relationship: CUBRT (CellSoft) = [CUBRT (manual) X ] This relationship accounted for 54.1 % of all variance. No other semen analysis variable had a sufficiently significant relationship with the transformed CellSoft data to be included in the analysis. Sperm Motility In contrast to the sperm concentration data, the values for the percent motile spermatozoa generated by CellSoft were usually higher than those obtained by manual sperm counting methods (Fig. 4). Again, although a highly significant linear relationship was obtained (r = , P < 0.001), this type of analysis was relatively uninformative as to the reliability of CellSoft's motility values. Plotting the difference between the manual and CellSoft methods against their average17 revealed a strong bias due to the high motility values generated by CellSoft (Fig. 5), and we again restructured the analysis with the manually derived values on the abscissa (Fig. 6). In this case, the simple arithmetic values for the differences were used, the transfor g; ~ ~......:: o 0 20 CellSoft = (1-397x Manual) "~"~..: i." J2":. ~"! ~~.. _ e,,; ::~<~.'~'.... :... ~ Manual Figure 4 Scatterplot of CellSoft percent sperm motility values against manual semen analysis percent sperm motility values for the 200 samples used in the present study. The broken line represents the line of correspondence (i.e., siope = 1.0) while the dotted line is that fitted using linear regression. 20.:.. ~ ~ 0... ~... ~... :..._:t...::_.:!.;..._.r_ SD e. e:.... ' e.-.._:._ ~. '.. ::: :-;'-.:-.:~:~--:----- mean ~......:~.:.: :.::.::~::::~~.' SD '.,,,, BO (Manual + ee/lsoft) /2 Figure 5 Plot of the differences between the CellSoft and manually derived values for percent sperm motility for the 200 samples used in the present study against the average percent sperm motility obtained using the two methods of measurement. The dashed line represents the mean difference between the two methods and the broken lines the 95% confidence limits (±2 SD) of the differences. 964 Mortimer et al. Automated semen analysis

6 if 20 ::;, c:: e. ~ -20 t;; ~ -40,I. to. _.. ~ :... _ so ~~:~~~~~~~-~-~- I r...:. ~ r'f'i) -: :. ":"," '20 ' 40' 60' 80 Manual + 2 SO mean - 2 SO Figure 6 Plot of the differences between the CellSoft and manually derived values for percent sperm motility for the 200 samples used in the present study against the percent sperm motility obtained by manual semen analysis. The dashed line represents the mean difference between the two methods and the broken lines the 95% confidence limits (±2 SD) of the differences. mations attempted did not improve the normalization of the data. The average difference was -17.5%, indicating that CellSoft usually gave a higher value for percent motility, the 95% range of -56.0% to 21.0%. Only 29 of the 200 samples (14.5%) had CellSoft values within 10% of the manual sperm motility counts. Multiple regression analysis using the CellSoft percent motility as the dependent variable produced the findings presented in Table 2. While semen analysis percent motility was the most significant predictor variable, accounting for 33.2% of all variance, semen analysis sperm concentration accounted for a further 16.6%. Normal morphologic features and clumping accounted for an additional 1.2% and 1.0% each. Overall, the multiple regression analysis (F 4 i95 = , P < 0.001) accounted for 52.0% of the total variance. Influence of Debris, Round Cells, and Leukocytes Linear regression analyses of the relationships between the amount of debris (scored as 0 to 4), the number of "round cells" (scored as 0 to 2), and the concentration of peroxidase-positive leukocytes (in millions/ml) were performed with use of the percent motile spermatozoa as dependent variable. All gave extremely small r values (0.086,0.043, and , respectively). Predictability of the Discrepancies between CellSoft and Semen Analysis Multiple regression analyses were also performed using the differences between CellSoft and manual measurements of both sperm concentra- tion and percent motility (Table 3). In both these analyses the semen analysis sperm concentration was the most significant predictor variable, with other semen analysis characteristics accounting for small amounts of additional variance. Overall, 63.7% of all the variance in concentration differences was accounted for (F 3 i96 = , P < 0.001) but only 29.9% of that for differences in % motility (F 3 i96 = , P < 0.001). DISCUSSION In