Abstract
Purpose :
Central Retinal Thickness (CRT) is a measure derived from Optical Coherence Tomography (OCT) scans and is an important outcome in clinical trials. Differences in segmentation algorithms among various commercial OCT machines leads to variability in CRT in the same eye and results in data analysis issues. To circumvent this issue, we have developed a custom segmentation algorithm called EdgeSelect (ES), to standardize CRT for large scale comparative analysis. In this project, we compare CRT measurements from commercial algorithms with the ES algorithm.
Methods :
We analyzed CRT measurements from both commercial OCT algorithms and the ES algorithm among 100 OCT images of healthy eyes, 100 diabetic macular edema (DME) eyes and 100 neovascular age-related macular degeneration (neoAMD) eyes. Commercial software analyzed includes Cirrus HD-OCT, Carl Zeiss Meditec, Inc. and Spectralis HRA-OCT, Heidelberg Engineering, Inc. Each OCT image was analyzed using proprietary, commercial software for CRT, after correcting for potential artifacts. All OCTs were converted to DICOM format, for viewing in a common platform and segmented using the semi-automated EdgeSelect algorithm at the following layers – internal limiting membrane (ILM), top and middle of retinal pigment epithelium (RPE) and Bruch’s membrane (BM). Paired CRT data from commercial OCT machines was compared with an equivalent algorithm using ES, defined as ILM – middle of RPE for Cirrus and ILM to BM for Heidelberg. In addition, commercial CRT was also compared with standardized, ES measured CRT, which was defined as ILM – top of RPE as standardized ES measurements are used for large scale reading center studies.
Results :
The mean difference in CRT between the commercial OCT algorithm and equivalent EdgeSelect algorithm is almost negligible in normal, DME eyes and neoAMD eyes. The mean difference between standardized CRT and commercial CRT was the most in neovascular AMD eyes.
Conclusions :
CRT measurement differences between equivalent algorithms in ES and Heidelberg show minimal variation across healthy and diseased retinas suggesting conversion of images to DICOM format and segmentation with ES yields similar results to previously validated, commercial software. Differences in standardized CRT measurements by ES compared to commercial software then reflects anatomical differences in where retinal segmentation occurs.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.