NeuroVase: AR Platform for Neurovascular Learning

• NeuroVase is an augmented reality educational system that uses tangible cue cards and interactive 3D models to deliver stroke-relevant neurovascular anatomy content.
• The platform integrates Unity 3D, Vuforia image tracking, and high-resolution MRI/MRA data to generate precise anatomical visualizations in an engaging dual-mode setup.
• User studies indicate strong improvements in engagement, usability, and learning gains, demonstrating its effectiveness as an active, spatially grounded learning tool.

Searching arXiv for "NeuroVase" and related papers to ground the article. arxiv_search(query="NeuroVase neurovascular", max_results=10, sort_by="relevance") arXiv search: NeuroVase (Jahani et al., 31 Mar 2026), neurovascular reconstruction (2002.01568), cerebrovascular segmentation (Abbas et al., 2023, Vayeghan et al., 23 Nov 2025, Galati et al., 1 Oct 2025), vascular generative modeling (Feldman et al., 17 Jun 2025). NeuroVase is a tablet-based mobile augmented reality learning system for neurovascular anatomy and stroke education that uses tangible cue cards as both standalone study aids and image-based AR markers for interactive 3D content delivery (Jahani et al., 31 Mar 2026). In the published literature, this specific educational platform is the primary formal referent of the name. In adjacent neurovascular computing literature, the same label is also invoked as a natural designation for organ-scale neurovascular reconstruction, vessel segmentation, planning, and vascular generative modeling; this suggests a broader conceptual association with systems that integrate neurovascular structure, visualization, and quantitative analysis (2002.01568, Vayeghan et al., 23 Nov 2025, Feldman et al., 17 Jun 2025).

1. Definition and scope

NeuroVase was introduced as a mobile AR platform motivated by the difficulty of teaching cerebrovascular anatomy, vascular territories, and stroke syndromes through static 2D diagrams and printed materials alone (Jahani et al., 31 Mar 2026). Its stated educational target is a strong spatial understanding of cerebral arterial anatomy, vascular territories, and their relation to stroke symptoms and severity. The system is explicitly framed around stroke-relevant neuroanatomy rather than general-purpose brain visualization.

The platform combines two components. The first is a mobile AR application running on a tablet; in the reported study, the device was a 10.9-inch Apple iPad. The second is a deck of tangible cue cards whose front faces function as image targets for AR tracking while their back faces provide concise text summaries for offline review. This “dual-mode setup” is central to the system’s identity because it couples physical study artifacts with interactive 3D visualization in a single workflow (Jahani et al., 31 Mar 2026).

A frequent misconception is that NeuroVase was presented as a general AR replacement for conventional instruction. The reported evidence is narrower. The controlled study showed that both AR and traditional paper-based methods significantly improved quiz scores, while NeuroVase yielded stronger usability, engagement, and enjoyment outcomes and at least comparable learning gains rather than an unequivocal superiority in raw knowledge improvement (Jahani et al., 31 Mar 2026).

2. System design, tracking, and model construction

The implementation uses Unity 3D (version 2023.1.16f1), C# for application logic and interaction, Vuforia Engine for image-based tracking, and Easy Volume Renderer to load and show MRI volumes (Jahani et al., 31 Mar 2026). AR content is anchored to physical cards. Each card front is designed as a Vuforia image target and contains a detailed anatomical illustration together with two QR codes in diagonal corners to increase feature density and high-contrast edges. When a card is detected, Vuforia estimates the 6-DOF pose of the card relative to the camera, and Unity renders the 3D model at that pose.

The cue-card system is divided into a master card and trigger cards. The master card is the primary anchor for the 3D brain model; when it is visible, the main virtual model appears above it and follows the card’s movement and rotation. Trigger cards are module-specific cards that change what is shown in AR. They are organized into three color-coded sets: 9 cards for Lobar Anatomy, 7 cards for Vascular Anatomy, and 4 cards for Vascular Territories. During the AR learning phase, participants were asked to look only at the front side; the back side was used for offline review after the post-test (Jahani et al., 31 Mar 2026).

The 3D anatomical content is assembled from multiple data sources. The lobar anatomy module uses structures extracted from the BCI-DNI brain atlas. The vascular anatomy and territory modules were built from one healthy adult subject with T1-weighted MRI at $1 \times 1 \times 1$ mm$^3$ resolution and Time-of-Flight MRA at $0.47 \times 0.47 \times 0.70$ mm$^3$ resolution. Brain surface extraction used BEaST, vessel extraction used a Frangi multiscale vesselness filter, and expert manual segmentation of arterial structures was performed in ITK-SNAP by a co-author with more than 10 years of neuroanatomy experience. Vascular territories were derived from the digital 3D arterial territories atlas of Liu et al. (2023), itself derived from lesion distributions in 1,298 acute stroke patients, and registered non-linearly to the subject MRI. The system uses the Level-2 subdivision with four major territories: ACA, MCA, PCA, and vertebrobasilar. All segmentations were converted to polygon meshes in .obj format and imported into Unity (Jahani et al., 31 Mar 2026).

