Purchase this article with an account.
N. O'Leary, D. F. Garway-Heath, D. P. Crabb; An Alternative Reference Standard for Structural Progression in Glaucoma: An in silico Model of Scanning Laser Tomography Image Series. Invest. Ophthalmol. Vis. Sci. 2010;51(13):4010.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
There is no established gold standard to measure structural progression of glaucomatous neuropathy in longitudinal image series from patients. It is, therefore, difficult to validate software progression algorithms. The purpose of the study was to develop an in silico model to simulate realistic variability in stable series and compare measurement variability from simulated series against that measured in real short-term time series. Stable series may be used to test progression algorithm specificity.
The 3 confocal image stacks from each of 5 sets of Heidelberg Retina Tomograph (HRT) II scans, obtained within a 6 week follow-up period, of 127 eyes from 66 patients were selected. For each eye, a simulated series was propagated from a baseline confocal image stack. Fixational eye movements were modelled to generate within scan variability. Photon counting and electronic measurement noise were estimated to simulate variability from the device itself. The simulated confocal stacks were imported into the HRT software to generate topography images. Comparison between real and simulated data was made using the mean of pixel height standard deviation (MPHSD), the image cross-correlation (CC) of pixel-wise variability and neuroretinal rim area (RA) variability.
The 95% limits of agreement of MPHSD between real and simulated data were 0.8 to 6.1µm within scans and -5.3 to 8.7µm between scans. The mean of CC between real and simulated spatial variability maps was 0.50 (0.12 SD) within scans and 0.54 (0.11 SD) between scans. RA coefficients of variation: 95% limits of agreement between real and simulated data were -0.19 to 0.15. Variability about anatomical features was well reproduced.
Simulation can realistically reproduce variability inherent in stable patient HRT data. Stability in clinical series is uncertain, whereas in the modelled series it is certain. Thus this method can produce benchmark datasets on which various statistical methods for detecting HRT progression can be compared.
This PDF is available to Subscribers Only