City Research Online

Analysis of natural eye movements to assess visual field loss in glaucoma

Asfaw, D. S. (2020). Analysis of natural eye movements to assess visual field loss in glaucoma. (Unpublished Doctoral thesis, City, University of London)

Abstract

Diagnosis and monitoring of glaucoma, an age-related eye conditions that can cause irreversible loss of vision, relies on assessment of the visual field (VF). In this thesis, I develop novel methods of detecting visual field loss from natural eye movements when watching videos or looking at pictures. I present data from a literature review and three empirical studies.

In the literature review, I identified and examined 26 papers that investigated eye movements of glaucoma patients while performing tasks such as reading, driving, and visual search. The review indicated eye movements are altered by glaucomatous visual field (VF) loss but identified inconsistency in how these alternations manifest between studies.

The first study investigated empirically whether glaucoma produces measurable changes in eye movements. Fifteen glaucoma patients with asymmetric vision loss viewed 120 images of natural scenes monocularly; once each with the better and worse eye. Eye movements were recorded using a remote eye tracker, and key eye-movement paramet-ers were computed and compared between eyes (better eye versus worse eye). These parameters included conventional metrics (saccade amplitude [SA], fixation counts and duration, and bivariate contour ellipse areas [BCEA]), as well as a novel metric I designed to measure saccadic sequences: the saccadic reversal rate (SRR). In the worse eye, SA and BCEA were smaller (p < 0.05), while SRR was greater (p < 0.05). There was also a significant correlation between the between eye difference in BCEA, and differences in mean deviation (MD; a measure of VF loss severity) values (p = 0.01), while differences in SRR were associated with differences in visual acuity (p = 0.01). Furthermore, between-eye differences in BCEA were a significant predictor of between eye differences in MD: for every 1 dB difference in MD, BCEA reduced by 6.2% (95% confidence interval [CI]: 1.6–10.3%).

The second study investigated whether changes in eye movements due to glaucoma are large enough to be clinically useful. I developed a gaze-contingent simulated VF loss paradigm, in which participants experienced a variable magnitude of simulated VF loss, based on a real glaucoma patient. Fifty-five people with healthy vision watched two short videos and three pictures, with either: no VF loss; moderate VF loss; or advanced VF loss. Eye movements were recorded using a remote eye tracker, and key eye movement parameters were computed (SA; spread of saccade endpoints as quantified using BCEA; location of saccade landing positions, and the similarity of fixations locations among participants as quantified using kernel density estimation). There were statistically significant differences between conditions, but these measures—alone or in combination—were not capable of identifying VF loss with sufficient diagnostic precision compatible with assumed clinical utility, when considered against a reference standard for measuring the VF (automated perimetry). I do, however, suggest ways in which performance could be improved.

The third study had two parts. First, I curated a dataset of eye movements of 46 patients with glaucoma and 32 controls, which I made available online for other researchers. Second, I presented a novel spatiotemporal saccadic movement analysis using a machine learning method, which I validated on the dataset. My novel method involved translating individual saccades into one of N values, based on its size and direction, and then using the relative presence of different permutations of saccadic sequences to classify individuals as patients or controls. This ‘n-gram’ approach has been successfully applied previously in other technical domains, such as automatic speech recognition. I evaluated the sensitivity and specificity of my method using tenfold cross-validation. Areas under the ROC curve (AUC) for the novel approach was 0.78 (95% CI 0.75–0.81), compared to 0.63 using simple eye movement summary statistics (e.g., SA), and AUC = 0.70 (95% CI: 0.66–0.74) using a published current reference standard.

Overall, results from this thesis provide more evidence for eye movements being disrupted by VF loss, that these changes are related to changes in clinical measures, and that it is possible to extract and process these measures using some novel methods automatically. In future, assessment of natural eye movements could be analysed to help detect glaucomatous VF loss, and in the Discussion, I outline what steps are required to reach this goal.

Publication Type: Thesis (Doctoral)
Subjects: R Medicine > RE Ophthalmology
Departments: Doctoral Theses
Doctoral Theses > School of Health Sciences Doctoral Theses
School of Health Sciences > Optometry & Visual Science
Date Deposited: 02 Nov 2020 16:26
URI: https://openaccess.city.ac.uk/id/eprint/25184
[img] Text - Accepted Version
This document is not freely accessible until 31 March 2023 due to copyright restrictions.

To request a copy, please use the button below.

Request a copy

Export

Downloads

Downloads per month over past year

View more statistics

Actions (login required)

Admin Login Admin Login