Purchase this article with an account.
G. Michelson, R. Chrastek, M. Wolf, T. Hothorn, B. Lausen, R. Lämmer, C.Y. Mardin, H. Niemann; Automated Outlining of the Optic Nerve Head for Glaucoma Screening . Invest. Ophthalmol. Vis. Sci. 2003;44(13):3372.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose:To evaluate a method for automated outlining of the external margin of the optic nerve head (ONH) and automatic glaucoma classification. Methods:Automatic outlining of the ONH: Outlining was carried out using the reflectivity images of the scanning Laser ophthalmoscope HRT I (Heidelberg Engineering, Germany). As the optic disk is of elliptical shape, the possible ONH margin was limited to a family of controlled continuity splines, known as active contours. The final contour was found by balancing the internal and external forces. The automatic outlined ONH margin was imported into the HRT database. The morphological parameters were calculated by regular HRT software (v.2.01). Classification: Automatic classification was based on morphological parameters. Three different automatic classifiers were used: linear discriminant analysis (LDA), classification trees (CTREE) and improved classification trees by bootstrap aggregation (bagging). The bias corrected 0.632+ bootstrap estimator was used to estimate the misclassification error, i.e. the proportion of incorrect classified subjects in percent. Each estimator was computed using 50 bootstrap replications. Patients: We evaluated 159 eyes of a case-control study. The subjects in the control- and glaucoma-group were matched by age and sex. The control group contained 77 subjects, the glaucoma-group 82 patients. Results:In 83% of eyes the automatic outlined ONH margin showed good coincidence with the manually outlined ONH margin. The error rates of the three classifiers using the automatic outlined ONH margin and the manually outlined ONH margin were similar for LDA and CTREE. The error rate estimation (ERE), the sensitivity (SE) and specificity (SP) for the automatic outlined ONH margin were 27.7% (LDA ERE), 25.2% (CTREE ERE), 22.2% (Bagging ERE), 85% (SE), 58% (SP), and for the manually outlined ONH margin 26.8% (LDA ERE), 22.0% (CTREE ERE), 13.4% (Bagging ERE), 88% (SE), and 66% (SP). Conclusions:The automatic outlining of the ONH margin is comparable with the manual outlining. The automatic classification achieved 77.8% of estimated correct classification.
This PDF is available to Subscribers Only