Abstract
Background/Objective Cerebral injury due to stroke in childhood increases the risk of higher-order visual processing (HOVP) deficits, like cerebral visual impairment (CVI), which can lead to severe behavioral and learning disabilities if left untreated. Using a virtual reality-based search task and structural Magnetic Resonance Imaging analysis, we assess the extent of functional vision deficits in childhood stroke patients and potential anatomical correlates.
Methods 20 childhood stroke patients and 38 healthy controls completed a dynamic visual search task using a virtual reality/eye-tracking (VR/ET) paradigm to quantify functional vision abilities between 2021 to 2024 (average 7.34 years after stroke). Virtual reality assessment measures, stroke imaging characteristics (vascular territory, type, and visual pathway involvement) and neuropsychological outcomes were analyzed between cohorts using standard statistical comparison methods and linear regression model.
Results All childhood stroke patients could complete virtual reality visual search tasks as demonstrated by success rates and task compliance in equal measure to controls. However, less accurate search and slower fixation rates together with less sensitivity to changes in task load and greater impairment in initiating a response to a target were observed in our patient cohort. On MRI lesion analysis injury involving the posterior visual pathways correlated with slower reaction time to fixation on a target independent of age, and the presence of attention deficit or other behavioral disorders.
Conclusions Bedside VR/ET assessment in children affected by stroke can detect signs of HOVP deficits. Imaging demonstrating involvement of the posterior visual pathway at the time of diagnosis strongly predicts development of impaired visual tracking abilities later in life. While detection of CVI and HOVP using current standard clinical and neuropsychological evaluations can be challenging, our study demonstrates that injury pattern on imaging at stroke onset can help identify children at risk of CVI. This can enable early diagnosis and timely accommodations facilitating functional vision development, critical to learning and skill acquisition.
Introduction
Stroke in childhood occurring during the perinatal period (28 weeks of gestation to 28 days of life) or pediatric (29 days to 18 years of age) is one of the leading causes of childhood disability1,2. More than 50% of perinatal and pediatric stroke survivors experience a cognitive/sensory disorder impacting their quality of life, making early detection critical to determining early interventions that could improve outcomes3-11. While motor deficits become apparent early on, most behavioral or cognitive deficits only become apparent as developmental stages are attained and vulnerabilities become outwardly observable, thus delaying implementation of rehabilitation interventions12-14. Critical to the development of higher cognitive functions are higher-order visual processing (HOVP) systems, defined as the cognitive abilities of integrating visual input for goal-oriented tasks15,16. Visual field loss17, oculomotor dysfunction, and visual perceptual impairments17-19 have been extensively studied in adult stroke survivors. Studies assessing HOVP found that up to 30% of survivors suffer from visual inattention, impaired functional vision (visual task-related ability) and/or difficulty with adaptation to changes in environmental stimuli8,20. Presence of HOVP impairments was determined by measuring the ability to modulate responses to visuospatial cues21, reaction time and accuracy in goal-oriented visuospatial tasks22, or assessing reading difficulties due to eye movement abnormalities23. Recently, virtual reality (VR) combined with eye tracking (ET) paradigms have been used to assess HOVP in a naturalistic environment while eliminating the reliance on the individual’s other abilities that stroke can affect, including verbal and motor abilities24,25.
At the anatomical level, HOVP requires function of the posterior visual pathway (PVP) - defined as the optic radiations (OR), lateral geniculate nucleus (LGN), and occipital lobe - and from there two main streams or pathways 1) the ventral stream to the temporal lobe (the “what pathway”), involved in object identification and recognition; and 2) the dorsal stream to the parietal lobe (“where pathway”), involved in object’s spatial location and visual input to motor functions26-28. Injury to the PVP, commonly identifiable in imaging after stroke, has been correlated with visual field deficits29,30, poor visual function31,32, and impaired visual acuity33.
