Computer-assisted interpretation of visual fields in glaucoma.

Visual field abnormality is an important diagnostic sign in glaucoma. Therefore, the presence or absence of visual field loss most often strongly influences diagnostic and therapeutic decisions in glaucoma management. Interpretation of visual field results is often difficult, however. Physiological variability of perimetric sensitivity values contributes to these difficulties. It has been our aim to develop improved computer-assisted methods for the recognition of early glaucomatous field loss. Our approach has therefore been to design techniques that are highly sensitive to small but significant departures from normality. We have investigated normal physiological variability in perimetric results and combined the obtained knowledge with pathophysiological models which are sensitive to the spatial patterns of field loss commonly seen in glaucoma. Thus, we have devised probability scores in order to take the complex physiological variability into account, and developed a hemifield analysis and an arcuate cluster analysis based on the normal anatomy of the retinal nerve fibre layer. A fundamental approach in the collection of normative data and the selection of glaucoma cases used in this project has been to select subjects using non-perimetric criteria (except for the removal of large field defects). Our objective here was to reduce bias from pre-conceived ideas of visual fields. This approach was used for (1) empirical studies on physiological variability, (2) development of analysis methods, and (3) evaluation of such methods. Glaucoma patients were selected based on evaluations of optic disc appearance. Normal subjects were never eliminated on the basis of perimetric results alone. The new methods developed in these studies have significantly improved discrimination between normal and glaucomatous field results, as compared with previously available techniques. Our results indicated that the usage of probability scores was the main source of this improvement, and that the location of observed field abnormalities and spatial modelling were other important factors. Candidate methods which did not properly combine spatial and normative analyses resulted in false positive defects in the mid-periphery and/or underestimated paracentral glaucomatous field defects. Similar approaches based on classification of visual field results in terms of significances, and on recognition of specific spatial patterns of field loss could be used for other groups of diseases having visual field abnormality as an important diagnostic sign.(ABSTRACT TRUNCATED AT 400 WORDS)