- Email: [email protected]
Comparison of three computer methods of sperm head analysis
FERTILITY AND STERILITY威 VOL. 80, NO. 3, SEPTEMBER 2003 Copyright ©2003 American Society for Reproductive Medicine Published by Elsevier Inc. Printed on acid-free paper in U.S.A.
Comparison of three computer methods of sperm head analysis Ariadne Rodrigues Goulart, M.Sc., Moema de Alencar Hausen, B.Sc., and Luiz Henrique Monteiro-Leal, Ph.D. Universidade do Estado do Rio de Janeiro, Departamento de Histologia e Embriologia, Laborato´rio de Microscopia e Processamento de Imagens, Rio de Janeiro, Brazil
Objective: Analysis of sperm heads using three different computer morphometrical tools and experimental conditions to find a more reliable and secure strategy among them. Design: Controlled experiments on sperm morphology analysis from volunteers. Setting: Laboratory of microscopy and imaging processing. Patient(s): Ten human semen samples donated by different zoospermic men. Intervention(s): Semen samples were collected by masturbation after ⱖ72 hours of abstinence. Main Outcome Measure(s): Spermatozoon head morphology was compared by the use of different videomicroscopy systems, three computer programs, and various staining conditions and manipulation by different operators. Nonbiological material in the form of latex beads was also used. Result(s): The data obtained suggest that the semiautomatic computer program is the most reliable and secure method for performing sperm analysis, besides the fact that it is a fast process compared with manual methods. Conclusion(s): Computer systems of sperm analysis should incorporate a step of interactive object identification to work properly, allowing the operator to confirm or correct possible computer misidentification. The latex beads were used to confirm the capability of all three computer programs to correctly evaluate nonbiological material. (Fertil Steril威 2003;80:625–30. ©2003 by American Society for Reproductive Medicine.) Key Words: Computer sperm analysis, image processing, interactive analysis, segmentation steps, staining conditions
Received September 10, 2002; revised and accepted February 24, 2003. Supported by SR-2 SubReitoria de Po´s-Graduac¸a˜o e Pesquisa da Universidade do Estado do Rio de Janeiro. (UERJ) Brazil. Reprint requests: Luiz Henrique Monteiro-Leal, Ph.D., Universidade do Estado do Rio de Janeiro, Departamento de Histologia e Embriologia, Av. Prof. Manoel de Abreu, 444, 3° andar, Maracana˜, Rio de Janeiro - RJ 20550-170, Brazil (FAX: 55-21-2268-9874; E-mail: [email protected]). 0015-0282/03/$30.00 doi:10.1016/S0015-0282(03) 00978-6
Over the last years, several papers have been published concerning computer methods of semen evaluation (1–3). Some of these methods were based on the strict criteria developed by Kruger and co-authors (4) and were used to predict the possibilities of fertilization success from specific IVF programs. Besides the fact that female factors also have an important influence on the production of positive results in fertilization attempts (5), the use of the strict method as an evaluation tool for semen quality was widely accepted. As pointed out by Coetzee et al. (6), the final goal for any semen analysis system is its successful application in the routine diagnosis of male fertility in vitro. To accomplish this goal, and to find a more useful and applicable method, scientists developed an automatic computer semen analyzer (2, 3). However, the analysis of sperm quality using these automatic methods, performed by
different technicians, with different staining procedures and in different laboratories, still showed substantial differences (7, 8). Moreover, it was pointed out that the establishment of certain quality controls, such as standardized and optimal sample and slide preparation, is imperative. Furthermore, careful focusing, the evaluation of the optimum number of cells (6), and specific training programs for the system operators (9) were also found to be critical. At the level of image processing and digital morphometry, an ideal computer system should be able to compensate for the variations between systems, or between operators, and adjust itself accordingly or should be adjustable to different illumination conditions, objectives, cameras, and computers. To collect new information about the possible variations and errors introduced by semen analysis systems, we created three different computer macros of sperm morphometry anal625
ysis (manual, semiautomatic, and fully automatic methods), which were based on the strict criteria (4). The results obtained suggest that the semiautomatic macro is the most reliable and most secure method for performing sperm analysis, besides the fact that it is a fast process.
MATERIALS AND METHODS The semen preparation was performed as follows: fresh semen was, after complete liquefaction, washed twice in Ham’s F-10 medium (GIBCO Lab., Grand Island, NY), as described by Lacquet et al. (10). The concentration of cells was estimated with the help of a Haemocytometer (Neubauer chamber; Sigma, St. Louis, MO). Dilution to a final concentration of approximately 100 ⫻ 106 cells per milliliter was made with Ham’s F-10. One droplet (25 L) of the latter solution was deposited on slides precleaned with 70% ethyl alcohol, and smears were performed with the help of another slide in angulations of 45%. The preparations were allowed to dry at room temperature and then fixed for 10 seconds in methanol (Solution I, Hemacolor Kit, Merck, Frankfurt, Germany). These slides were stained with hematoxylin and eosin (Merck, Rio de Janeiro, Brazil) or methylene blue 2% in an aqueous solution (Vetec Chemistry, Rio de Janeiro, Brazil) for 10 minutes and then rinsed with water to remove the excess. The slides belonging to one patient were stained with the different dyes in parallel. To check the possible influence of different conditions on the final aspect of the cells, we also analyzed sperm, which were prepared according to the above methodology but with the omission of staining. Using the differential interference contrast, these unstained cells were compared with the same sperm specimens, which were, after this first analysis, stained with the Hemacolor Kit (Merck). The same cells were found after the staining procedure by the use of reference points. The results presented here were obtained after the comparative analysis of 200 sperm heads from 10 individuals (1 slide per person, checked 2 times with each macro), using the microscope stage and the translational control knobs, to prevent the same cell from being analyzed more than once. We used two different systems of video microscopy: system-assigned A, composed of a state-of-the-art scientific Zeiss Axiophot microscope (Zeiss, Oberkochen, Germany) equipped with a highly coherent illumination system (using Koehler illumination and aplanatic/achromatic condenser), high numerical aperture (NA), and aberration-corrected objectives (Plan-APOCHROMAT 63 ⫻ 1.4 N.A. objective; Zeiss), combined with a digital camera SV-Micro (Sound Vision SV-Micro, Zeiss-Vision, Hallbergmoos, Germany) and directly connected to a computer; and the system-assigned B, consisting of a typical laboratory instrument, the KF2 Zeiss microscope (Zeiss), without coherent illumination 626
Goulart et al.