Cerebral visual impairment (CVI), an HOVP disorder characterized by visual dysfunction resulting from early damage and/or maldevelopment of visual pathways and processing centers of the brain15,34-36 is the leading cause of congenital visual impairment in developed countries28,37. While visual acuity is regularly assessed in children, functional vision is not typically included in early ophthalmological or neurological assessments in perinatal and pediatric stroke survivors, as it requires a multidisciplinary approach and the child’s ability to complete complex testing batteries38. Early detection of CVI in children is critical as early implementation of interventions can maximize rehabilitative neuroplasticity 8,39-41. Implementing individualized educational programs (IEPs) to eliminate environmental clutter42, mask surrounding text while reading, and encourage educators to wear distinct, bright clothing to aid in identification33, is recommended for children with CVI. If left unaccommodated for, functional vision deficits can create additional barriers to success and impede children’s access to education and rehabilitative therapies, development, and functional independence43.
In this study we tested childhood stroke survivors with a VR/ET paradigm previously used to characterize HOVP deficits in CVI-diagnosed participants25,44-46 and correlated their performance with stroke lesion characteristics on brain imaging with neuropsychological testing scores to identify anatomical correlates to their HOVP abilities.
Methods
Participants
The study included three cohorts of participants: childhood stroke survivors (perinatal and pediatric stroke), children formally diagnosed with CVI, and controls (neurotypical development with normal visual acuity and no CVI). Participants were recruited and enrolled from June 2021-January 2024. Members of the stroke cohort were recruited from the MGH (Massachusetts General Hospital) Multidisciplinary Stroke Clinic in Boston, MA, during clinic visits. They completed the VR assessment following their consent. CVI participants had varied suspected causes of CVI including ischemic hypoxic/anoxic injury, periventricular leukomalacia, premature birth complications, and seizure. Members of the CVI and control cohorts were recruited from the Boston area. Details of CVI and control participant recruitment is described in Manley et al. 202246.
The study was approved by the local Institutional Review Board, Mass General Brigham Human Research Committee (MGBHRC) prior to participant enrollment and included both a retrospective medical records protocol (2015P002030) and a prospective protocol (2019P000070). For the retrospective protocol, participant consent was waived by the MGBHRC. For the prospective protocol, written informed consent was obtained from all participants (or guardians of participants). This study was carried out in accordance with the Code of Ethics of the World Medical Association (Declaration of Helsinki) for experiments involving humans.
VR Testing
The virtual hallway environment was developed using the Unity 3D game engine version 5.6 (Unity Technologies, San Francisco, CA, USA). Participants sat a viewing distance of 50-60cm (about 1.97 ft) away from a 27” laptop monitor, with the eye tracker unit (Tobii 4C; 90 Hz sampling frequency. Tobii Technology AB, Stockholm, Sweden) mounted on the lower portion of the monitor. The virtual hallway stimulus subtended 58 × 32 degrees of visual angle.
Participants were able to move their heads freely but reminded to keep their gaze on the screen throughout testing. Three-dimensional object models were created using Blender modeling software (Blender Foundation), and 3D human models were created in Adobe Fuse CC and rigged for animation in Adobe Mixamo (Adobe Systems Inc., San Jose, CA, USA)47.
The virtual hallway is a dynamic search task in which participants are instructed to scan a screen, displaying the hallway of a fictitious school with a crowd of animated characters walking toward the observer (Fig. 1). From a fixed, first-person perspective, participants are instructed to locate and pursue the principal each time she appears on the screen from one of the eight entrances until she exits the scene. The interval between a target disappearing and reappearing in the hallway from trial to trial varied between 5 and 15 sec. The duration of the target’s visibility was determined by its starting point and path length. This varied between 5 and 17 sec for the closest and furthest starting points, respectively. Crowd density (the number of characters on screen at a given time) was manipulated by three categories: low (average of 5 ± 5 people), medium (10 ± 5), and high (15 ± 5), with each level of crowd density presented equally and in a pseudo-random fashion. Each trial run lasted about 3.5 minutes, and participants completed 2 trial runs with a brief break in between. For further details regarding the design of the visual stimulus and task, see Bennett et. al., 202246.
The dynamic search task tracks participants’ eye movements as they are tasked with searching and pursuing a character (denoted as the principal, circled) as various “distractor” characters walk in and out of the hallway on screen. Crowd density fluctuates throughout the trial from Low (5 ± 5 people), Medium (10 ± 5 people), and High (15 ± 5 people) task difficulty. The calibrated eye tracker collects visuomotor data, then processed as the outcome metrics of the task.