comparison with routine laboratory methods for semen analysis CellSoft has unacceptably high variability in its determinations of sperm concentration and the percent motile spermatozoa. Similar findings have also been reported recently from other laboratories.1o,n On the other hand, the routine semen analysis methods used in the present study have been demonstrated previously to have excellent precision.15 Indeed, our hemocytometry procedure is that now recommended by the WHO.6 The discrepancies between sperm concentration values obtained using this method and those generated by CellSoft may be due to a number of possible sources of error. Firstly, CellSoft uses a Makler chamber which, according to the original description of the device, has coefficients of variation ranging from 21.6% for the 10 to 20 X 106/m l range to 6.1 % for the 100 to 200 X 106/ m l range.19 Our average difference between duplicate hemocytometric determinations is <1.0 X 106/m l over the range 20 to 150 X 106/ml (95% range of differences = -7.2 to +6.9 X 106/ ml).15,20 We do not use the Makler chamber in our routine laboratory because of its lower precision and the simple fact that even very experienced technicians find it difficult to use for samples with high sperm concentrations. Although a wide range of sperm concentrations was apparently studied by Mathur et al. B (judging from the size of the standard errors presented) there were only 67 samples in total. These authors quoted linear regression coefficients (r values) of 0.80 and 0.93 for the 35 "fertile" and 32 "infertile" semen samples in comparing their routine semen analysis with CellSoft. However, these statistics only showed that the observations fell close to a straight line, there was no information whatsoever regarding the actual relationship between the two methods of measurement. Values for the slope and intercept are essential since if the slope is significantly +1.0 then the two methods are not giving Mortimer et a1. Automated semen analysis 965

7 r Table 3 Multiple Regression Analyses Using the Difference Between CellSoft and Manual Semen Analysis Sperm and Percent Motile Characteristics as Dependent Variables and Semen Analysis Characteristics as the Predictor Variables Dependent variable ~ concentration ~ % motile Semen analysis predictor variable Clumping % motile Constant % normal forms % vital ("live") Constant Multiple B R b.r 2 (nonstandardized) the same answers (similarly the intercept should be zero). Nor was there any information regarding the size and variability of the differences between the two methods of measurement. More appropriate and sensitive statistical analyses are essential (c.f. ref. 17) before the statement that there was "excellent agreement between the two measurements" can be accepted; it cannot be justified from the data analysis provided in the original paper.s The scatter of concentration values obtained by Vantman et al. also demonstrates the inappropriateness of linear regression for this type of study. 11 Secondly, there are two potential sources of error originating in the image analysis technology used by CellSoft: (1) the correct identification of all the spermatozoa in a field and (2) the identification of motile as opposed to immotile cells. With regard to the first category, clumped spermatozoa, which are commonly seen in clinical material, would be digitized as a single image that, being too large for a sperm head according to the pixel size range set in the system parameters, would be rejected from the analysis. Even clumps of as few as two to three cells would digitize to give a single object whose area exceeded the maximum allowed in the system parameters for a spermatozoon (i.e., 25 pixels). This would simultaneously decrease the sperm concentration and, because almost exclusively immotile spermatozoa are involved in clumping, increase the motile fraction. Any debris or other cells similar in size to a sperm head that appear refractile in the phase contrast image used for digitization would be included as immotile spermatozoa in the analysis. With regards to the motile versus immotile question, CellSoft defines a motile object as one showing a curvilinear velocity that exceeds the preset threshold over the first four frames. Because the threshold of 10 /-lm/sec corresponds to a real movement between frames of 0.33 /-lm (10 /-lm/30 Hz), this requires that the object centroid move only 0.5 pixels