Endothelial cell density (ECD) assessment with the noncontact Topcon SP2000P specular microscope is known to be reliable when the automated mode is followed by manual corrections (touched‐up mode). We compared its agreement with Rhine‐Tec, a new noncontact specular microscope, in 270 eyes of 160 patients, by comparing the ECD measured in the automated and touched‐up modes with that of Topcon touched‐up. Good agreement existed between either touched‐up modes with a mean difference of only 2 cells/mm2 95% CI (−27; 23) whereas agreement with the Rhine‐Tec automated mode was poor with an overestimation by a mean of 226 cells/mm2 95% CI (172; 281).
Background
The Topcon SP2000P non‐contact specular microscope (Topcon, Tokyo, Japan) is widely used to measure corneal endothelial cell density (ECD). It uses a cell contour recognition algorithm based on contrast differences, and ECD derived from an automated delineation of cell boundaries with manual correction of inaccurately drawn cells (“touched‐up” mode) has been validated.1,2 A new commercially available non‐contact specular microscope Rhine‐Tec (Rhine‐Tec, Krefeld, Germany) determines ECD by a cell‐centre method consisting of identification of centres of contiguous cells either in an automated or a touched‐up mode using an operator‐defined cell centre. We aimed to determine its agreement with the Topcon touched‐up mode.
Methods
The central endothelium of 270 eyes of 160 subjects (healthy subjects and those operated on for cataract or penetrating keratoplasty to explore a wide range of ECDs) was imaged, first with the Topcon microscope and immediately afterwards with the Rhine‐Tec specular microscope. On each sharpest image, the ECD and percentage of hexagonal cells were determined on the largest possible area by a trained ophthalmologist blinded to the clinical antecedents and choosing the device (Topcon or Rhine‐Tec) randomly. Both automated and touched‐up modes were used for Rhine‐Tec, and the Topcon touched‐up served as the reference.1,2 Agreement was determined using the Bland–Altman method.3 Means were compared using paired t tests (or non‐parametric Wilcoxon's test when required) and results considered significant when p<0.05.
Results
Mean (standard deviation (SD)) number of cells counted was 129 (46) (range 24–219, median 144) for Topcon and 137 (44) (range 47–279, median 136) for Rhine‐Tec (p<0.001). Despite a good correlation between the Topcon touched‐up and Rhine‐Tec automated modes (y = 0.37x+1497, r = 0.8, p<0.001), the agreement was poor, with Rhine‐Tec overestimating ECD by a mean of 226 cells/mm2 95% CI (172 to 281) (fig 1). A continuous bias in ECD determination characterised by overestimation of low ECDs and underestimation of high ones (p<0.001) was observed with the Rhine‐Tec automated mode, with maximal effects of this drift visible for low (<1800) and high (>2500 cells/mm2) ECDs. Correlation between the either touched‐up modes was good (y = 0.82x+377, r = 0.93 p<0.001) and agreement was far better with a mean difference of 2 cells/mm2 95% CI (−27 to 23) and narrower bounds of agreements (fig 2). Although the tendency to overestimate low and underestimate high ECDs persisted, it was significantly lower. Mean (SD) percentage hexagonality (calculated in touched‐up modes) was 58 (12) (range 14–100, median 59) and 41 (8) range 14–69, median 41) for Topcon and Rhine‐Tec, respectively (p<0.001).
Figure 1 Difference against mean graph showing the agreement between the endothelial cell densities (ECDs) obtained with the Topcon touched‐up mode compared with the Rhine‐Tec automated mode. ECD measured in Topcon touched‐up mode ranged from 417 to 3263 cells/mm2, with 84 eyes (31%) having <1800 (low ECD), 109 eyes (40%) having between 1800 and 2500 cells/mm2 and 77 eyes (29%) having >2500 cells/mm2 (high ECD). The solid horizontal line indicated the mean difference between Rhine‐Tec and Topcon (226 cells/mm2). The agreement was poor with wide 95% limits of agreement (calculated as mean difference (±1.96 SD of the difference) of the difference) set between −663 and +1116 cells/mm2 (dashed lines). The linear regression line (bold) indicated the significant overestimation of low ECDs and underestimation of high ECDs by Rhine‐Tec.
Figure 2 Difference against mean graph showing the agreement between the endothelial cell densities (ECDs) obtained with the Topcon and Rhine‐Tec touched‐up modes. The solid horizontal line indicated the mean difference between Rhine‐Tec and Topcon (2 cells/mm2). The agreement was considered acceptable, with 95% limits of agreement set between −412 and +408 cells/mm2 (dashed lines). The feeble slope of the linear regression line (bold) showed that the drift in counting (low to high) according to the ECD was reduced compared with figure 1.
Comments
Both non‐contact specular microscopes assess ECD on small cell samples. As observed with Topcon,1,2,4 the Rhine‐Tec automated mode is also not reliable. However, after manual corrections, Rhine‐Tec shows an acceptable agreement with Topcon. Regarding the two bounds of agreement on this sample size, a consistent individual variation remains. The reliability of hexagonality assessment using Topcon was already dealt with.1 The difference observed with Rhine‐Tec remains unexplained and is probably linked to a different image analysis algorithm (proprietary data not available). Information regarding the coefficient of variation of the cell area would be helpful to further evaluate this device.
Footnotes
Competing interest: None.
Presented in part at the 18th annual meeting of the European Eye Bank Association held on 20–21 January 2006 in Venice, Italy, and 112th annual meeting of the French Society of Ophthalmology held on 6–10 May 2006 in Paris, France.
References
- 1.Cheung S W, Cho P. Endothelial cells analysis with the Topcon specular microscope SP‐2000P and IMAGEnet system. Curr Eye Res 200021788–798. [DOI] [PubMed] [Google Scholar]
- 2.van Schaick W, van Dooren B T, Mulder P G.et al Validity of endothelial cell analysis methods and recommendations for calibration in Topcon SP‐2000P specular microscopy. Cornea 200524538–544. [DOI] [PubMed] [Google Scholar]
- 3.Bland J M, Altman D G. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 19861307–310. [PubMed] [Google Scholar]
- 4.Vecchi M, Braccio L, Orsoni J G. The Topcon SP 1000 and IMAGEnet systems. A comparison of four methods for evaluating corneal endothelial cell density. Cornea 199615271–277. [DOI] [PubMed] [Google Scholar]