Discriminant function and regression analysis in Heidelberg Retina Tomograph (HRT) and HRT II
A. F. Scheuerle, R. O. W. Burk, T. Bartke, H. E. Voelcker
Objective: Evaluate three types of automated optic disc classification ("glaucoma" vs "normal"), all based on 3-D topometric parameters of HRT and HRT II: reference plane dependent discriminant function (DF1), reference plane independent discriminant function (DF2) to predict the presence of glaucomatous visual field defects (GVD) and Moorfields regression analysis (RA).
Methods: We compared 10o images (HRT, 2562 pixels) and 15o images (HRT II, 3842 pixels) obtained from 20 eyes of 20 glaucomatous patients and 15 eyes of 15 normal controls as defined by the result of static w/w 6o threshold perimetry. Differences between classification rates within the diagnostic groups were considered significant at an alpha-level of p<0,05.
Results: RA showed a single "abnormal" segment in 2 of 15 control eyes in HRT, in HRT II 15/15 control eyes were considered "normal". In the GVD group RA classified 18/20 as "glaucoma" in HRT and 17/20 in HRT II. In contrast to the global classification rate, RA showed less damaged segments for 8 eyes (6 glaucoma, 2 normal) in HRT II compared to HRT. Vice versa 2 glaucomatous eyes showed more damaged segments in HRT II compared to HRT. DF1 classified 17/20 as glaucoma in both HRT and HRT II, DF2 17/20 in HRT and 15/20 in HRT II. DF2 classification rate was significantly lower for HRT II. The difference of discriminant function (HRT HRT II) was 0,79 ± 0,39 (mean ± SD) for DF1and 0,41 ± 0,12 for DF2.
Conclusion: Results indicate that the automated classification procedures DF1, DF2 and RA, implemented in the HRT software, may be applied for the interpretation of HRT II data. While the averaged DF2 value (prediction of GVD irrespective of optic disk appearance) is constant, it leads to a decreased sensitivity with respect to the classification rate in HRT II data set as compared to HRT.
Eye clinic University of Heidelberg, INF 400, D-69120 Heidelberg