The Sampling Error From Specular Microscopy Examinations and Their Reliability Indexes

Abstract

surgery and the second after 23 months. The analysis tool built into the Topcon SP-3000P SM reported the ECD as 2282 cells per square millimeter after 12 months and 1800 cells per square millimeter after 23 months. However, no single cell changed position, shape, or size in the overlapping area (colorcoded stripe-wise comparator image not shown). We conclude that the ECD loss of 482 cells per square millimeter is an artifact as a result of sampling errors.Cell-by-cell comparisons may help in tackling the problem of insufficient sampling sizes when assessing endothelial stability. Our novel method uses all morphological information from the SM images rather than limiting analysis to the cell dotting, as is the case in ECD counting. We therefore suggest comparison of subsequent SM images cell by cell in clinical trials instead of focusing on ECD alone. For this reason, we have developed an image registration program for SM images (manuscript in preparation). FIGURE 2. Superimposition (average) of 2 SM images after homologous limbokeratoplasty. The first image was recorded 12 months and the second one 23 months after surgery.

Further Analysis of the Predictability of Corneal Endothelial Cell Density Estimates When Polymegethism Is Present

“…Overall, it was concluded that a different approach was needed to 'sample sufficient endothelial cells ….. to give a reasonably reliable production of the true mean' (in cell area or ECD). A similar perspective was also adopted in much later studies, [12][13][14] but in which it was claimed that the socalled 'relative error' (in ECD estimates) could routinely be overcome by increasing the number of cells measured. These later studies included analysis of small field images obtained from the Topcon SP-3000P microscope, but the specific basis of the error calculation (now set at 5 % rather than 10 %) was not provided, neither was any obvious analysis of the relationship of this sampling error (and its correction) to the extent of endothelial polymegethism.…”

“…3 With the recognition that the endothelium could change from being homogeneous to heterogeneous in appearance came consideration of how repeatable the estimates of ECD might be according to the number of cells measured, with different studies addressing slightly different aspects of this so-called reliability of cell morphometry. [10][11][12][13][14][15] This issue is distinctly different from assessments of errors in imaging software that can lead to very substantial uncertainty in ECD and especially COV estimates. 16 In consideration of endothelial image analysis for clinical trials, no specific recommendations have actually been made on the number of cells that should be measured, and an ideal scenario might be considered to be that to measure as many cells as possible.…”

“…20 From the opposite perspective, it has been implied that any 'error' in ECD estimates associated with increased COV can be simply overcome by measuring more cells from more than one image. 13,14 Overall, therefore, a case can be made that the potential importance of higher variability in cell areas (increased COV on cell area values) on ECD estimates needs further clarification.…”