Abstract
In 2019, we faced a pandemic due to the coronavirus disease (COVID-19), with millions of confirmed cases and reported deaths. Even in recovered patients, symptoms can be persistent over weeks, termed Post-COVID. In addition to common symptoms of fatigue, muscle weakness, and cognitive impairments, visual impairments have been reported. Automatic classification of COVID and Post-COVID is researched based on blood samples and radiation-based procedures, among others. However, a symptom-oriented assessment for visual impairments is still missing. Thus, we propose a Virtual Reality environment in which stereoscopic stimuli are displayed to test the patient's stereopsis performance. While performing the visual tasks, the eyes' gaze and pupil diameter are recorded. We collected data from 15 controls and 20 Post-COVID patients in a study. Therefrom, we extracted features of three main data groups, stereopsis performance, pupil diameter, and gaze behavior, and trained various classifiers. The Random Forest classifier achieved the best result with 71% accuracy. The recorded data support the classification result showing worse stereopsis performance and eye movement alterations in Post-COVID. There are limitations in the study design, comprising a small sample size and the use of an eye tracking system.
Introduction: The Need for Functional Biomarkers
While the acute phase of COVID-19 is primarily respiratory, "Post-COVID" syndrome—defined by symptoms persisting beyond 12 weeks—frequently involves the central nervous system. Patients often report "brain fog," chronic fatigue, and cognitive dysfunction. Visual impairments and altered oculomotor behavior are increasingly recognized as part of this neurological profile. However, current diagnostic methods often rely on invasive blood tests or radiation-heavy imaging like CT scans. There is a critical need for non-invasive, symptom-oriented tools to assess these functional deficits. This research explores whether a brief VR-based visual task can serve as a functional biomarker for the neurological impact of Post-COVID.
Methodology
The researchers developed a VR environment using Unreal Engine 4, delivered through an HTC Vive Pro Eye headset equipped with an integrated Tobii eye-tracker (120 Hz).
The Stereopsis Task
Participants were presented with a rhombus-like arrangement of four virtual "footballs." In each trial, one ball was rendered with a specific disparity, making it appear slightly closer to the user than the others. The objective was to identify the closer ball as quickly as possible using keyboard inputs.
The difficulty was modulated by varying the "disparity difference"—the angular difference in how each eye perceives the object's position. Smaller disparities require higher stereoacuity to detect.
Experimental Protocol
The study involved a pilot cohort of 35 participants: 20 Post-COVID patients (average symptom duration ~389 days) and 15 age-matched healthy controls (20–35 years) to avoid age-related vision decline as a confounding factor. The experimental flow included initial calibration followed by multiple trials to ensure data robustness.
Data Analysis and Classification
The team extracted 73 features across three primary categories:
- Stereopsis Performance: Reaction times (mean, median, variance) and accuracy.
- Pupil Diameter: Statistical features and the "Low/High Index of Pupillary Activity" (LHIPA) to quantify cognitive load.
- Gaze Behavior: Number of fixations and angular velocity modeled asymptotically.
These features were used to train various classifiers, including Support Vector Machines (SVM), k-Nearest Neighbors (kNN), and Random Forest (RF), using a nested cross-validation approach.
Key Results
The Random Forest (RF) classifier proved most effective, achieving an overall accuracy of 71% using stereopsis performance data alone. The model demonstrated high sensitivity, correctly identifying 80% of Post-COVID patients (16 out of 20). However, specificity was lower at 60%, suggesting a tendency toward over-diagnosis in the control group. Gaze behavior alone also showed predictive power, reaching 66% accuracy with an SVM (RBF kernel).
Discussion: Neurological Fingerprints
The recorded data revealed several distinct patterns in the Post-COVID group:
- Oculomotor Fatigue: Patients showed a decrease in eye movement velocity, likely reflecting the muscle weakness and systemic fatigue characteristic of the syndrome.
- Cognitive Processing Delay: Reaction times were consistently higher in the Post-COVID group across all difficulty levels, providing a quantitative measure for the subjective "slowness" reported by patients.
- Visual Search Patterns: Post-COVID patients exhibited longer fixation durations on the background, suggesting increased difficulty in scene processing or decision-making.
- Pupillary Response: Interestingly, pupil diameter was less effective for classification. The authors suggest that the increased cognitive load (which typically increases pupil oscillation) might be masked by a sluggish pupillary response inherent to the disease.
Limitations and Future Outlook
The study's primary limitation is the small sample size (N=35), which necessitates further validation with larger, more diverse cohorts. Technical constraints included the VR headset's pixel density, which limited the minimum detectable disparity, and the requirement for participants to remove corrective eyewear, potentially affecting data for those with high prescriptions. Future work will explore time-series analysis of eye-tracking data and the application of this VR tool to other neurological conditions, such as mild traumatic brain injury.
Summary updated