Outcome Measures
Eye gaze behavior was processed using an in-house Matlab script (MathWorks, Natick, MA, USA). Metrics used for analysis were success rate, reaction time, gaze error, search area, and off-screen percentage. Success rate is the percentage representing how many times the participant located and fixated on the target (the principal) in each trial. Reaction time is defined as the first moment gaze within the target’s outer contour and remains fixated for the presentation. Gaze error is a measure of fixation accuracy assessed as the distance between the center of the target and the participant’s gaze position. Search area represents the visual search area explored and serves as an index for search precision; it’s measured by an ellipse-shaped 95% confidence interval fitted to the captured eye-tracking data. Off-screen percentage measures task compliance, assessing the proportion of gaze points falling outside the bounds of the screen on each trial.
Retrospective Medical Record Review
Childhood stroke participants’ medical records and brain multimodal magnetic resonance imaging (MRI) were reviewed to characterize and segment the cohort based on different stroke characteristics including stroke laterality (left/right/bilateral), age of stroke onset (perinatal/pediatric), territory affected (anterior cerebral artery (ACA)/medial cerebral artery (MCA)/posterior cerebral artery (PCA)/cerebellar/other), and stroke type (hemorrhagic/ischemic/venous infarction). A neuroradiologist (A.L.) blinded to clinical information and the VR/ET task results assessed the MRI closest to the incident of stroke to determine if the injury involved the PVP. The PVP was assessed on multi-modal axial and/or coronal planes, and assigned a binary categorization of “yes” or “no.” This determination was based on whether chronic or acute stroke lesions involved the PVP. To further characterize behavioral abilities, data regarding attention-deficit/hyperactivity disorder (ADHD), and neuropsychological evaluations were extracted after consultation with a psychology fellow (A.S). We extracted percentiles for three standardized, normative-based neuropsychological instruments assessing visuospatial construction through graphomotor integration/replicating drawings48, part-to-whole construction using blocks to replicate designs49-52, and visuospatial rotation and synthesis to mentally construct puzzles49-52 given their correlations with HOVP functions.
Statistical Analysis
Statistical analyses were carried out using GraphPad Prism version 10.0.2(171) (Boston, MA, USA). The Kolmogorov-Smirnov test was used to test for normality. For each behavioral outcome, an unpaired t-test, ANOVA, Mann-Whitney U, or Kruskal-Wallis test (dependent on normality) was conducted with cohorts as a between-subjects factor and task difficulty (i.e., low, medium, and high) as a within-subjects factor. For ANOVA and Kruskal-Wallis analyses, Dunn’s test to correct for multiple comparisons was carried out on mean differences of behavioral outcomes to explore the directionality of the observed effects. Between-group comparisons were planned between cohorts. Within-group comparisons were planned for reaction time based on task difficulty in childhood stroke and control cohorts. A multivariable linear regression model was created to determine if the following metrics correlated with the reaction time in the VR task for the stroke patients: age at VR testing, PVP injury (binary yes/no), and ADHD (binary yes/no). Results were considered significant if p<0.05.
Results
Childhood stroke cohort characteristics and feasibility of VR task
We studied participants in 3 cohorts: 1) a confirmed pediatric or perinatal stroke diagnosis as confirmed by MRI and no formal diagnosis of CVI (n=20, mean age: 9.9 years, range: 6-15 years; median: 9; inter-quartile range (IQR): 4) 2) participants formally diagnosed with CVI (n=17, mean age: 19.1 years, range: 8-31 years; median: 20; IQR: 8), and 3) controls with neurotypical development and no history of ophthalmic (e.g., strabismus, amblyopia) or neurodevelopmental conditions (n=38, mean age: 20.9 years, range: 11-33 years; median: 20; IQR: 4) (Table 1). All participants had normal or corrected-to-normal visual acuity and no visual field deficits. On average, patients had an MRI 71.9 days after the stroke. Of the childhood stroke cohort, 11 patients have experienced perinatal stroke, and 9 pediatric stroke. VR/ET testing happened 7.34 years after the stroke. Majority of stroke lesions were ischemic (65%), involved the MCA vascular territory (60%), and affected males (70%). Neuropsychological evaluations were available in 15 stroke patients, between the ages of 5 and 15 years, and the evaluation closest to the time of VR testing was analyzed. Further stroke cohort characteristics are detailed in Table 2.
