Keratoconus and Keratoconus Suspect Classification Modeling Using Scanning-Slit Videokeratography
Narrative Responses:
Purpose
To assess the efficacy of curvature, elevation, and thickness measures, as well as previously published algorithms, of slit-scanning videokeratography in order to differentiate populations with a clinical diagnosis of normal, suspect keratoconus, and keratoconus.
Methods
617 metrics derived from slit-scanning videokeratography (Orbscan, Technolas Perfect Vision) were calculated for a data set that included sub-groups classified as diseased if they had characteristic keratoconic slit-lamp findings in the study (KCN, n=338) eye or fellow (KCF, n=74) eye only. Keratoconus suspects (KCS, n=78) had irregular videokeratopography, but no keratoconic slit-lamp findings in either eye. Normal eyes (NRM, n=114) were sampled from refractive surgery patients with no evidence of ectasia upon follow-up. For each measure, a non-parametric estimate of the area under the Receiver Operator Curve (AUC) was calculated, from which a discriminant model for ectasia diagnosis was developed.
Results
Anterior curvature, anterior elevation, and posterior elevation metrics had higher AUCs (≥0.97) for detecting keratoconic disease (KCN), relative to pachymetry-derived metrics such as corneal volume or spatial distribution (≥0.92). Although single item metrics performed quite well, combining metrics improved the ability to identify and differentiate sub-groups, specifically KCF (AUC ≥ 0.99) and KCS (AUC ≥ 0.82).
Conclusion
Evaluation of keratoconus and suspect corneas highlights the importance of both anterior and posterior videokeratographic analysis in the preoperative assessment of keratorefractive surgery candidates.