Computer sperm analysis
and with a normal noncorrected, low NA and an oil objective (CP-ACROMAT 1.25 N.A.; Zeiss), and of an analogue color camera (JVC TK-C1380; JVC, Kanagawa, Japan) and a computer equipped with frame grabber Meteor RGB (Matrox, Quebec, Canada). The final magnification obtained (the magnification observed on the monitor) was ⫻5.500 using the scientific instrument, ⫻3,000 in the less sophisticated equipment, and ⫻2,700 using the scientific equipment in the montage with the ⫻40 objective and zoom intermediate lens. All macros were readjusted and recalibrated with the help of a calibrated ruler. The macros used in this work to evaluate the semen morphology were prepared by our group and were run in the Zeiss image-processing system KS400 (Zeiss-Vision). Details of the semiautomatic macro can be obtained elsewhere (11). The manual macro is fully dependent on the operator, who has to trace the maximal and minimal diameters of the cells with the help of the computer mouse (see also Fig. 1). The fully automatic method is a variation of the semiautomatic macro, in which the step of segmentation (object recognition) is performed totally independent of the operator. The thresholds on which this operation was based were obtained after the analysis of the staining pattern of 200 spermatozoa that had been stained with freshly prepared hematoxylin and eosin. In all macros, the program automatically creates data sheets listing the values of minimum and maximum diameters and also the object shape, that is the shape of the sperm head, obtained in the semiautomatic and fully automatic programs. This is done by using the following formula: 4 (area/perimeter2). On the basis of these values, the system uses a classification relying on the strict criteria of Kruger et al. (4) for Diff-Quick staining. The sperm head is classified as perfect only when it fulfils all normality categories. It presents to the operator, besides the above-mentioned morphometric data, the shape and the final fraction or percentage of normal (perfect) and abnormal cells. These percentages of normal cells were used in all comparisons and statistics performed here. To control the analytical ability of each system, nonbiological material with a simple morphology was used. We prepared slides with droplets of 2-m latex beads (mean diameter, 2.16 m; Sigma Chemical Co., St. Louis, MO), which were dissolved in phosphate-buffered saline. The beads were analyzed using the same protocols applied to the spermatozoa preparations. To analyze the features of each macro program and the influence of other variables such as the staining methodology, microscope, and camera, the system operation was performed by two different technicians with different skills and background in what concerns the recognition of normal and abnormal sperm morphology. The technician identified Vol. 80, No. 3, September 2003
FIGURE 1 Operational scheme for the work with three different semen analysis programs. This figure presents a cartoon with the representation of the major steps concerning the computer semen analysis. After the acquisition of the image and its digitalization, the operator chooses one of the three computer methods: the manual (manual), the semiautomatic (semiautomatic), and the fully automatic (full-automatic). In the manual method, the operator should trace the sperm head limits (M) with the help of the computer mouse. In the semiautomatic method, the operator interacting with the computer adjusts the window in the segmentation step (A) and creates a binary image (B) with the corrected sperm heads that have been identified. In the fully automatic method, the operator has no control on or choice of interactions with the system.
TABLE 1 Comparison between macros using different staining conditions.
Staining H&E Meth. blue
Manual macro
Semiautomatic macro
Fully automatic macro
37.1 ⫾ 9.0 21.5 ⫾ 6.4
28.1 ⫾ 10.1 26.5 ⫾ 5.9
19.3 ⫾ 2.0 2.3 ⫾ 1.8
Note: Percentages and standard deviation of normal spermatozoa in 200 analyses per slide from the same 10 patients. H&E ⫽ hematoxylin and eosin staining; Meth. blue ⫽ methylene blue staining. Goulart. Computer sperm analysis. Fertil Steril 2003.
comparison of two groups of data, which are presented in Tables 2 and 3.
RESULTS To analyze the competence of the computer programs to obtain confident results in what concerns the sperm morphology, we created three different computer macros, operating in the KS 400 image-processing system (Fig. 1). In the manual macro, after automatic acquisition and image cali-
TABLE 2 Comparison among macros using the same staining (hematoxylin and eosin) and involving different technicians.
Technician
Manual macro
Semiautomatic macro
Fully automatic macro
1 2
33.8 ⫾ 8.4 32.5 ⫾ 8.5
23.2 ⫾ 5.3 24.3 ⫾ 6.5
15.5 ⫾ 9.2 11.5 ⫾ 7.0
Note: Percentages and standard deviations of normal spermatozoa in 200 analyses per slide from the same 10 patients. Goulart. Computer sperm analysis. Fertil Steril 2003.
Goulart. Computer sperm analysis. Fertil Steril 2003.
as number 1 has ⱖ4 years of experience with semen analysis and computer techniques. The technician identified as number 2 has ⱕ1 year of experience with the system. The results obtained using the different macros, systems, and technicians were statistically analyzed using analysis of variance (ANOVA) (Graphpad Software Inc., San Diego, CA) for multiple samples. In this study the ANOVA was used for comparisons of the results obtained in the three macros, as was the case in Table 1. Alternatively, the Student’s t test (Graphpad Software Inc.) was used for the FERTILITY & STERILITY威
TABLE 3 Comparison between macros using different microscope systems.
