Automatic image processing applied to corneal endothelium cell count and shape characterization

Abstract

Corneal endothelium cell count, as well as cell hexagonality percent characterization, are of great importance nowadays to detect anomalies and pathologies of human eye, such as glaucoma. Prevalent technologies used are mainly based in both microscopy and a later image analysis. However, automatic cell count made by microscopes’ built-in software is rather inconsistent, therefore many laboratories opt for using manual count as the most reliable alternative. This count is a tedious and time-consuming task, that can lead to human error, for this reason, several proposals to automate the process have been made. Present communication shows a procedure for the automatic pre-processing, segmentation and analysis of the images obtained by a confocal microscope, using watershed transform, and the graphics user interface (GUI) created with Matlab® to apply this procedure. In order to quantify the procedure’s quality, 30 corneal endothelium images with a number of cells between 90 and 170 were analysed, resulting in a mean error in cell count of 4.3%, which can be considered a reasonably good result. However, results achieved for hexagonality percent using this method, and with the available image quality, are not as good as expected, which invites to improving image quality, focusing in areas with better cell homogeneity or even considering the application of other algorithms, such as neural networks, for future works.