Endometrial cell counts in recurrent miscarriage: a comparison of counting methods

Aims: Studies of uterine natural killer (uNK) cells require reliable measurements of uNK cell density among diverse endometrial tissue. The aim of this study was to compare cell counting manually with two computer-aided methods based on a public domain software package, ImageJ. Methods and results: Immunohistochemistry (IHC) of CD56(+) uNK cells was performed on endometrium from recurrent miscarriage patients. Numbers of stromal cells per high-power field (HPF) were counted by two observers using: (i) manual tally counter and graticule; (ii) ImageJ 'point picker' tool; and (iii) ImageJ 'particle analysis' tool. Coefficients of variation (CV) and Bland-Altman plots were used to evaluate interobserver differences. Evaluation of %uNK using ImageJ particle analysis for stromal cell counts and point picker tool for uNK counts was undertaken. Point picker and particle analysis were significantly better than manual counting [ interobserver CVs mean (standard deviation) 6.1% (3.3%); 4.7% (3.9%), 8.2% (6.5%), respectively]. Mean inter-and intra-observer CVs for %uNK were 10.3% (6.6%), 8.5% (4.9%) and 6.8% (4.3%), respectively. Bland-Altman analysis revealed no systematic differences in cell counts with the number of cells in the image for each method. Conclusions: Compared to manual cell counting, computer-aided image analysis yields more reproducible results for the assessment of uNK cells density using IHC.