Automated Detection of the Stromal Demarcation Line Using Optical Coherence Tomography in Keratoconus Eyes After Corneal Cross-linking

Abstract

Purpose: To evaluate the role of a novel automated detection software as compared to human operators in assessing the presence and depth of stromal demarcation line on optical coherence tomography (OCT) in keratoconus eyes post cross-linking. Design: Reliability analysis study. Methods: Two independent operators and an automated detection software examined corneal OCTs of 25 eyes of 25 patients post corneal cross-linking using the Dresden protocol, at 3 months postoperatively. Operators evaluated the presence of the demarcation line and measured its depth by looking at OCT images (128 cuts) on 2 separate occasions 1 week apart. The automated software examined all 128 cuts of each OCT measurement. Results: The mean corneal demarcation line depth was 321.54 ± 47.71, 322.86 ± 45.77, and 309.21 ± 40.98 μm, as computed by the automated detection software and the human operators, respectively. The intraclass correlation coefficients (ICC) between the automated detection software and Observers #1 and #2 were 0.884 and 0.847, respectively (P <.001). The ICC for interoperator reproducibility was 0.890, and for intraoperator repeatability for Operator #1 and Observer #2 were 0.922 and 0.925, respectively. The ICC for intersoftware repeatability was 1. Bland-Altman plots showed a good agreement between both observers and the software, with adequate 95% limits of agreement. Conclusion: Detection of the demarcation line by human operators is repeatable and reproducible, but it can be further optimized and standardized by an ultrafast and accurate automated software detection tool, providing a reliable indicator for treatment success. © 2018 Elsevier Inc.