Simulating Vision Impairment in Virtual Reality -- A Comparison of Visual Task Performance with Real and Simulated Tunnel Vision

TL;DR

This study investigates the validity of simulating gaze-contingent tunnel vision in VR by comparing performance and gaze behavior between eight RP patients and eight visually healthy participants with simulated tunnel vision across three VR tasks. It leverages data from a prior RP study for the patient group and an at-home VR setup with gaze-contingent masks for the simulated group, applying TOST equivalence testing and linear mixed modeling to assess task-based equivalence and learning effects. The main contributions show that VR-based simulation can reproduce RP-like performance at the group level in most tasks, but gaze-related measures diverge and only gradually converge with extended exposure, highlighting both the utility and limits of VR simulations for studying visual field defects. The findings inform the design of VR-based accessibility evaluations and training tools, suggesting cautious use of simulations for gaze-behavior analyses and the potential for broader, at-home study populations when focusing on quantitative task performance.

Abstract

Purpose: In this work, we explore the potential and limitations of simulating gaze-contingent tunnel vision conditions using Virtual Reality (VR) with built-in eye tracking technology. This approach promises an easy and accessible way of expanding study populations and test groups for visual training, visual aids, or accessibility evaluations. However, it is crucial to assess the validity and reliability of simulating these types of visual impairments and evaluate the extend to which participants with simulated tunnel vision can represent real patients. Methods: Two age-matched participant groups were acquired: The first group (n=8 aged 20-60, average 49.1, sd 13.2) consisted of patients diagnosed with Retinitis pigmentosa (RP). The second group (n=8, aged 27-59, average 46.5, sd 10.8) consisted of visually healthy participants with simulated tunnel vision. Both groups carried out different visual tasks in a virtual environment for 30 minutes per day over the course of four weeks. Task performances as well as gaze characteristics were evaluated in both groups over the course of the study. Results: Using the "two one-sided tests for equivalence" method, the two groups were found to perform similar in all three visual tasks. Significant differences between groups were found in different aspects of their gaze behavior, though most of these aspects seem to converge over time. Conclusion: Our study evaluates the potential and limitations of using Virtual Reality technology to simulate the effects of tunnel vision within controlled virtual environments. We find that the simulation accurately represents performance of RP patients in the context of group averages, but fails to fully replicate effects on gaze behavior.

Figures (9)

• Figure 1: Examples of the three visual tasks carried out by the participants, from neugebauer2023gazeTraining. (a-c) show the tasks without tunnel vision simulation. (d-f) show the tasks with an exemplary simulated deficient visual field of $\sim$15° diameter, giving an impression of the visuals displayed to the participant group with simulated tunnel vision. (a,d) Target Tracking; (b,e) Search task; (c,f) Navigation task. The interactive area in the Target Tracking and Search task (visbile in a and b) have dimensions of 80° horizontally and 60° vertically.
• Figure 2: Example image of one of the obstacle course layouts used in the navigation task, from neugebauer2023gazeTraining. s: Starting position; o: Obstacles (example selection); g: Goal area.
• Figure 3: Schematic of the study layout, showing the introductory in-person session, the task execution phase of 20 sessions of 30 minutes each over four weeks, and a concluding in-person session.
• Figure 4: Visualization of the concept of the Dynamic Visual Field. The array of black pixels marks a large visual area that could potentially be observed by the participant. The blue circle marks the VF of the patient. Orange arrows indicate gaze movement over a specified time window. Light-grey pixels mark the area that has been observed at any point during the moving time window. The Dynamic Visual Field is defined as the ratio between light-grey pixels and total number of pixels at the end of the time window, meaning in the right-most image.
• Figure 5: Results for the task performance in the three visual tasks. Dotted lines show the average results for each individual participant, continuous lines show the average regression of all trials. Results for trial duration and number of collisions in the navigation task are shown separately. Non-logarithmic trial duration values are shown for better visualization despite statistical evaluation only considering the logarithm of trial duration for better normal distribution. Each plot includes the resulting p-values of the statistical models conducted: Eq describes the equivalence between groups found with TOST, Dif describes the effect between groups found with an LMM, and LR describes the interaction effect between the Training Session and Group parameter, meaning differences in the learning rate between groups. Interaction terms are not feasible in the nbGLMM, thus no p-value for LR is reported under (d).