3. Curriculum and interaction model

The pedagogical design is structured into three sequential modules. The first, Lobar Anatomy, covers the frontal, parietal, temporal, and occipital lobes together with the brainstem and cerebellum. The second, Cerebral Arterial System, covers the internal carotid artery, anterior cerebral artery, middle cerebral artery, posterior cerebral artery, vertebral and basilar arteries, and the Circle of Willis. The third, Vascular Territories, links each major arterial territory to its 3D extent, associated functional regions, frequency of strokes in that territory, and typical stroke symptoms and syndromes (Jahani et al., 31 Mar 2026).

This sequencing is not incidental. The system explicitly uses lobar anatomy as scaffolding for arterial anatomy and uses arterial anatomy as scaffolding for vascular territories. In operational terms, trigger cards are module-specific, and cards from other modules have no effect when the application is in a different module. This enforces a structured learning order rather than a free-form gallery of models (Jahani et al., 31 Mar 2026).

Interaction occurs through three channels. Physical cards select or anchor content. Touch gestures on the tablet rotate, scale, and translate the 3D model: one-finger swipe rotates, two-finger pinch or unpinch scales, and two-finger swipe moves the model in the AR scene. The master card also allows physical repositioning of the virtual model by moving the card itself. In the in-module AR view, detection of a trigger card causes the relevant structure or territory to be highlighted while a text panel updates with explanatory information (Jahani et al., 31 Mar 2026).

The paper does not formalize the pedagogy in a heavy learning-theory vocabulary, but it explicitly aligns the system with active learning, constructivism, embodied and tangible learning, and spatial learning. A plausible implication is that NeuroVase is best understood as a spatially grounded instructional environment in which conceptual progression, object manipulation, and physical referents are deliberately coupled rather than merely co-present (Jahani et al., 31 Mar 2026).

4. Evaluation, usability, and empirical findings

NeuroVase was evaluated in a controlled user study with 40 STEM undergraduate or graduate participants who had limited cerebrovascular knowledge. The AR group comprised 20 participants, with age $29.0 \pm 4.3$ years; the control group also comprised 20 participants, with age $29.7 \pm 5.3$ years. The control condition used traditional paper-based educational materials with 2D textbook images, static diagrams, and text identical to the app’s content, organized into the same three sections as the AR condition (Jahani et al., 31 Mar 2026).

The study procedure consisted of introduction and consent, an 18-question pre-study quiz, a 30-minute learning phase, the same 18-question post-study quiz, and then questionnaires. In the AR condition, participants received a brief tutorial, studied the three modules in fixed order, and used the app with the cue-card fronts only during the learning phase. The AR group completed the System Usability Scale and five custom UX questions; the control group completed three UX questions. Statistical analysis used the Mann–Whitney U test for quiz comparisons, a one-sample $t$-test comparing SUS to the threshold of 68, Wilcoxon signed-rank tests for SUS sub-items and UX items against neutrality, and a significance level of $p < 0.05$ (Jahani et al., 31 Mar 2026).

The quiz results showed substantial gains in both conditions. The AR group improved from $40.83\% \pm 7.71\%$ pre-study to $70.28\% \pm 14.11\%$ post-study, while the control group improved from $^3$0 to $^3$1. Both within-group improvements were significant at $^3$2. Average improvement was $^3$3 for the AR group and $^3$4 for the control group. Pre-study scores were significantly higher in the AR group ($^3$5); post-study scores were higher in the AR group but not significantly ($^3$6); and the gain difference was not significant ($^3$7) (Jahani et al., 31 Mar 2026).

Usability results were stronger. The mean SUS score was $^3$8 for $^3$9, significantly above the threshold 68 at $0.47 \times 0.47 \times 0.70$0. All SUS items differed significantly from neutral. Custom UX ratings in the AR group were all significantly above neutral at $0.47 \times 0.47 \times 0.70$1: enjoyment $0.47 \times 0.47 \times 0.70$2, engagement $0.47 \times 0.47 \times 0.70$3, perceived usefulness $0.47 \times 0.47 \times 0.70$4, effectiveness of AR visualization $0.47 \times 0.47 \times 0.70$5, and cue card design quality $0.47 \times 0.47 \times 0.70$6. By contrast, control-group UX responses were roughly neutral. Between-group comparisons showed significantly higher enjoyment ($0.47 \times 0.47 \times 0.70$7) and engagement ($0.47 \times 0.47 \times 0.70$8) for the AR condition (Jahani et al., 31 Mar 2026).

