Comparison of methods for detecting keratoconus using videokeratography
N. Maeda, S. D. Klyce and M. K. Smolek
Lions Eye Research Laboratories, Louisiana State University, Medical Center School of Medicine, New Orleans, USA.
BACKGROUND: The detection of keratoconus patterns on videokeratography is
important for screening candidates for refractive surgery and for studying
the genetic basis of keratoconus. OBJECTIVE: We compared three quantitative
approaches to identifying keratoconus from videokeratographic information
to examine the limitations and capabilities of each test and to determine
their suitability for use in the clinical setting. METHODS:
Videokeratographs typical of clinically diagnosed keratoconus (n = 44) and
of various non-keratoconus conditions (n = 132, including normal,
with-the-rule astigmatism, contact lens-induced corneal warpage,
photorefractive keratectomy, keratoplasty, and pellucid marginal
degeneration) were selected. Three methods for detecting keratoconus were
used: keratometry (average Simulated Keratometry [SimK] readings > 45.7
diopters [D]); the modified Rabinowitz-McDonnell test (central corneal
power > 47.2 D and/or Inferosuperior Asymmetry [I-S] value > 1.4 D);
and an expert system classifier (classification based on discriminant
analysis and classification tree with eight topographic indexes).
Sensitivity and specificity were calculated for each test. RESULTS:
Sensitivities were 84% for keratometry, 96% for the modified
Rabinowitz-McDonnell test, and 98% for the expert system classifier.
Specificities for the three methods were 86%, 85%, and 99%, respectively.
In terms of sensitivity, the expert system classifier was significantly
better than keratometry (P = .04). In terms of specificity, the expert
system classifier was significantly better than either of the other methods
(P = .001). CONCLUSIONS: For screening candidates for refractive surgery,
where high sensitivity is needed, either the modified Rabinowitz-McDonnell
test or the expert system classifier is suitable. For diagnosing
keratoconus, where high specificity is more useful, the expert system
classifier is more appropriate than the other two methods.
