Purchase this article with an account.
P.H. Artes, B.C. Chauhan, D.P. Crabb; The Effect of Spatial Filtering on the Analysis of Glaucomatous Visual Field Loss . Invest. Ophthalmol. Vis. Sci. 2005;46(13):3732.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose: To investigate the effect of spatial filtering (Gardiner et al., 2004) on visual field variability and systematic error (bias), and on the detectability of visual field loss in total deviation probability maps. Methods:Visual field series (20 tests in 6–monthly intervals, 24–2 full threshold) from 84 eyes of 50 glaucoma patients (mean MD –4.5, range +0.1 to –17.0 dB) were processed with a new spatial filter which takes into account the physiological patterns of interdependence between neighboring test locations, as well as the close relationship between sensitivity and visual field variability. Since the true visual field is unknown, the running median of 5 consecutive tests (best available estimate, BAE) was used as a standard for comparison at each point in time. Pointwise differences between filtered / unfiltered visual fields and the corresponding BAEs were analyzed in terms of variability (SD of differences) and bias (mean difference), stratified for sensitivity. Variability and bias were then compared between filtered and unfiltered series. Total deviation (TD) probability maps were established with reference to filtered and unfiltered visual field data of 114 eyes of 114 normals (mean MD –1.1 dB, range +2.0 to –5.2 dB) and compared between filtered, unfiltered, and BAE versions of one randomly selected test of each glaucoma eye. Results:Filtered visual fields were considerably less variable, particularly at sensitivities <20 dB (SD 4.1 dB at 10 dB, cf 6.0 dB in unfiltered fields, p<0.001). This gain in precision was achieved at the expense of increased bias – the filtered visual fields tended to underestimate visual field loss (mean differences ≈2.5 dB at sensivities near 0 dB). However, spatial filtering of the healthy control fields led to a slight shift (mean, +0.9 dB) of the normal limits, and most TD maps showed a small though significant increase in the size (+15%, p<0.01) and significance (p<0.01) of the defects after filtering. Conclusions:Spatial filtering provided a considerable reduction of visual field variability which is likely to enhance the detectability of change. The concurrent slight underestimation of visual field loss did not appear to impair the detectability of visual field loss in total deviation probability maps. We are currently exploring the benefits of spatial filtering for the earlier detection of visual field progression.
This PDF is available to Subscribers Only