between frames (since 1 pixel = /-lm). Spurious motile objects could be created easily by a small amount of specimen "flow" or by collisions at the start of the digitization grab period between a motile spermatozoon and either an immotile spermatozoon or other cell or even a piece of refractile debris. Obviously, this would further increase the motile fraction of the sample. The CellSoft system parameter settings used in the various studies remain as another potential source of differences between evaluation studies. Our settings were established at the outset of the study as a result of discussions with Cryo Resources staff and other CellSoft users at that time. Obviously, we could not modify the system parameter settings during this evaluation study. However, subsequent to this study we have evaluated the use of a higher threshold velocity for motility in an attempt to define a subpopulation of more progressively motile cells. 21 A very strong concentration dependency of the values generated by CellSoft was evident. This problem has already been reported by other workers. 1O,11,22,23 CellSoft's overestimation of sperm concentration at the lower end of the range is probably due to the reduced likelihood of clumping in these samples and possibly to the incidence of refractile debris, etc. being more significant in comparison to the numbers of spermatozoa per field. At high sperm concentrations, there would probably be both more clumping and more collisions, causing proportionately greater reductions in the apparent sperm concentration and increases in the percentage motile. Of great significance in this respect is that the most motile cells will be preferentially excluded by collisions, and consequently a valid detailed evaluation of sperm movement characteristics can be performed only on appropriately diluted samples. Such an evaluation is in progress. In general, the simple regressions demonstrated 966 Mortimer et ai. Automated semen analysis

8 there to be little apparent effect of round cells, leukocytes, or, overall, debris upon CellSoft's assessment of a sample. The multiple regression analyses confirmed that clumping was a small but significant factor. In terms of a useful range for CellSoft, it seems that samples with sperm concentrations greater than 100 X 106/ml cannot be analyzed under standard conditions. Furthermore, since the likelihood of collisions will be directly related to sperm concentration' it would be best to set an even lower limit of about 50 X 10 6 /ml. Vantman et al.ll have demonstrated empirically that this limit should be 40 X 106/ml, a requirement with which we concur. Problems related to the incidence of refractile debris becoming significant at low sperm concentrations may be correctable if a narrower cell size range (e.g., 10 to 25 pixels) were to be used. Obviously, the only way CellSoft could then be used in a routine laboratory setting would be to dilute the semen sample to obtain a sperm concentration of the order of 30 to 50 X 10 6 / ml. The diluent used for hemocytometry clearly cannot be used for this purpose (as suggested by Mathur 22 ) since it contains formaldehyde and would kill all the spermatozoa. Instead, either homologous seminal plasma (prepared from an aliquot of the sample being analyzed with centrifugation) or perhaps pooled seminal plasma obtained from many normal semen samples must be used. Because dilution with balanced salt solutions or culture media would cause significant changes in sperm movement this would not be acceptable if movement analysis was also. a required feature of the CellSoft evaluation. This is also the recommendation of other workers.ll This and other studies have clearly demonstrated that CellSoft version 3.2 has severe limitations as an automated semen analyzer. lo,ll Indeed, as a result ofthe present study, we are left with the inescapable conclusion that CellSoft version 3.2 cannot be accepted as a reliable method for automated semen analysis. While at least some of the problems may be circumvented by dilution of semen samples using homologous cell-free seminal plasma, they have been acknowledged by Cryo Resources, and a revised version of CellSoft is being developed. We have recently received an early version of this new CellSoft program that is now undergoing extensive evaluation. The problems of detecting all the immotile spermatozoa and discriminating them from debris using presently available image analysis