Automated keratoconus detection using the EyeSys videokeratoscope

Purpose: To evaluate the effectiveness of indices derived from the EyeSys S ystem 2000 in detecting keratoconic corneas. Setting: Department of Ophthalmology, Hopital Saint Antoine, Paris VI Unive rsity, Paris, Methods: Topographies of 208 corneas were evaluated. The corneas were from 8 groups of patients classified by the following diagnoses: normal, regular astigmatism, cataract, radial keratotmy, photorefractive keratectomy, myop ic keratomileusis, penetrating keratoplasty (PKP), and keratoconus. Nine st atistical indices derived from EyeSys data, 2 Holladay Diagnosis Summary in dices (coefficient of uniformity and coefficient of as phercity [Asph]), an d our refractive power symmetry index were studied. A training set of 104 c orneas was used to determine the most efficient threshold value of each ind ex based on sensitivity and specificity curves. Decision trees combining 2 indices were generated. Sensitivity and specificity were calculated in a va lidation set composed of the remaining 104 corneas. Results: Based on the results of the training set, the optimum indices were SDSD (standard deviation of the standard deviations of the radii of curvat ure of each ring) and Asph. In the validation set, the decision tree using these indices featured 88.5% sensitivity and 94.9% specificity the 4 false- positive cases were in corneas in the PKP group of patients. Conclusions: Clinically apparent keratoconus can be detected among normal c orneas and irregular corneal shape patterns using the EyeSys System 2000 da ta and a decision tree combining 2 indices. J Cataract Refract Surg 2000; 2 6:675- 683 (C) 2000 ASCRS and ESCRS.