Systema
Manual macro
Semiautomatic macro
Fully automatic macro
A B
35.2 ⫾ 9.0 20.7 ⫾ 7.9
27.2 ⫾ 6.7 30.1 ⫾ 7.8
15.4 ⫾ 8.9 4.9 ⫾ 3.4
Note: Percentages and standard deviations of normal spermatozoa in 200 analyses per slide from the same 10 patients. a A, Scientific state-of-the-art microscope; B, basic laboratory microscope. Goulart. Computer sperm analysis. Fertil Steril 2003.
627
bration, the system allowed the operator to indicate the object to be measured by tracing the maximum and minimum cell diameters (in this case: the sperm head) with the help of the computer mouse. After the end of this operation, the computer automatically accumulated the data and allowed the acquisition of a new image. In the semiautomatic macro, the first steps were the same as already discussed in relation to the manual macro. However, in the step of object recognition, the system offers to the operator a predefined threshold of segmentation. If the automatic recognition was not perfect, the operator had the possibility to interact with the system. The operator could correct and adjust the thresholds to be sure that nothing but the sperm head was recognized by the computer as the object to be measured (Fig. 1A and B). This threshold step was fully dependent on the color characteristics of the sample, as it was not possible to use another morphological recognition tool. In the automatic macro, the range of the thresholds was previously determined. In other words, the operator had no opportunity to adjust it and thereby to correct possible mistakes (Fig. 1). Table 2 shows the data obtained when the three different computer programs were operated by different technicians with different skills. The statistical analysis of these data by unpaired Student’s t test showed that the differences in the values obtained by the two technicians in the automatic analysis (P value of .2880) were considered not significant. The same holds true for both the semiautomatic (P value of .6834) and the manual (P⫽.7351) analyses. On the other hand, the results obtained for each technician in each macro were considerably different. The percentage of normal cells was smaller in the application of the automatic and larger in the manual method. The semiautomatic system presented results that were in between the others (Table 2). The statistical (unpaired Student’s t test) comparison showed that these differences were statistically significant between the automatic and the manual ones (P⫽.0004) but were not significant between the semiautomatic and the manual macros (P⫽.0513). Table 1 shows the results obtained by the same technician working with the same samples and using two different staining conditions. The data and the statistical analysis using the ANOVA one-way analysis of variance and the Tukey-Kramer comparison test showed that there is a statistically significant difference between the results obtained by different staining conditions for the same slide sets (10 different slides from 10 individuals, with 200 sperm heads analyzed per slide), in the case of the automatic macros (P⫽.0001) and also in the case of the manual one (P⫽.0003). In contrast, the analysis showed that there were no statistically significant differences between the semiautomatic 628
Goulart et al.
Computer sperm analysis
data obtained with the two different staining conditions (P⫽.6717). We also checked the capability of the semiautomatic method to first analyze unstained samples observed by differential interference contrast and then to analyze the same cells stained by Hemacolor. Despite minor differences in diameters observed before and after staining, no great difference in the final results was obtained (data not shown). The analysis done with the three computer programs using two different microscope sets, one (system A) composed of a state-of-the-art microscope equipped with digital camera, and the other one (system B) composed of a more simple laboratory microscope equipped with a color analogue camera, showed the following results. The fully automatic macro presents discrepant data in a comparison of the results obtained from the same 10 samples (10 different slides from 10 individuals, with 200 sperm heads analyzed per slide; Table 3). The statistical analysis (unpaired t test) of the values obtained for the number of normal spermatozoa in the 10 individuals showed a statistically significant variance (P⫽.0026) between the two microscope systems, when one used the fully automatic macro (Table 3). The statistical analysis performed with the data from the manual macro shows that there was also a statistically significant difference (P value of .0019) between the results, when the same set of slides was analyzed in the two different microscope setups. On the other hand, the analysis of slide sets, with the semiautomatic method in the two different microscope systems, presents the smallest variability between medians compared with the case of the other two computer methods. The statistical analysis (unpaired t test) confirms that there was no statistically significant difference between the semiautomatic analyses performed (P⫽.3840). The ability of the three different programs to evaluate nonbiological material was tested by the use of 2-m latex beads, which were observed by standard transmitted light microscopy. The results showed that the three systems were able to determine the correct ratio between maximum and minimum diameter (around 1), but manual analysis presented the greater variability (data not shown). We also analyzed the mean time needed for one operator to analyze one sample (spermatozoa heads) using the three different computer methods and a totally manual procedure) using the microscope system A. It is possible to see that the fully automatic macro is the fastest, followed by the semiautomatic and then by the computer manual one. The findings also showed that to perform the morphometric analysis of the sperm head without the help of the computer, one needs approximately twice the time that is needed when having help from the machine, even if the manual macro is used (data not shown). Vol. 80, No. 3, September 2003
DISCUSSION The aim of this work was not to present a new method of computer semen analysis. On the contrary, our objective here was to collect new information on possible reasons for the variations of results obtained using the automatic semen analysis systems (7, 8). A semiautomatic program of sperm analysis developed by our group (11) was the starting point for the creation of the other computer macros. Our results show that fully automatic systems are perfect for evaluating nonvariable, preferably nonbiological materials, such as latex beads. The fully automatic systems in these cases show a greater reproducibility, in comparison to manual systems, and are equal to semiautomatic programs in terms of correctness. Nevertheless, latex beads, on the other hand, show no differences either in color or shape. So the segmentation steps created are efficient to identify all objects, as observed in our assay. The same did not hold true for biological specimens. As could be concluded from the data in Table 1, the fully automatic method was not able to correctly recognize the sperm heads. In other words, the system that was calibrated for the use of one staining method, in this case hematoxylin and eosin, did not correctly work when another technique was applied (Table 1). This is not the case, however, when one uses the semiautomatic