Immersive virtual reality gameplay detects visuospatial atypicality, including unilateral spatial neglect, following brain injury: a pilot study

Abstract

Background: In neurorehabilitation, problems with visuospatial attention, including unilateral spatial neglect, are prevalent and routinely assessed by pen-and-paper tests, which are limited in accuracy and sensitivity. Immersive virtual reality (VR), which motivates a much wider (more intuitive) spatial behaviour, promises new futures for identifying visuospatial atypicality in multiple measures, which reflects cognitive and motor diversity across individuals with brain injuries.

Methods: In this pilot study, we had 9 clinician controls (mean age 43 years; 4 males) and 13 neurorehabilitation inpatients (mean age 59 years; 9 males) recruited a mean of 41 days post-injury play a VR visual search game. Primary injuries included 7 stroke, 4 traumatic brain injury, 2 other acquired brain injury. Three patients were identified as having left sided neglect prior to taking part in the VR. Response accuracy, reaction time, and headset and controller raycast orientation quantified gameplay. Normative modelling identified the typical gameplay bounds, and visuospatial atypicality was defined as gameplay beyond these bounds.

Results: The study found VR to be feasible, with only minor instances of motion sickness, positive user experiences, and satisfactory system usability. Crucially, the analytical method, which emphasized identifying 'visuospatial atypicality,' proved effective. Visuospatial atypicality was more commonly observed in patients compared to controls and was prevalent in both groups of patients-those with and without neglect.

Conclusion: Our research indicates that normative modelling of VR gameplay is a promising tool for identifying visuospatial atypicality after acute brain injury. This approach holds potential for a detailed examination of neglect.

Keywords: Brain injury; Classification; Cognitive assessment; Immersive virtual reality; Unilateral visuospatial neglect.

Fig. 1

The Attention Atlas. a Visual search localisation gameplay. Array elements were positioned on a spherical surface with an origin at the headset position, calibrated at the start of each level [31]. This panel depicts the full field level. b Coordinates. Array elements were presented on a spherical grid. Each game level used a subset of possible positions, which could appear within the central field of view (FOV) or towards or beyond its edge, requiring head and eye movements for target localisation. For the axes level, the most eccentric horizontal and vertical positions fell outside the central FOV. c Cue/array trial structure. Cues and arrays were presented until an element was selected. d Game levels are depicted from the first-person perspective. Stimuli are scaled for clarity. Those with eight elements were presented on a single concentric ring presented 15° from central vision. The tutorial, excluded from analysis, was a search of a red “T” among blue “Ls”. Axes was a search for a “T” among “Ls” positioned horizontally or vertically on separate trials. Stimuli were food, cards, and balloons on separate trials. For food, the target and distractors are randomly selected at the beginning of each trial from 121 food icons. The queen of diamonds was the target for cards. For balloons, the target was a balloon without a string located among balloons with strings. Depth presented elements simultaneously at one of two depths: a near-surface (2 m) and a far-surface (4 m). Full field presented elements in four concentric rings. Free viewing depicted a low-resolution polygon forest, which surrounded the player in 360°. We instructed players to “look around” and report what they could see

Fig. 2

Feasibility of The Attention Atlas. a Motion sickness during VR on the SSQ. b Game experience on the GEQ-R (1 = not at all, 2 = slightly, 3 = moderately, 4 = fairly, 5 = extremely). This and subsequent box plots show the median, the 50th and 75th percentiles (at the upper/lower hinges), and 1.5 × IQR from the hinges (whiskers). In this and subsequent plots, points represent players. c System usability on the SUS. Shaded regions show acceptability cut-offs. This and subsequent half-violin plots show the kernel density estimation

Fig. 3

Summary of visuospatial atypicality. a Patient summary matrix. b Control summary matrix. The six primary performance metrics (response accuracy, RT, headset latitude mean, headset longitude mean, controller latitude mean, controller longitude mean) reflect the mean number of outliers across game levels. The vertical axis plots game ID, and the horizontal axis plots attention metrics. Black rectangles show outlier absence (i.e., typical performance), and non-black squares with red outliers show outlier presence (i.e., atypical performance). Yellow squares show missing data. For the first two metrics (accuracy and RT), outlier Boolean proportion (0–1) reflected the mean across game levels. LF = left/right contrast; UD = up/down contrast; 25.0° diff. = 12.5°/25.0° contrast; 37.5° diff. = 25.0°/37.5° contrast; lat. = latitude; lon. = longitude; M = mean

Fig. 4

Mean visuospatial atypicality. Patients versus controls

Fig. 5

An illustrative case. This figure depicts raycasts for patient 17. a Headset and controller maps. For each level, longitude and latitude coordinates were converted into 2D histograms with bounds of -50° and + 50° and a bin width of 1°. Raycast maps were limited to search array rather than cue periods. Points represent array element locations. Blue points represent unselected locations, and pink points represent selected locations. b-c Raycast distributions. Boxplots, kernel density estimates, and individual frame measurements are depicted. b Raycast latitude means. c Raycast longitude means