Purpose: This prospective study was designed to test the reproducibility of a new automated technique for analyzing the corneal endothelium and to assess the validity of the technique by comparing it with a standard method.
Subjects and methods: We used a contact specular microscope combined with a video camera (Tomey EM-1000) and a computer (IBM compatible PC, 486DX33) with suitable software (Tomey EM-1100, version 0.94). Video images of the corneal endothelium (area: 0.312 mm2) were passed directly into the computer input by means of a frame grabber and were automatically processed. The area to be analyzed could be varied by location and size (5580-135,150 microns2), depending on the quality of the image. Healthy corneas of 67 volunteers (age: 30.9 +/- 8.6 years) were examined. One examiner measured cell density three times in each of 42 eyes (retest-stability); three different examiners made one measurement in each of 25 eyes (objectivity). We evaluated the cell density determined by the computer after automated analysis and assessed the corrected cell density. This second result was obtained after the examiner had corrected the processed image by drawing in cell boundaries that the computer had not recognized or erasing cell boundaries the computer had sketched in by mistake. Additionally, a photograph of the corneal endothelium (specular microscope Bio Optics LSM 2000 A) was obtained from 40 volunteers to be used for manual cell counting applying a "fixed-frame" technique (validity).
Results: The corrected values showed a high retest-stability (reliability coefficient r = 0.943) and a high objectivity (r = 0.904). The values obtained by the automated method (2415 +/- 214 cells/mm2) did not differ significantly from those obtained by manual cell counting (2431 +/- 228 cells/mm2) (P = 0.898). The uncorrected values (2252 +/- 190 cells/mm2) were on average 7.2 +/- 2.6% lower than the corrected ones (177 +/- 69 cells/mm2). Retest-stability (r = 0.856) and objectivity (r = 0.737) of the uncorrected values were satisfactory. The uncorrected value was significantly lower than the value of manual cell counting (P < 0.001). The size of the analyzed area (range 12,750-84,708 microns2; average 31,438 +/- 10,655 microns2) had no significant effect on cell density (Spearman's correlation coefficient k = -0.150, P = 0.093).
Conclusion: The automated method for analyzing the corneal endothelium quickly produces valid, reproducible results in normal corneas, provided that the correction mode of the software is applied.