method. The interaction between the operator and the machine was sufficient to correct the differences caused by the staining method, in such a way that the comparison between the data presents no variability independent of the differences in the characteristics of the dyes applied in this work. Moreover, the analysis of the same sperm heads before and after the staining procedure revealed that using the interactive protocol, one could minimize the effects of the staining step in the preparations to obtain a corrected evaluation. It is almost impossible to standardize microscope and camera conditions among laboratories; therefore, we would expect that the variability in color, resolution, and contrast created by the devices involved could mislead the automatic system. An indication of this problem is obvious from data in Tables 1 and 3. Again, the fully automatic system was not able to compensate for the variations among microscopes, camera modules, and staining procedures. The operator can correct, only in the semiautomatic programs, the changes in microscope systems. So the program can always correctly identify the object, whatever the microscope conditions are (Table 3). When the same set of slides was analyzed by the manual computer method, again some differences were observed (Table 3). This probably means that an additional factor influencing the results was the ability of the operator to trace the correct sperm head. The use of latex beads (data not shown) showed that the manual method presented the strongest variability. This FERTILITY & STERILITY威
could be interpreted as a manifestation of the negative influence exerted by the operator. The results of manual analysis of the sperm morphometry in all cases represented, independent of the staining, the microscope, or the operator, values of normal or perfect spermatozoa higher than those of the automatic and the semiautomatic programs. This is probably due to a superestimation of the number of normal cells, possibly caused by an involuntary tendency of the operator to draw a perfectly shaped and sized spermatozoon, even if this was not the case. During the experiments using the semiautomatic macro, we observed that in 20%–30% of all cases, the computer was able to correctly identify the whole sperm head. Nevertheless, in some cases the computer misinterpreted the entity of head, part of the midpiece together with part of the tail, and in other cases only a part of the head, as the sperm head. This is probably the reason that in Table 2, even when the staining conditions were the same, the fully automatic macro failed to present correct results, underestimating the percentage of normal sperm heads. This phenomenon was even more pronounced when the macro, calibrated with one particular staining method, was used to analyze sperms stained with an additional method (Table 1). The intervention of the technician in this series of analysis was fundamental for the corrected identification of the object by the computer, and, of course, for the generation of correct results. The systems created by us analyze the overall morphological aspect of the sperm head. Some other programs (e.g., IVOS, Spermorph) are designed to acquire information also from the acrosome and midpiece. When these other parameters were included, the variation observed was enormous (data not shown). After the establishment of the best method (semiautomatic), we are now adjusting our system and using high numerical aperture objectives and condensers to be capable of presenting reliable data, even when the variations in the analyzed structures are in the range of only a few hundred nanometers, as these other aspects suggested by the World Health Organization laboratory manual (12). Wherever the source of variance is, that is, the stain, microscope, illumination, objective, camera, or computer (or integrated systems), it is imperative that the software permits the intervention of the technician in its most important step, that is the recognition of the object. We conclude from this work that fully automatic systems perfectly work with nonbiological material, but sometimes fail to recognize more complex objects of interest, especially when the morphological aspect, their shape, cannot be reliably used as an identification tool, in addition to color and size. The manual computer method is the most tedious, laborious, and slow method, besides the fact that it totally depends on the routine and skills of the operator. The data presented here show that the use of semiautomatic methods is the best strategy because they present a compromise between the fast automatic method and the 629
permanent involvement of the operator. Therefore, semiautomatic evaluation should be considered as the best solution for morphometrical sperm head analysis computer systems.
Acknowledgments: The authors thank Helmut Troester, Ph.D. (DKFZ Heidelberg, Germany), for the critical revision of the manuscript and Adriano Caldeira de Arau´ jo, Ph.D. (UERJ-Brazil), for assistance with the statistical analysis.
References 1. Davis RO, Gravance CG, Casey PJ. Automated morphometric analysis of stallion spermatozoa. Am J Vet Res 1993;54:1808 –11. 2. Kruger TF, DuToit TC, Franken DR, Acosta AA, Oehninger SC, Menkveld R, et al. A new computerized method of reading sperm morphology (strict criteria) is as efficient as technician reading. Fertil Steril 1993;59:202–9. 3. Harr R. Characterization of spermatozoa by planar morphometry. Clin Lab Sci 1997;10:190 –6. 4. Kruger TF, Menkveld R, Stander FS, Lombard CJ, Van der Merwe JP, van Zyl JA, et al. Sperm morphologic features as a prognostic factor in in vitro fertilization. Fertil Steril 1986;46:1118 –23.
630
Goulart et al.
Computer sperm analysis
5. Montanaro Gauci M, Kruger TF, Coetzee K, Smith K, Van Der Merwe JP, Lombard CJ. Stepwise regression analysis to study male and female factors impacting on pregnancy rate in an intrauterine insemination programme. Andrologia 2001;33:135–41. 6. Coetzee K, Kruger TF, Lombard CJ. Repeatability and variance analysis on multiple computer assisted (IVOS*) sperm morphology readings. Andrologia 1999;31:163–8. 7. Menkveld R, Lacquet FA, Kruger TF, Lombard CJ, Sanchez Sarmiento CA, de Villiers A. Effects of different staining and washing procedures on the results of human sperm morphology evaluation by manual and computerized methods. Andrologia 1997;29:1–7. 8. Barroso G, Mercan R, Ozgur K, Morshedi M, Kolm P, Coetzee K, et al. Intra- and inter-laboratory variability in the assessment of sperm morphology by strict criteria: impact of semen preparation, staining techniques and manual versus computerized analysis. Hum Reprod 1999; 14:2036 – 40. 9. Franken DR, Smith M, Menkveld R. The development of a continuous quality control programme for strict sperm morphology among subSaharan African laboratories. Hum Reprod 2000;15:667–71. 10. Lacquet FA, Kruger TF, Du Toit TC, Lombard CJ, Sanchez Sarmiento CA, De Villiers A, et al. Slide preparation and staining procedures for reliable results using computerised morphology (IVOS). Arch Androl 1996;36:133–8. 11. Monteiro-Leal LH, Goulart AR, Lopes VWS, Ferreira J, Lourencini R, Campanati L. Novo me´ todo semi-automa´ tico para avaliac¸ a˜ o da morfologia de espermatozo´ ides humanos. J Bras Urol 1999;25:235–40. 12. World Health Organization. Laboratory manual for the examination of human semen and sperm-cervical mucus interaction. 4th ed. Cambridge, UK: Cambridge University Press, 1999:4 –33.