No significant differences in VR/ET metrics were observed between perinatal and pediatric stroke cohorts (Mann-Whitney U tests: success rate: p=0.334, reaction time: p=0.992; gaze error: p=0.552; search area: p=>0.999; off-screen percentage: p=0.807) (Supp. Fig. 1). Both cohorts were therefore combined as a single childhood stroke cohort for subsequent analysis. This cohort’s ability to perform VR/ET tasks was confirmed similar to controls success rate (stroke: 94.72±6.62; controls: 97.66±3.03; Mann-Whitney U: p=0.162) and off-screen percentage values (stroke: 0.36±0.42; controls: 0.38±0.29; Mann-Whitney U: p=0.285) (Fig. 2A).
Despite similar task success compared to controls (A), stroke patients showed significant differences in (B) reaction time (p<0.0001), (C) gaze error (p<0.0001), and (D) search area (p<0.0001). (E) When assessing changes to reaction time across the varying task load, control participants have a steeper change in reaction time than childhood stroke patients, as shown by slope values of best-fit lines. Further, the childhood stroke cohort has a higher initial visual orienting response, as calculated by the extrapolated y-intercept, than control participants.
Participants’ age showed no effect on VR performance assessed by reaction time, despite ranges in age both within and across the childhood stroke and control cohorts [stroke cohort linear regression: p=0.7882 and R2 = 0.004355; control cohort linear regression p=0.7339 and R2 = 0.003344]. Regarding visuoperceptual-related tasks evaluated during neuropsychological evaluations available in 15 stroke patients, we found no significant correlation between scores and VR/ET metrics.
Childhood stroke patients show significant deficits in VR search task and decreased adaptive capacity to increased complexity of the visual environment
Childhood stroke patients had significantly higher reaction time to fixation (mean±SD: stroke: 2617.76±948.72; controls: 1499.82±492.21; t-test: p<0.0001), a larger gaze error (stroke: 6.27± 1.56; controls: 4.67±0.79; t-test: p<0.0001;), and a larger search area compared to controls (stroke:11.20±3.51; controls: 6.97±2.17; t-test p<0.0001) (Fig. 2B-D). Reaction time was assessed across the “low,” “medium,” and “high” task difficulty levels in both cohorts. Visual search was tested at increasing visual environment complexity from low to medium to high levels. While there was no significant difference across task complexity levels for reaction time in the childhood stroke cohort (low: 2371.81±1347.31; medium: 2921.16±1293.13; high: 2659.44±1000.67; Kruskal-Wallis: p=0.264), the control cohort exhibited significant differences (low: 1301.09±797.54; medium: 1509.96±582.61; high: 1688.93±581.55; Kruskal-Wallis: p=0.002). This resulted in a less steep slope for reaction time in the stroke cohort (slope=143.8 vs 193.92 of controls) denoting an impaired visual processing adaptive capacity. Additionally, the initial visual orienting response, or the reaction time to identify the target in the absence of distractors, as defined by the extrapolated y-intercept value of the reaction time at different levels of task complexity, was higher (y-intercept=2363.2) in childhood stroke patients than controls (y-intercept=1112.1) (Fig. 2E).
