Purchase this article with an account.
SK Gardiner, DP Crabb, FW Fitzke, RA Hitchings; Developing a Physiologically Accurate Spatial Filter for Visual Field Data . Invest. Ophthalmol. Vis. Sci. 2002;43(13):3906.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose: To develop a spatial filter for reducing noise in perimetric visual field data, which takes into account the physiological structure of the eye. Methods: From a uniquely large retrospective database of 96000 visual fields, obtained from Humphrey perimetry, the covariances between sensitivities at different points in the eye were obtained. Algorithms for predicting the sensitivity of a point based on those elsewhere in the field were obtained by a series of regressions on these covariances, taking into account second-order interactions between points. The filtered value of the point is a linear combination of this predicted value and the raw value, where the predicted value is given more weight at points where the predictions are more accurate (based on the correlation between predicted and raw values at each point, and the number of other points used in the prediction). The effect of the filter on the noise was tested by examining series of visual fields and measuring the effect on the R-Squared values of pointwise linear regression slopes. Results: The locations of the points used to derive the predicted value at each point were found to conform to the known structure of the retinal nerve fibre layer (unlike the fixed-window approach of the Gaussian filter). R-Squared values were generally higher for the filtered data; the number of points with R-Squared≷0.5 over 303 series of length 20 doubled when the filter was applied. This indicates that the filtered data fit more closely to a straight line over time during long series, thus providing a better estimate of the true behaviour of the eye. Conclusion: The filter developed here significantly reduces the amount of noise present in series of visual field data, without the loss of discriminatory power ('blurring') associated with a simple Gaussian filter.
This PDF is available to Subscribers Only