Vol. 80, No. 3, September 2003
Comparison of three computer methods of sperm head analysis Ariadne Rodrigues Goulart, M.Sc., Moema de Alencar Hausen, B.Sc., and Luiz Henrique Monteiro-Leal, Ph.D. Universidade do Estado do Rio de Janeiro, Departamento de Histologia e Embriologia, Laborato´rio de Microscopia e Processamento de Imagens, Rio de Janeiro, Brazil
Objective: Analysis of sperm heads using three different computer morphometrical tools and experimental conditions to find a more reliable and secure strategy among them. Design: Controlled experiments on sperm morphology analysis from volunteers. Setting: Laboratory of microscopy and imaging processing. Patient(s): Ten human semen samples donated by different zoospermic men. Intervention(s): Semen samples were collected by masturbation after ⱖ72 hours of abstinence. Main Outcome Measure(s): Spermatozoon head morphology was compared by the use of different videomicroscopy systems, three computer programs, and various staining conditions and manipulation by different operators. Nonbiological material in the form of latex beads was also used. Result(s): The data obtained suggest that the semiautomatic computer program is the most reliable and secure method for performing sperm analysis, besides the fact that it is a fast process compared with manual methods. Conclusion(s): Computer systems of sperm analysis should incorporate a step of interactive object identification to work properly, allowing the operator to confirm or correct possible computer misidentification. The latex beads were used to confirm the capability of all three computer programs to correctly evaluate nonbiological material. (Fertil Steril威 2003;80:625–30. ©2003 by American Society for Reproductive Medicine.) Key Words: Computer sperm analysis, image processing, interactive analysis, segmentation steps, staining conditions
Received September 10, 2002; revised and accepted February 24, 2003. Supported by SR-2 SubReitoria de Po´s-Graduac¸a˜o e Pesquisa da Universidade do Estado do Rio de Janeiro. (UERJ) Brazil. Reprint requests: Luiz Henrique Monteiro-Leal, Ph.D., Universidade do Estado do Rio de Janeiro, Departamento de Histologia e Embriologia, Av. Prof. Manoel de Abreu, 444, 3° andar, Maracana˜, Rio de Janeiro - RJ 20550-170, Brazil (FAX: 55-21-2268-9874; E-mail: [email protected]). 0015-0282/03/$30.00 doi:10.1016/S0015-0282(03) 00978-6
Over the last years, several papers have been published concerning computer methods of semen evaluation (1–3). Some of these methods were based on the strict criteria developed by Kruger and co-authors (4) and were used to predict the possibilities of fertilization success from specific IVF programs. Besides the fact that female factors also have an important influence on the production of positive results in fertilization attempts (5), the use of the strict method as an evaluation tool for semen quality was widely accepted. As pointed out by Coetzee et al. (6), the final goal for any semen analysis system is its successful application in the routine diagnosis of male fertility in vitro. To accomplish this goal, and to find a more useful and applicable method, scientists developed an automatic computer semen analyzer (2, 3). However, the analysis of sperm quality using these automatic methods, performed by
different technicians, with different staining procedures and in different laboratories, still showed substantial differences (7, 8). Moreover, it was pointed out that the establishment of certain quality controls, such as standardized and optimal sample and slide preparation, is imperative. Furthermore, careful focusing, the evaluation of the optimum number of cells (6), and specific training programs for the system operators (9) were also found to be critical. At the level of image processing and digital morphometry, an ideal computer system should be able to compensate for the variations between systems, or between operators, and adjust itself accordingly or should be adjustable to different illumination conditions, objectives, cameras, and computers. To collect new information about the possible variations and errors introduced by semen analysis systems, we created three different computer macros of sperm morphometry anal625
ysis (manual, semiautomatic, and fully automatic methods), which were based on the strict criteria (4). The results obtained suggest that the semiautomatic macro is the most reliable and most secure method for performing sperm analysis, besides the fact that it is a fast process.
MATERIALS AND METHODS The semen preparation was performed as follows: fresh semen was, after complete liquefaction, washed twice in Ham’s F-10 medium (GIBCO Lab., Grand Island, NY), as described by Lacquet et al. (10). The concentration of cells was estimated with the help of a Haemocytometer (Neubauer chamber; Sigma, St. Louis, MO). Dilution to a final concentration of approximately 100 ⫻ 106 cells per milliliter was made with Ham’s F-10. One droplet (25 L) of the latter solution was deposited on slides precleaned with 70% ethyl alcohol, and smears were performed with the help of another slide in angulations of 45%. The preparations were allowed to dry at room temperature and then fixed for 10 seconds in methanol (Solution I, Hemacolor Kit, Merck, Frankfurt, Germany). These slides were stained with hematoxylin and eosin (Merck, Rio de Janeiro, Brazil) or methylene blue 2% in an aqueous solution (Vetec Chemistry, Rio de Janeiro, Brazil) for 10 minutes and then rinsed with water to remove the excess. The slides belonging to one patient were stained with the different dyes in parallel. To check the possible influence of different conditions on the final aspect of the cells, we also analyzed sperm, which were prepared according to the above methodology but with the omission of staining. Using the differential interference contrast, these unstained cells were compared with the same sperm specimens, which were, after this first analysis, stained with the Hemacolor Kit (Merck). The same cells were found after the staining procedure by the use of reference points. The results presented here were obtained after the comparative analysis of 200 sperm heads from 10 individuals (1 slide per person, checked 2 times with each macro), using the microscope stage and the translational control knobs, to prevent the same cell from being analyzed more than once. We used two different systems of video microscopy: system-assigned A, composed of a state-of-the-art scientific Zeiss Axiophot microscope (Zeiss, Oberkochen, Germany) equipped with a highly coherent illumination system (using Koehler illumination and aplanatic/achromatic condenser), high numerical aperture (NA), and aberration-corrected objectives (Plan-APOCHROMAT 63 ⫻ 1.4 N.A. objective; Zeiss), combined with a digital camera SV-Micro (Sound Vision SV-Micro, Zeiss-Vision, Hallbergmoos, Germany) and directly connected to a computer; and the system-assigned B, consisting of a typical laboratory instrument, the KF2 Zeiss microscope (Zeiss), without coherent illumination 626
Goulart et al.
