651. The Location Consistency Index Helps to Distinguish Eyes With Subclinical Keratoconus From Normal Eyes.
- Author
-
Wang M, Shi C, Zhou Y, Ye Y, Fan X, Huang H, Yu X, Lu F, and Shen M
- Subjects
- Cornea, Corneal Pachymetry, Corneal Topography, Humans, ROC Curve, Retrospective Studies, Keratoconus diagnosis
- Abstract
Purpose: To develop a novel index that combines the locations and magnitudes of corneal alterations to improve discrimination of eyes with subclinical keratoconus from normal eyes., Methods: A Scheimpflug-based tomography system was used to image 252 eyes (normal: 78 eyes, subclinical keratoconus: 71 eyes, and keratoconus: 103 eyes) of 252 patients from two clinical centers. Coordinates and magnitudes of the maximum corneal protrusion alterations were extracted from curvature, elevation, and pachymetry maps. A location consistency index (LCI) was calculated from the Euclidean distances among these locations. A logistic regression model, named the location consistency enhanced score (LCES), which combined the LCI and the magnitudes of these maximum alterations, was trained and tested in two different datasets., Results: The LCI in eyes with subclinical keratoconus was 7.8 ± 2.6 µm, which was significantly different from that in normal eyes (11.8 ± 3.9 µm) and eyes with keratoconus (5.8 ± 2.4 µm) (all P < .001). The LCI could differentiate eyes with subclinical keratoconus from normal eyes with a sensitivity of 67.6%, specificity of 83.3%, and area under the receiver operating characteristic curve (AUC) of 0.81. Combining the magnitudes of these maximum alterations with the LCI for the LCES yielded a sensitivity of 90.0% and a specificity of 74.4% for differentiating eyes with subclinical keratoconus from normal eyes (AUC: 0.91)., Conclusions: The LCI can assist in differentiating eyes with subclinical keratoconus from normal eyes. The LCES is a potential new index to assist in a confirmatory test of eyes with subclinical keratoconus. [ J Refract Surg . 2022;38(1):35-42.] .
- Published
- 2022
- Full Text
- View/download PDF