Abstract
Abstract: :
Purpose: To evaluate the effectiveness, separately and in combination, of a meridional index and a circumferential (mire) index in the detection of keratoconus. Methods: 200 raw data records of Orbscan (Orbtek/Bausch & Lomb) anterior surface elevation maps were randomly selected from a clinical database. 90 had been classified clinically as manifestly keratoconic; the remainder were of mixed diagnosis. Missing image data were approximated by polynomial interpolation. The semi Meridional Steepness Ratio (sMSR) was calculated as the ratio of gradients of the inferior and superior hemi-meridians at 290º at a radial distance of 2.0mm from the apex. The Fourier Series Harmonic Ratio (FSHR) was determined as the ratio of the magnitudes of the first and second harmonic terms in the Fourier series representation of the 2.3mm circumferential mire. A descriminant model based on the optimal combination of the sMSR and the FSHR (as identified from the ROC (receiver-operator curve) was determined using an adaptive artificial neural network (AANN). All computer programs were written in MATLAB (Mathworks Inc.) Results: Sensitivities and specificities of 1.0 and 0.76 for the sMSR and 0.95 and 0.88 for the FSHR in the delineation of clinical keratoconus were obtained. In combination within the AANN, a nomographic axis was established which allowed the optimisation of sensitivity for any prescribed specificity. Conclusion: Low ordered (minimal parameter) indices of keratoconus achieving clinically valuable sensitivity and specificity can be derived from the anterior surface elevation topography alone.
Keywords: 450 keratoconus • 599 topography • 429 image processing