Validation of an Artificial Intelligence-based Tool - The Screening Corneal Objective Risk of Ectasia Integrated into Anterion for Detection of Corneal Ectasia/Risk of Ectasia.

Abstract

Purpose: The purpose of this study was to validate the artificial intelligence-based Screening Corneal Objective Risk of Ectasia (SCORE) for the detection of corneal ectasia/risk of ectasia and to find the mean SCORE value in normal eyes.

Methods: This prospective observational study included 320 eyes of 160 patients undergoing screening for refractive surgery and keratoconus. All were subjected to Anterion (Heidelberg, Germany) corneal tomography. The maps were read by trained cornea specialists masked with respect to system-generated SCORE. The SCORE values were retrieved by another masked observer. The SCORE was labeled true positive if the eye was tomographically positive for ectasia/ risk of ectasia with or without clinical evidence of ectasia (i.e., clinician's diagnosis as ectasia/ risk of ectasia) and SCORE was >/=0. The SCORE test was labeled as true negative if all the maps were normal and SCORE was <0. A false-positive test was where the eye was normal but SCORE was ≥0. A false-negative test was one tomographically there was definite evidence of ectasia/ risk; with or without clinical evidence of ectasia (i.e.,clinician's diagnosis as ectasia/ at risk of ectasia) but SCORE was <0. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated.

Results: The mean SCORE in normal eyes was -0.7 ± 0.6. Sensitivity was 83.3% and specificity was 80%. PPV and NPV were 46% and 96% respectively. In the keratoconus subgroup, specificity and PPV were 100%; sensitivity was 79% and NPV was 76.3%.

Conclusion: SCORE in Anterion has moderate sensitivity and specificity for detection of ectasia/risk of ectasia. It performs best in the detection of corneal ectasia. It needs to be interpreted with caution.