Computer sperm analysis
and with a normal noncorrected, low NA and an oil objective (CP-ACROMAT 1.25 N.A.; Zeiss), and of an analogue color camera (JVC TK-C1380; JVC, Kanagawa, Japan) and a computer equipped with frame grabber Meteor RGB (Matrox, Quebec, Canada). The final magnification obtained (the magnification observed on the monitor) was ⫻5.500 using the scientific instrument, ⫻3,000 in the less sophisticated equipment, and ⫻2,700 using the scientific equipment in the montage with the ⫻40 objective and zoom intermediate lens. All macros were readjusted and recalibrated with the help of a calibrated ruler. The macros used in this work to evaluate the semen morphology were prepared by our group and were run in the Zeiss image-processing system KS400 (Zeiss-Vision). Details of the semiautomatic macro can be obtained elsewhere (11). The manual macro is fully dependent on the operator, who has to trace the maximal and minimal diameters of the cells with the help of the computer mouse (see also Fig. 1). The fully automatic method is a variation of the semiautomatic macro, in which the step of segmentation (object recognition) is performed totally independent of the operator. The thresholds on which this operation was based were obtained after the analysis of the staining pattern of 200 spermatozoa that had been stained with freshly prepared hematoxylin and eosin. In all macros, the program automatically creates data sheets listing the values of minimum and maximum diameters and also the object shape, that is the shape of the sperm head, obtained in the semiautomatic and fully automatic programs. This is done by using the following formula: 4 (area/perimeter2). On the basis of these values, the system uses a classification relying on the strict criteria of Kruger et al. (4) for Diff-Quick staining. The sperm head is classified as perfect only when it fulfils all normality categories. It presents to the operator, besides the above-mentioned morphometric data, the shape and the final fraction or percentage of normal (perfect) and abnormal cells. These percentages of normal cells were used in all comparisons and statistics performed here. To control the analytical ability of each system, nonbiological material with a simple morphology was used. We prepared slides with droplets of 2-m latex beads (mean diameter, 2.16 m; Sigma Chemical Co., St. Louis, MO), which were dissolved in phosphate-buffered saline. The beads were analyzed using the same protocols applied to the spermatozoa preparations. To analyze the features of each macro program and the influence of other variables such as the staining methodology, microscope, and camera, the system operation was performed by two different technicians with different skills and background in what concerns the recognition of normal and abnormal sperm morphology. The technician identified Vol. 80, No. 3, September 2003
FIGURE 1 Operational scheme for the work with three different semen analysis programs. This figure presents a cartoon with the representation of the major steps concerning the computer semen analysis. After the acquisition of the image and its digitalization, the operator chooses one of the three computer methods: the manual (manual), the semiautomatic (semiautomatic), and the fully automatic (full-automatic). In the manual method, the operator should trace the sperm head limits (M) with the help of the computer mouse. In the semiautomatic method, the operator interacting with the computer adjusts the window in the segmentation step (A) and creates a binary image (B) with the corrected sperm heads that have been identified. In the fully automatic method, the operator has no control on or choice of interactions with the system.
TABLE 1 Comparison between macros using different staining conditions.
Staining H&E Meth. blue
Manual macro
Semiautomatic macro
Fully automatic macro
37.1 ⫾ 9.0 21.5 ⫾ 6.4
28.1 ⫾ 10.1 26.5 ⫾ 5.9
19.3 ⫾ 2.0 2.3 ⫾ 1.8
Note: Percentages and standard deviation of normal spermatozoa in 200 analyses per slide from the same 10 patients. H&E ⫽ hematoxylin and eosin staining; Meth. blue ⫽ methylene blue staining. Goulart. Computer sperm analysis. Fertil Steril 2003.
comparison of two groups of data, which are presented in Tables 2 and 3.
RESULTS To analyze the competence of the computer programs to obtain confident results in what concerns the sperm morphology, we created three different computer macros, operating in the KS 400 image-processing system (Fig. 1). In the manual macro, after automatic acquisition and image cali-
TABLE 2 Comparison among macros using the same staining (hematoxylin and eosin) and involving different technicians.
Technician
Manual macro
Semiautomatic macro
Fully automatic macro
1 2
33.8 ⫾ 8.4 32.5 ⫾ 8.5
23.2 ⫾ 5.3 24.3 ⫾ 6.5
15.5 ⫾ 9.2 11.5 ⫾ 7.0
Note: Percentages and standard deviations of normal spermatozoa in 200 analyses per slide from the same 10 patients. Goulart. Computer sperm analysis. Fertil Steril 2003.
Goulart. Computer sperm analysis. Fertil Steril 2003.
as number 1 has ⱖ4 years of experience with semen analysis and computer techniques. The technician identified as number 2 has ⱕ1 year of experience with the system. The results obtained using the different macros, systems, and technicians were statistically analyzed using analysis of variance (ANOVA) (Graphpad Software Inc., San Diego, CA) for multiple samples. In this study the ANOVA was used for comparisons of the results obtained in the three macros, as was the case in Table 1. Alternatively, the Student’s t test (Graphpad Software Inc.) was used for the FERTILITY & STERILITY威
TABLE 3 Comparison between macros using different microscope systems.
Systema
Manual macro
Semiautomatic macro
Fully automatic macro
A B
35.2 ⫾ 9.0 20.7 ⫾ 7.9
27.2 ⫾ 6.7 30.1 ⫾ 7.8
15.4 ⫾ 8.9 4.9 ⫾ 3.4
Note: Percentages and standard deviations of normal spermatozoa in 200 analyses per slide from the same 10 patients. a A, Scientific state-of-the-art microscope; B, basic laboratory microscope. Goulart. Computer sperm analysis. Fertil Steril 2003.
