Systematic detection of keratoconus in OCT: corneal and epithelial thickness maps

• Published in: Journal of cataract and refractive surgery Vol. 48; no. 12; pp. 1360 - 1365
• Main Authors: Yücekul, Burcu, Dick, H Burkhard, Taneri, Suphi
• Format: Journal Article
• Language: English
• Published: United States 01-12-2022
• Subjects: Cornea; Corneal Pachymetry - methods; Corneal Topography; Humans; Keratoconus - diagnosis; Retrospective Studies; ROC Curve; Tomography, Optical Coherence - methods
• ISSN: 0886-3350; 1873-4502
• DOI: 10.1097/j.jcrs.0000000000000990

To detect keratoconus (KC) only by analyzing the corneal and epithelial map parameters and patterns in optical coherence tomography (OCT).
Tertiary care refractive surgery center.
Retrospective data collection.
Corneal and epithelial thickness maps of normal, manifest, and subclinical keratoconic eyes (according to the Belin-Ambrosio display, Pentacam) were evaluated using spectral-domain OCT (Zeiss Cirrus 5000 HD). A new 2-step decision tree was developed based on previous studies with another OCT device. In the first step, if at least 1 of the 4 independent parameters (pachymetry minimum, pachymetry minimum-median, pachymetry superonasal-inferotemporal, and epithelial superonasal-inferotemporal) overruns the cutoff values, the eye was suspicious for KC. In the second step, if the epithelial map showed concentric thinning and the thinnest point of the cornea and epithelium is coincident, the eye was classified as keratoconic.
172 manifest keratoconic eyes (108 patients), 21 subclinical keratoconic eyes (20 patients), and 172 normal eyes (90 age-matched participants) were included in this study. Step 1 captured 100% of manifest and subclinical keratoconic eyes. Step 2 ruled out all suspicious but normal cases and, falsely, 2 subclinical keratoconic eyes. Our 2-step decision tree reached 100% specificity, 100% sensitivity in manifest KC, and 90.4% sensitivity in subclinical KC.
Pachymetric and epithelial map parameters and patterns in OCT can be used in the diagnosis of KC, including subclinical cases, yielding a high level of agreement with the commonly used diagnostic reference, the Belin-Ambrosio display. Further improvements by refining our algorithm and including an automated evaluation in the software are desirable.