Validation and use of simulated corneal endothelium specular microscopy images

Abstract

Abstract Purpose The aim of the current study was to develop simulation software capable of producing life like images of the corneal endothelium (CE) captured by clinical specular microscopy (CSM). Exact morphological data would come with each image. Once evaluated for authenticity, simulated images would be used for evaluation of the precision of methods of morphometry of corneal endothelium density (CED). Methods Software for simulation of images of the CE captured by CSM was developed. Several unique images of the CE were created, spanning a range of CED. 12 images, evenly spread over a CED range of 1959‐3838cells/mm2 were selected. The images were uploaded to a PC with border tracing software for CE morphometry. The semi‐automated morphometry software was used to analyze the CED of the 12 images. The CED results given by the analysis and the CED given from the creation of the images were analyzed by linear regression. Results An expected correlation between CED given from the creation of the simulated images and the CED given from the semi‐automated analysis was found. Assuming CED given by the model were exact, the standard deviation of the residual variance of the analyzed CED was estimated as a 95% confidence interval to 15.8‐91.0cells/mm2, corresponding to a variation coefficient of 0.5‐3.1%. Average CED of the 12 samples was 2899cells/mm2. Conclusion A simulator capable of realistic CE structures is a useful tool. Not only can it be used to estimate the accuracy of current methods of CE morphometry – it can also be used for development of new methods of morphometry, due to the virtually unlimited access to unique realistic images that come with exact and pre‐defined morphological data.