627
bration, the system allowed the operator to indicate the object to be measured by tracing the maximum and minimum cell diameters (in this case: the sperm head) with the help of the computer mouse. After the end of this operation, the computer automatically accumulated the data and allowed the acquisition of a new image. In the semiautomatic macro, the first steps were the same as already discussed in relation to the manual macro. However, in the step of object recognition, the system offers to the operator a predefined threshold of segmentation. If the automatic recognition was not perfect, the operator had the possibility to interact with the system. The operator could correct and adjust the thresholds to be sure that nothing but the sperm head was recognized by the computer as the object to be measured (Fig. 1A and B). This threshold step was fully dependent on the color characteristics of the sample, as it was not possible to use another morphological recognition tool. In the automatic macro, the range of the thresholds was previously determined. In other words, the operator had no opportunity to adjust it and thereby to correct possible mistakes (Fig. 1). Table 2 shows the data obtained when the three different computer programs were operated by different technicians with different skills. The statistical analysis of these data by unpaired Student’s t test showed that the differences in the values obtained by the two technicians in the automatic analysis (P value of .2880) were considered not significant. The same holds true for both the semiautomatic (P value of .6834) and the manual (P⫽.7351) analyses. On the other hand, the results obtained for each technician in each macro were considerably different. The percentage of normal cells was smaller in the application of the automatic and larger in the manual method. The semiautomatic system presented results that were in between the others (Table 2). The statistical (unpaired Student’s t test) comparison showed that these differences were statistically significant between the automatic and the manual ones (P⫽.0004) but were not significant between the semiautomatic and the manual macros (P⫽.0513). Table 1 shows the results obtained by the same technician working with the same samples and using two different staining conditions. The data and the statistical analysis using the ANOVA one-way analysis of variance and the Tukey-Kramer comparison test showed that there is a statistically significant difference between the results obtained by different staining conditions for the same slide sets (10 different slides from 10 individuals, with 200 sperm heads analyzed per slide), in the case of the automatic macros (P⫽.0001) and also in the case of the manual one (P⫽.0003). In contrast, the analysis showed that there were no statistically significant differences between the semiautomatic 628
Goulart et al.
Computer sperm analysis
data obtained with the two different staining conditions (P⫽.6717). We also checked the capability of the semiautomatic method to first analyze unstained samples observed by differential interference contrast and then to analyze the same cells stained by Hemacolor. Despite minor differences in diameters observed before and after staining, no great difference in the final results was obtained (data not shown). The analysis done with the three computer programs using two different microscope sets, one (system A) composed of a state-of-the-art microscope equipped with digital camera, and the other one (system B) composed of a more simple laboratory microscope equipped with a color analogue camera, showed the following results. The fully automatic macro presents discrepant data in a comparison of the results obtained from the same 10 samples (10 different slides from 10 individuals, with 200 sperm heads analyzed per slide; Table 3). The statistical analysis (unpaired t test) of the values obtained for the number of normal spermatozoa in the 10 individuals showed a statistically significant variance (P⫽.0026) between the two microscope systems, when one used the fully automatic macro (Table 3). The statistical analysis performed with the data from the manual macro shows that there was also a statistically significant difference (P value of .0019) between the results, when the same set of slides was analyzed in the two different microscope setups. On the other hand, the analysis of slide sets, with the semiautomatic method in the two different microscope systems, presents the smallest variability between medians compared with the case of the other two computer methods. The statistical analysis (unpaired t test) confirms that there was no statistically significant difference between the semiautomatic analyses performed (P⫽.3840). The ability of the three different programs to evaluate nonbiological material was tested by the use of 2-m latex beads, which were observed by standard transmitted light microscopy. The results showed that the three systems were able to determine the correct ratio between maximum and minimum diameter (around 1), but manual analysis presented the greater variability (data not shown). We also analyzed the mean time needed for one operator to analyze one sample (spermatozoa heads) using the three different computer methods and a totally manual procedure) using the microscope system A. It is possible to see that the fully automatic macro is the fastest, followed by the semiautomatic and then by the computer manual one. The findings also showed that to perform the morphometric analysis of the sperm head without the help of the computer, one needs approximately twice the time that is needed when having help from the machine, even if the manual macro is used (data not shown). Vol. 80, No. 3, September 2003
DISCUSSION The aim of this work was not to present a new method of computer semen analysis. On the contrary, our objective here was to collect new information on possible reasons for the variations of results obtained using the automatic semen analysis systems (7, 8). A semiautomatic program of sperm analysis developed by our group (11) was the starting point for the creation of the other computer macros. Our results show that fully automatic systems are perfect for evaluating nonvariable, preferably nonbiological materials, such as latex beads. The fully automatic systems in these cases show a greater reproducibility, in comparison to manual systems, and are equal to semiautomatic programs in terms of correctness. Nevertheless, latex beads, on the other hand, show no differences either in color or shape. So the segmentation steps created are efficient to identify all objects, as observed in our assay. The same did not hold true for biological specimens. As could be concluded from the data in Table 1, the fully automatic method was not able to correctly recognize the sperm heads. In other words, the system that was calibrated for the use of one staining method, in this case hematoxylin and eosin, did not correctly work when another technique was applied (Table 1). This is not the case, however, when one uses the semiautomatic method. The interaction between the operator and the machine was sufficient to correct the differences caused by the staining method, in such a way that the comparison between the data presents no