Purchase this article with an account.
R.D. Fechtner, M. Fingeret, A.S. Khouri, M.J. Sinai; Diagnostic Accuracy of HRT 3 Moorfields Regression Analysis and Glaucoma Probability Score . Invest. Ophthalmol. Vis. Sci. 2006;47(13):3631.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To compare the diagnostic accuracy of the HRT 3 Moorfields Regression Analysis (MRA) and a new machine learning classifier, the Glaucoma Probability Score (GPS).
Normal and glaucoma patients were enrolled from two clinical sites, the NY VA in St. Alban’s NY and New Jersey Medical School, Newark, NJ. All subjects underwent a full eye examination, standard achromatic perimetry (SITA– standard 24–2) and confocal scanning laser ophthalmoscopy (Heidelberg Retinal Tomograph HRT 2 hardware running HRT 2 or HRT 3 software). Glaucoma was defined by the existence of reproducible glaucomatous visual field defect. Normal subjects had normal visual fields and IOP <24 mm Hg). All images were exported to the HRT3 software for analysis. The contour line was drawn on each image by a glaucoma specialist. GPS analysis does not require a contour line. Sensitivity and specificity values were determined for the MRA, the GPS, and two additional discriminant function analyses available in software (FSM and RB analyses). ROC analysis was performed on stereometric and GPS parameters but could not be performed on MRA because it exports classifiers and not continuous variables.
73 Normal eyes and 81 glaucoma eyes were enrolled. Sensitivity, specificity and area under ROC curve are reported in the table below.
The new machine learning classifier in the HRT 3, the GPS, performs similarly to the MRA. Both of these measures provided good sensitivity and specificity. GPS does not require the operator to draw a contour line, thus eliminating a potential source of error.
This PDF is available to Subscribers Only