Purchase this article with an account.
P. S. Kollbaum, J. Pepose, M. Qazi, A. Mahmoud, C. Roberts, M. Twa, M. Merchea; Detection of Corneal Irregularity With Automated Corneal Topography Indices. Invest. Ophthalmol. Vis. Sci. 2008;49(13):1030.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Keratoconus (KCN) and form fruste keratoconus (FFKCN) are contraindications to many refractive surgical procedures. It is imperative to identify and exclude these conditions preoperatively. Several curvature, elevation, and thickness measures available on the ORBSCAN have been previously described to screen for the presence of corneal abnormalities. The purpose of the current study is to evaluate these screening indices and develop an improved, more robust screening measure.
ORBSCAN exams from 209 KCN, 150 FFKC, and 174 normal eyes were classified by custom automatic classification software based on 52 pass/fail binary indices. This classification was compared to the clinical diagnosis of KCN, FFKCN, or normal. The clinical diagnosis of KCN was based on the presence of biomicroscopic signs. Fellow eyes of subjects with KCN were classified as FFKCN if they had no slit lamp signs of the disease. A univariate analysis was done to determine the area under the curve (AUC) for the 52 binary measures and 125 continuous variable measures. A split-sample cross-validation procedure was used to evaluate a multivariate logistic regression model developed to differentiate between KCN and normal eyes. The ability of this model to detect irregularity was then evaluated on FFKCN eyes.
Based on previously proposed indices comparing the entire data set of KCN and normal eyes, AUCs greater than 96.5% were achieved for 2 indices: sum of the 12 "ORBSCAN screening indices" greater than 6, and a posterior elevation (PE) greater than 40 microns above the best-fit sphere (BFS). The continuous measure of the maximum PE value alone had an AUC of 98.8%. A stepwise logistic regression model consisting of the maximum PE value and the mean pachymetry value of the central 5 mm achieved an AUC of 99.1%. The average of 100 split-sample validation sets yielded an AUC of 98.7%. The AUC for detecting FFKCN was 93.2%.
A single index evaluating the PE performed quite well in detecting corneal irregularity. Detection ability can be further improved by combining PE and pachymetry information in a logistic regression model.
This PDF is available to Subscribers Only