technology may well prevent any automated system from providing values for total sperm concentration and the proportions of immotile, nonprogressive and progressively motile cells with the accuracy and reliability of traditional manual methods. Because of the optical and digital image analysis principles fundamental to all systems that use machine vision, all automated analyzers will be influenced to a greater or lesser extent by one or more of the problems discussed above. The authors only have limited experience with other automated analyzers and consequently precise relationships cannot be defined at the present. However, since it is only the concentration of progressively motile spermatozoa in semen, along with their movement characteristics, that is of clinical significance, perhaps we should take this opportunity to consider just what andrologists require of these automated analyzers. Now is the time for users to influence the developers and manufacturers of automated analyzers to provide us with the most useful tools possible within the constraints of currently available technology. Acknowledgments. The authors thank Miss Ruth Tan and Mrs. Kathy Moore of the Diagnostic Semen Laboratory for their assistance in performing the semen analyses and preparing the videotapes. Weare grateful to Mr. David Persaud of the Nat Christie Unit for the Study of Human Reproduction for assistance with the statistical analyses. REFERENCES 1. Mortimer D: The value of conventional semen analysis. In Advances in Perinatal and Reproductive Medicine, Edited by WC Cheng, SL Tan. Singapore, PG Publishing Pte Ltd. In press, Eliasson R: Standards in evaluation of human semen. Andrologie 3:49, Eliasson R: Analysis of semen. In Progress in Infertility, 2nd ed, Edited by SJ Behrman, RW Kistner. Boston, Little, Brown & Co, 1975, p Freund M, Peterson RN: Semen evaluation and fertility. In Human Semen and Fertility Regulation in Men, Edited by ESE Hafez. St. Louis, CV Mosby, 1976, p Belsey MA, Eliasson R, Gallegos AJ, Moghissi KS, Paulsen CA, Prasad MRN: Laboratory Manual for the Examination of Human Semen and Semen-Cervical Mucus Interaction. Singapore, Press Concern, World Health Organization: WHO Laboratory Manual for the Examination of Human Semen and Semen-Cervical Mucus Interaction. Cambridge, Cambridge University Press, Tyler JPP, Harrison KL, Crockett NG: Semen analysis: an Australian survey. Aust J Med Lab Sci 6:59, Mathur S, Carlton M, Ziegler J, Rust PF, Williamson HO: Mortimer et al. Automated semen analysis 967

9 A computerized sperm motion analysis. Fertil Steril46:484, Knuth UA, Yeung C-H, Nieschlag E: Computerized semen analysis: objective measurement of semen characteristics is biased by subjective parameter setting. Fertil Steril48:118, Knuth UA, Nieschlag E: Comparison of computerized semen analysis with the conventional procedure in 322 patients. Fertil Steril49:881, Vantman D, Koukoulis G, Dennison L, Zinaman M, Sherins RJ: Computer-assisted semen analysis: evaluation of a method and assessment of the influence of sperm concentration on linear velocity determination. Fertil Steril 49: 510, Lorton SP: Sperm motion analysis. (Letter) Fertil Steril4 7: 885, Mortimer D: Computerized semen analysis. (Letter) Fertil Steril49:182, Mortimer D: The Male Factor in Infertility Part I: Semen Analysis. Current Problems in Obstetrics, Gynecology and Fertility, Vol VIII, No 7. Chicago, Year Book Medical Publishers, 1985, 87 pp 15. Mortimer D, Shu MA, Tan R: Standardization and quality control of sperm concentration and sperm motility counts in semen analysis. Hum Reprod 1:299, SPSS Inc: SPSS' User's Guide. New York, McGraw-Hill Book Co, 1983, 806 pp 17. Bland JM, Altman DG: Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1:307, Mortimer D, Lenton EA: Distribution of sperm counts in suspected infertile men. J Reprod Fertil 68:91, Makler A: A new chamber for rapid sperm count and motility estimation. Fertil Steril30:313, Mortimer D, Shu MA, Tan R, Mortimer ST: A technical note on diluting semen for the haemocytometric determination of sperm concentration. Hum Reprod. In press 21. Mortimer D, Mortimer ST: Influence of system parameter settings on human sperm motility analysis using CellSoft. Hum Reprod, 3:621, Nieschlag E: Computerized semen analysis. (Letter [Reply]) Fertil Steril49:183, Mathur S: Computerized semen analysis (Letter [Reply]) Fertil Steril49:184, Mortimer et al. Automated semen analysis