Laterality of stroke lesion alters VR outcomes
For participants in the childhood stroke cohort with a supratentorial stroke lesion, we assessed if stroke laterality exacerbated HOVP deficits as measured by the VR dynamic hallway test. No significant differences were found in gaze error and search area between left-sided (n=8), right-sided (n=5), or bilateral stroke lesions (n=3). Patients with right hemisphere lesions showed significantly higher success rates than bilateral stroke patients (right hemisphere stroke: 98.89±2.49; bilateral stroke: 96.67±8.48; Kruskal-Wallis: p=0.047) and faster reaction times (right hemisphere stroke: 1792.16±570.32; left hemisphere stroke: 3305.25±721.06; Kruskal-Wallis: p=0.014) than left hemisphere stroke patients (Supp. Fig. 2). Beyond laterality, there were no significant differences in the VR metrics based upon vascular territories affected - MCA, PCA, ACA, cerebellar or other territories (success rate: p=0.156, reaction time: p=0.486; gaze error: p=0.305; search area: p=0.53; off-screen percentage: p=0.727) (Supp. Fig. 4) or type of stroke– ischemic, hemorrhagic, or venous infarction (success rate: p=0.418, reaction time: p=0.832; gaze error: p=0.973; search area: p=0.348; off-screen percentage: p=0.496) (Supp. Fig. 5).
Posterior visual pathway involvement at time of stroke diagnosis predicts higher-order visual processing deficits
We found that stroke patients with lesions involving the PVP on MRI (n=10, of which 4 are pediatric and 6 are perinatal stroke) correlated with longer reaction times on the VR assessment (p=0.012) after controlling for age at VR assessment and possible confounding attention disorder diagnoses (Fig. 3). Specifically, reaction time in the VR assessment increased by 1121ms in those with PVP injury as compared to stroke patients with uninvolved PVP, after controlling for age at VR assessment and the diagnosis of behavioral disorders. See Figure 3D for further details of this multivariable linear regression model. Notably, laterality of PVP injury and time of stroke injury diagnosis (acute or subacute) did not show significant differences in reaction time (laterality: p=0.094; acute vs. subacute imaging: p=0.267).
(A) Diffusion-weighted imaging (DWI) representative of an ischemic stroke shows acute involvement of the left LGN (top left) and left occipital lobe (bottom left). T2-weighted (top right and bottom right) imaging representative of ischemic stroke with acute involvement of left optic tract and OR, respectively. Arrows illustrate where injury maps in visual pathway. (B) When controlling for age at VR testing and ADHD, childhood stroke patients with stroke injury involving the PVP significantly correlated with reaction time in the VR search task (p=0.0123). (C) Multivariable regression analysis shows a significant positive correlation between reaction time and stroke injury involving the PVP.
Childhood stroke survivors have similar VR/ET performance to CVI-diagnosed participants
We compared the VR performance metrics of the childhood stroke cohort to a cohort of formally diagnosed CVI participants (mean age: 19.1 years, range: 8-31 years) (n=17). CVI-diagnosed patients had an increased reaction time compared to stroke patients (p=0.0175; stroke: 2617.76±948.72; CVI: 4084.07±2158.77). In the remaining 4 metrics assessed from the dynamic hallway VR assessment, the childhood stroke and CVI cohorts showed similar impairments (success rate: p=0.296; gaze error: p=0.0513; search area: p=0.133; off-screen percentage: p=0.561) (Supp. Fig. 3). This preliminary analysis indicates that the HOVP abilities of childhood stroke patients align more closely with those of CVI-diagnosed participants than healthy controls.
Discussion
In this study, we demonstrated that VR/ET is a sensitive and feasible non-invasive method to assess HOVP in a cohort of children affected by childhood stroke. While no clinical diagnosis of HOVP disorders or CVI was present at the time of assessment all childhood stroke patients in our study showed significantly decreased visual processing abilities comparable to children diagnosed with CVI by multidisciplinary assessment. Our VR/ET paradigm has the potential to screen childhood stroke participants for HOVP deficits in a naturalistic and engaging test. For the first time, we show that involvement of the PVP on brain imaging at the time of stroke diagnosis is associated with significantly increased incidence of HOVP, providing a bedside tool to identify patients at risk to guide timely diagnosis and initiation of adaptive and habilitating interventions. If medical reports highlight PVP involvement in childhood stroke injuries, providers could further examine HOVP deficits. For example, MRI PVP involvement could be coupled with formal assessments of functional vision to guide accommodative practices for children’s visual processing deficits during early school years critical to learning and higher cognitive function development.
