Kruger T F, du Toit T C, Franken D R, Menkveld R, Lombard C J
University of Stellenbosch, Tygerberg, Republic of South Africa.
Fertil Steril. 1995 Jan;63(1):134-41. doi: 10.1016/s0015-0282(16)57308-7.
To correlate the percentage normal morphology reported by different observers and a computerized method (IVOS; Hamilton-Thorne Research, Beverly, MA) on a slide-by-slide basis using strict criteria: [1] Experienced observer (T.F.K.) versus experienced observer (R.M.), [2] experienced observer (T.F.K.) versus sperm morphology analyzer (IVOS), and [3] repeatability of normal and abnormal cells (IVOS). DESIGN SETTING, PATIENTS: Slides from 30 different patients from the Tygerberg IVF program were selected randomly. Microscopic fields and sperm cells were chosen randomly and percent normal morphology was recorded (objectives 1 and 2). The same slides were used and a cell-by-cell repeatability was done as outlined (objective 3).
Experiment 1 (objective 1): there was no significant bias between T.F.K. and R.M. The limits of agreement were 8.6% and -7.3%. The SDs were not significantly different (P = 0.1283). The Spearman correlation coefficient between readers was 0.83. Experiment 2 (objective 2): the same findings were reported but the limits of agreement were 12.1% and -15.5%. The Spearman correlation coefficient was 0.85. The limits of agreement was tighter below 20% normal forms (+8.4 and -6.6). Experiment 3 (objective 3) (repeatability): 255 cells were analyzed three times in succession. Estimating pairwise agreement, the kappa statistic for the pairs are 0.85, 0.80, and 0.85, respectively, which compares favorably with the second canonical moment of 0.8329 (kappa = 0.83).
The computer's ability to classify normal morphology per slide is promising. Below 20% normal forms, the limit of agreement is tight. Because of the 6% higher reading compared with the manual method, different thresholds possibly will be developed to identify subfertile from fertile patients. The computer gives excellent repeatability of normal and abnormal cells. Based on results obtained, this system can be of clinical value both in IVF units and andrology laboratories but more clinical data is required in this field.
依据严格标准，逐张玻片地对比不同观察者报告的正常形态百分比与一种计算机化方法（IVOS；汉密尔顿 - 索恩研究公司，马萨诸塞州贝弗利）：[1] 经验丰富的观察者（T.F.K.）与经验丰富的观察者（R.M.）对比，[2] 经验丰富的观察者（T.F.K.）与精子形态分析仪（IVOS）对比，以及[3] 正常和异常细胞的重复性（IVOS）。设计、地点、患者：从泰格堡体外受精项目的30名不同患者的玻片中随机选取。随机选择显微镜视野和精子细胞并记录正常形态百分比（目标1和2）。使用相同玻片，按所述方法逐细胞进行重复性检测（目标3）。
实验1（目标1）：T.F.K.和R.M.之间无显著偏差。一致性界限为8.6%和 -7.3%。标准差无显著差异（P = 0.1283）。读者之间的斯皮尔曼相关系数为0.83。实验2（目标2）：报告了相同的结果，但一致性界限为12.1%和 -15.5%。斯皮尔曼相关系数为0.85。正常形态低于20%时一致性界限更窄（+8.4和 -6.6）。实验3（目标3）（重复性）：对255个细胞连续分析三次。估计两两一致性时，各对的kappa统计量分别为0.85、0.80和0.85，与第二标准矩0.8329（kappa = 0.83）相比表现良好。
计算机逐张玻片分类正常形态的能力很有前景。正常形态低于20%时，一致性界限较窄。由于与手工方法相比读数高6%，可能需要制定不同阈值来区分不育和可育患者。计算机对正常和异常细胞具有出色的重复性。基于所得结果，该系统在体外受精单位和男科实验室均可能具有临床价值，但该领域还需要更多临床数据。
