Endothelial cell recognition capability of central and paracentral corneal areas using automatic specular microscopy counting

Abstract

Purpose : Automatic counting of endothelial cells (EC) is subject to errors due to the acquired image quality and to the accuracy of the software, which can distort the specular microscopy (SM) results. This study aims to compare the ability to recognize endothelial cells using automatic counting of a non-contact (NC) SM between the central area (CA) and paracentral areas (PCA) from the endothelial mosaic.

Methods : Thirty-six right eyes of 36 patients without any ocular abnormalities or previous surgeries were evaluated using NC SM CEM-350 (NIDEK©). Nine endothelial images were obtained in each examination: 1 image in CA (Group I) and 8 in PCA (Group II). The SM software automatically counted the cells in each image. The ability of the software to recognize endothelial cells was analyzed by comparing the following outcome measurements: 1 – number (N) of non-counted EC groups and total number of EC in these groups; 2 – Erroneously counted EC: Split EC (one cell counted as two or more) and the average amount of EC created by this division, N of cell clusters (multiple cells counted as one) and the average amount of EC per cluster (Figure 1); 3 – Non-evaluated area of the image, calculated as 1-[counted cells x average cell area (µm²)/image size (µm²)]. We also compared the results of the SM parameters: counted cells (CC), endothelial cell density (ECD), average cell area (AVG), coefficient of variation (CV) and hexagonality percentage (HEX). Descriptive statistics and two-tailed Student’s t test was used for statistical analysis (p<0.05). The same examiner performed all examinations.

Results : Results from EC recognition capability analysis (N of non-counted EC groups, total number of EC in these groups and erroneously counted EC) and SM parameters (CC, ECD, ACA, CV and HEX) as well as the comparison between groups I and II are presented in table 1.

Conclusions : EC automatic counting presented inaccuracies in both groups, although they were more significant in group I. This study suggests manual cell counting or individualized assessment with correction of errors to minimize distortions in the results of specular microscopy.

This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.

 

Three types of cell recognition errors. A=Non-counted cells; B=Cell cluster; C=Split cell; D=image C before automatic counting.