Creation and validation of a classifier model for keratoconus detection using inter-ocular asymmetry based parameters

Session Details

Session Title: Corneal Biomechanics

Session Date/Time: Monday 09/10/2017 | 08:00-10:30

Paper Time: 08:06

Venue: Room 2.1

First Author: : G.Prakash UNITED ARAB EMIRATES

Co Author(s): :    V. Jhanji   R. Philip   R. Bacero   D. Srivastava           

Abstract Details

Purpose:

To explore if normative limits of inter-ocular asymmetry can be used to differentiate between normal and keratoconic eyes, and to look at screening benefits of including this parameter as an adjuvant to conventional cutoffs in a multiresponse, graded classifier model for keratoconus Vs normal eyes.

Setting:

Cornea and Refractive Surgery Services, NMC Eye Care, NMC Specialty Hospital, Abu Dhabi, United Arab Emirates

Methods:

In the first part of the study, medical records of 150 patients (100 bilateral normal, 50 bilateral keratoconus) were randomly selected and reviewed. Scheimpflug scans (Sirius, CSO, Italy) were evaluated. Corneal thinnest point, apex and central characteristics (curvature, corneal thickness (CT)), and 6-mm anterior corneal higher order aberrations (HOARMS) were evaluated. In the second part, the classifier model was built using statistical cutoffs of 1 and 2 standard deviations from the normal eyes. The sensitivity and specificity of inter-ocular difference and inter-ocular average was used to validate the classifier on 50 new patients (25 each keratoconus and normal).

Results:

As expected, normal eyes had lower mean curvature and HOARMS, and higher pachymetry. Both measures of inter-ocular symmetry: absolute difference and “absolute difference per unit” (absolute difference/ average of two eyes) were also significantly lower in normal eyes .The absolute difference values were: HOARMS 1.6±1.3µ vs. 0.14±0.3µ, Apex Keratometry 6.1±6.5D vs. 0.8±0.7D and Thinnest CT 38.9±42.1µ vs. 7.0±7.5µ for keratoconic and normal eyes, respectively. These three-variable based classifier was tested on 50 new cases and had sensitivity of 1.0 for both the average and difference classifiers, however specificity of 0.95 for average and of 0.29 for difference.

Conclusions:

An inter-ocular difference based classifier considering the absolute difference between apex keratometry, thin CT and 6mm corneal anterior HOARMS can be used as a highly sensitive screening tool differentiating between normal and keratoconic eyes. As these differences are independent of the mean value of these parameters, early cases of asymmetric keratoconus can be differentiated from asymmetric normal corneas by this model. The poorer specificity of the difference based classifiers is on expected lines because keratoconus can be symmetrical and advanced. However, it is less probable for normal eyes to be asymmetrical beyond statistical limits as shown in our study.

Financial Disclosure:

NONE