Our study is consistent with findings on visual processing impairments in adult stroke survivors, as we identify similar delays in task-related visuospatial tasks and inability to sufficiently modulate their performance to changes in environmental stimuli21,22. While visuospatial processing has been primarily associated with the brain’s right hemisphere53,54, our findings support literature that the left hemisphere is also essential for tasks of executive functions55.
Identification of HOVP disorders and CVI in children may impact their life-long developmental potential given their increased potential for neuroplasticity. Given that focal and diffuse white matter injury, when addressed early in development and close to stroke onset, the young brain can use remapping to compensate for functional loss at the injury site56.
As previously shown, special education programming can mitigate HOVP and/or CVI-related impairments that affect a child’s ability to access academic curriculum and make developmental progress. Services and accommodations, including visual stimulation therapy57, decluttering classroom environments42, and additional visual cues to signal visual attention33 yield benefits. Similarly, orientation and mobility services can ensure safe ambulation and attention to spatial surroundings, promoting independence and inclusion. These accommodations will help stroke patients complete goal-based tasks despite their decreased adaptive capacity to increased visually complex environments as identified in the VR/ET paradigm. Recognizing ways to tailor learning environments to the needs of stroke survivors can improve their overall experience and given age-dependent plasticity, early diagnosis and intervention can improve successful remapping56.
Therefore, utilizing accommodations recommended for CVI-diagnosed patients, such as masking texts when reading and eliminating unnecessary visual stimuli on worksheets, may improve childhood stroke survivors’ quality of life.
The limitations in our study include lacking age-matched stroke and control cohorts. However, we found that age didn’t affect success rate, reaction time, search area, or off-screen percentage metrics but did affect gaze error. A low sample size for the stroke cohort limited the analysis of whether stroke type or vascular territory affected contributed to HOVP. We speculate the lack of correlation between VR/ET metrics and neuropsychological testing scores may be due to low statistical power. Additionally, a binary assessment of the PVP by lesions on MRI (present or absent) doesn’t account for the degree of injury or loss of function. Our data is inclusive of various stroke subtypes, thus reflecting a clinically relevant sample of childhood stroke patients. To our knowledge, this is the first study that characterizes visual processing deficits and their correlation to anatomical injury at time of diagnosis providing a predictive tool to guide diagnostic workup and earlier meaningful interventions.
Future work evaluating this technique in a larger cohort would validate our findings and permit further analysis of outcomes of different stroke factors and neuropsychological testing correlates. Assessment and mapping of the functional connectivity of brain regions could provide detailed insight into the correlations between the severity of higher-order visual impairment and critical visual anatomical networks beyond our assessment of correlation to the PVP. Lastly, investigating differences in visual processing deficits in patients with diffuse and focal white matter injury could provide insight into white matter’s role in remapping abilities and recovery.
Data Availability
All data produced in the present study are available upon reasonable request to the authors
Source of funding
National Institute of Health/ National Eye Institute (NIH/NEI R01 EY030973) to LBM
Disclosures
none
Supplementary Figures
Comparing the performance metrics of pediatric and perinatal stroke patients did not show any significant differences across the 4 VR metrics, allowing us to combine these 2 cohorts into one aggregate childhood stroke cohort.
(A) Childhood stroke patients with right hemispheric injury exhibited significantly increased success rates as compared to stroke patients with bilateral injury (p=0.0469). (B) Stroke patients with right hemispheric injury also had significantly faster reaction times than patients with left hemispheric injury (p=0.0137). There were no statistically significant differences in gaze error (C) or search area (D) on the VR task for patients with left hemispheric, right hemispheric, or bilateral injuries.
Success rate (A), gaze error (C), and search area (D) on the VR search task did not significantly differ between the childhood stroke and CVI-diagnosed cohorts. (B) CVI-diagnosed participants exhibited increased reaction time compared to stroke patients (p=0.0175).
Dividing childhood stroke cohort according to the vascular regions affected (ACA, MCA, PCA, cerebellar, or other) did not show any significance across any of the 4 VR metrics captured.
Dividing childhood stroke cohort according to their stroke type (ischemic, hemorrhagic, and venous thrombosis) did not show any significance across any of the 4 VR metrics captured.
Footnotes
↵* The authors share first authorship;
References
