Purchase this article with an account.
Cindy Cai, Sina Vahedi, Inna Stroh, Gelareh Dastjerd, Hajra Channa, Mehraz Malakooti, Allen O Eghrari; Assessment of a standardized protocol for corneal endothelial cell analysis and interrater reliability across educational levels using the flex-center method. Invest. Ophthalmol. Vis. Sci. 2016;57(12):1938.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The flex-center method is a semi-automated process of measuring corneal endothelial cell density and morphology from images obtained from specular microscopy. It is unknown whether educational level impacts measurement reliability. We sought to assess the reliability of a standardized training protocol for the flex-center method of analysis across a spectrum of graders with different educational backgrounds.
Images from 20 random healthy individuals were collected using the Konan Specular Microscope (NSP-9900, Irvine, CA, USA). We developed a manual to train five graders on the flex-center method. Konan Cellchek software was used to extract measures including cell density (CD, cells/mm2), average cell area (CA, mm2), coefficient of variation (CV), and percent of hexagonal cells (HEX). We calculated the intraclass correlation (ICC) based on a two-way mixed effect model to assess inter-grader reliability of two sets of images: one set composed of the clearer image of a participant’s two eyes, and a second set of the more challenging image, with varying degrees of opacity and blur. We utilized test-retest data to assess intra-grader reliability.
Five reviewers, ranging from completion of secondary school to doctorate degrees, evaluated images from 40 eyes. Among the eyes with the clearer image, the average number of cells counted was high across all graders (mean cells per photo: range 100 to 146.1). Inter-grader reliability as determined by the ICC for CD (0.92, 95% CI: 0.87-0.97) and CA (0.95, 0.91-0.98) was high. The ICC was less robust for CV (0.46, 0.26-0.69) and HEX (0.71, 0.55-0.85). Inter-grader reliability also remained high for CD (0.91, 0.84-0.96) and CA (0.90, 0.82-0.95) in the set of more challenging images. Intra-grader reliability upon test-retest again was high for CD (0.97, 0.93-0.99) and CA (0.97, 0.93-0.99) but less so for CV (0.66, 0.32-0.85) and HEX (0.86, 0.69-0.94).
This standardized training protocol produces higher inter-grader reliability measures for determination of CD and CA than previously reported across a variety of educational backgrounds, and remains strong in a set of challenging images. These findings provide support for future large-scale analyses that require a diverse set of reviewers.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.
This PDF is available to Subscribers Only