variability independent of the differences in the characteristics of the dyes applied in this work. Moreover, the analysis of the same sperm heads before and after the staining procedure revealed that using the interactive protocol, one could minimize the effects of the staining step in the preparations to obtain a corrected evaluation. It is almost impossible to standardize microscope and camera conditions among laboratories; therefore, we would expect that the variability in color, resolution, and contrast created by the devices involved could mislead the automatic system. An indication of this problem is obvious from data in Tables 1 and 3. Again, the fully automatic system was not able to compensate for the variations among microscopes, camera modules, and staining procedures. The operator can correct, only in the semiautomatic programs, the changes in microscope systems. So the program can always correctly identify the object, whatever the microscope conditions are (Table 3). When the same set of slides was analyzed by the manual computer method, again some differences were observed (Table 3). This probably means that an additional factor influencing the results was the ability of the operator to trace the correct sperm head. The use of latex beads (data not shown) showed that the manual method presented the strongest variability. This FERTILITY & STERILITY威
could be interpreted as a manifestation of the negative influence exerted by the operator. The results of manual analysis of the sperm morphometry in all cases represented, independent of the staining, the microscope, or the operator, values of normal or perfect spermatozoa higher than those of the automatic and the semiautomatic programs. This is probably due to a superestimation of the number of normal cells, possibly caused by an involuntary tendency of the operator to draw a perfectly shaped and sized spermatozoon, even if this was not the case. During the experiments using the semiautomatic macro, we observed that in 20%–30% of all cases, the computer was able to correctly identify the whole sperm head. Nevertheless, in some cases the computer misinterpreted the entity of head, part of the midpiece together with part of the tail, and in other cases only a part of the head, as the sperm head. This is probably the reason that in Table 2, even when the staining conditions were the same, the fully automatic macro failed to present correct results, underestimating the percentage of normal sperm heads. This phenomenon was even more pronounced when the macro, calibrated with one particular staining method, was used to analyze sperms stained with an additional method (Table 1). The intervention of the technician in this series of analysis was fundamental for the corrected identification of the object by the computer, and, of course, for the generation of correct results. The systems created by us analyze the overall morphological aspect of the sperm head. Some other programs (e.g., IVOS, Spermorph) are designed to acquire information also from the acrosome and midpiece. When these other parameters were included, the variation observed was enormous (data not shown). After the establishment of the best method (semiautomatic), we are now adjusting our system and using high numerical aperture objectives and condensers to be capable of presenting reliable data, even when the variations in the analyzed structures are in the range of only a few hundred nanometers, as these other aspects suggested by the World Health Organization laboratory manual (12). Wherever the source of variance is, that is, the stain, microscope, illumination, objective, camera, or computer (or integrated systems), it is imperative that the software permits the intervention of the technician in its most important step, that is the recognition of the object. We conclude from this work that fully automatic systems perfectly work with nonbiological material, but sometimes fail to recognize more complex objects of interest, especially when the morphological aspect, their shape, cannot be reliably used as an identification tool, in addition to color and size. The manual computer method is the most tedious, laborious, and slow method, besides the fact that it totally depends on the routine and skills of the operator. The data presented here show that the use of semiautomatic methods is the best strategy because they present a compromise between the fast automatic method and the 629
permanent involvement of the operator. Therefore, semiautomatic evaluation should be considered as the best solution for morphometrical sperm head analysis computer systems.
Acknowledgments: The authors thank Helmut Troester, Ph.D. (DKFZ Heidelberg, Germany), for the critical revision of the manuscript and Adriano Caldeira de Arau´ jo, Ph.D. (UERJ-Brazil), for assistance with the statistical analysis.
References 1. Davis RO, Gravance CG, Casey PJ. Automated morphometric analysis of stallion spermatozoa. Am J Vet Res 1993;54:1808 –11. 2. Kruger TF, DuToit TC, Franken DR, Acosta AA, Oehninger SC, Menkveld R, et al. A new computerized method of reading sperm morphology (strict criteria) is as efficient as technician reading. Fertil Steril 1993;59:202–9. 3. Harr R. Characterization of spermatozoa by planar morphometry. Clin Lab Sci 1997;10:190 –6. 4. Kruger TF, Menkveld R, Stander FS, Lombard CJ, Van der Merwe JP, van Zyl JA, et al. Sperm morphologic features as a prognostic factor in in vitro fertilization. Fertil Steril 1986;46:1118 –23.
630
Goulart et al.
Computer sperm analysis
5. Montanaro Gauci M, Kruger TF, Coetzee K, Smith K, Van Der Merwe JP, Lombard CJ. Stepwise regression analysis to study male and female factors impacting on pregnancy rate in an intrauterine insemination programme. Andrologia 2001;33:135–41. 6. Coetzee K, Kruger TF, Lombard CJ. Repeatability and variance analysis on multiple computer assisted (IVOS*) sperm morphology readings. Andrologia 1999;31:163–8. 7. Menkveld R, Lacquet FA, Kruger TF, Lombard CJ, Sanchez Sarmiento CA, de Villiers A. Effects of different staining and washing procedures on the results of human sperm morphology evaluation by manual and computerized methods. Andrologia 1997;29:1–7. 8. Barroso G, Mercan R, Ozgur K, Morshedi M, Kolm P, Coetzee K, et al. Intra- and inter-laboratory variability in the assessment of sperm morphology by strict criteria: impact of semen preparation, staining techniques and manual versus computerized analysis. Hum Reprod 1999; 14:2036 – 40. 9. Franken DR, Smith M, Menkveld R. The development of a continuous quality control programme for strict sperm morphology among subSaharan African laboratories. Hum Reprod 2000;15:667–71. 10. Lacquet FA, Kruger TF, Du Toit TC, Lombard CJ, Sanchez Sarmiento CA, De Villiers A, et al. Slide preparation and staining procedures for reliable results using computerised morphology (IVOS). Arch Androl 1996;36:133–8. 11. Monteiro-Leal LH, Goulart AR, Lopes VWS, Ferreira J, Lourencini R, Campanati L. Novo me´ todo semi-automa´ tico para avaliac¸ a˜ o da morfologia de espermatozo´ ides humanos. J Bras Urol 1999;25:235–40. 12. World Health Organization. Laboratory manual for the examination of human semen and sperm-cervical mucus interaction. 4th ed. Cambridge, UK: Cambridge University Press, 1999:4 –33.
Vol. 80, No. 3, September 2003