Qualitative comments were broadly consistent with the quantitative results. In the AR group, positive themes included 3D visualization helpfulness, ease of use, and smooth interaction design. Reported issues included marker tracking glitches, occasional content overload, and requests for more detailed anatomy, clarification of medical jargon, repositionable text boxes, blood-flow animations, and improved rotation controls. This pattern indicates that the principal strengths were spatial visualization and interaction, whereas the principal limitations were tracking robustness and information-presentation tuning (Jahani et al., 31 Mar 2026).

5. Position within the broader neurovascular computing literature

Although NeuroVase is formally an educational AR system, the surrounding neurovascular literature associates the same name with a wider class of systems that integrate segmentation, reconstruction, planning, and simulation. In organ-scale microscopy, DVNet was presented as a memory-efficient fully convolutional 3D encoder-decoder for cellular and microvascular segmentation in multi-terabyte KESM volumes, enabling quantitative metrics for organ-scale neurovascular analysis; its reported workflow includes patch-wise semantic segmentation followed by downstream cellular and vascular reconstruction steps (2002.01568). In a clinical MRI setting, NeuroVascU-Net targeted automatic 3D segmentation of cerebral vasculature from T1-weighted contrast-enhanced MRI in neuro-oncology patients and reported a Dice score of 0.8609, precision of 0.8841, and 12,429,814 parameters on a test set of 27 patients (Vayeghan et al., 23 Nov 2025). In TOF-MRA, CV-AttentionUNet combined vesselness preprocessing, spatial attention gates, deep supervision, and Tversky loss for 3D cerebrovascular segmentation, reporting strong performance on TubeTK-derived datasets (Abbas et al., 2023). A multi-domain StyleGAN2-based framework subsequently extended brain-vessel segmentation across centers, modalities, and vessel types through label-preserving feature disentanglement and semi-supervised domain adaptation (Galati et al., 1 Oct 2025).

Generative modeling constitutes another adjacent branch. A recursive variational neural network for 3D blood vessel generative modeling represented vascular trees as binary hierarchies of centerline points and radii and generated anatomically plausible cerebral vessels, including aneurysm-bearing examples, from a learned latent space (Feldman et al., 17 Jun 2025). This suggests that the broader “NeuroVase” idea can extend beyond visualization and segmentation toward synthesis, augmentation, and simulation.

The educational system and these computational systems address different layers of the same domain. NeuroVase proper focuses on human spatial learning of arterial anatomy, vascular territories, and stroke syndromes (Jahani et al., 31 Mar 2026). The segmentation and generative systems focus on extracting, reconstructing, or synthesizing vascular structure from imaging data (2002.01568, Vayeghan et al., 23 Nov 2025, Abbas et al., 2023, Galati et al., 1 Oct 2025, Feldman et al., 17 Jun 2025). A plausible implication is that the term now occupies a small but growing semantic field spanning pedagogy, image computing, and neurovascular representation.

6. Limitations, interpretations, and prospective development

The reported NeuroVase study has several explicit limitations. The participant population consisted of STEM students rather than medical students or practicing clinicians; the sample size was limited to $0.47 \times 0.47 \times 0.70$9 per group; the AR group had higher baseline pre-test scores; the content focused on major lobes, major arteries, and four large vascular territories; and the outcome measures emphasized knowledge tests and user experience rather than long-term retention, diagnostic speed, or clinical decision-making (Jahani et al., 31 Mar 2026). These constraints delimit the strength of claims that can be made about professional training efficacy.

Technical limitations were also clear. Image-based tracking was sensitive to lighting conditions, viewing angle, and card occlusion, producing occasional recognition glitches. The platform was implemented on an iPad and uses standard tablet AR rather than stereo head-mounted displays. The paper does not report frame rate or measured latency. Future work proposed by the authors includes improving marker design and lighting guidance, refining gestures, allowing repositioning of text panels, increasing anatomical detail, and expanding stroke content with more explicit syndromes and case-based modules (Jahani et al., 31 Mar 2026).

In the broader neurovascular sense, future development appears likely to couple systems of the NeuroVase type with computational pipelines for vessel extraction and analysis. The surrounding literature already supports organ-scale microvascular reconstruction (2002.01568), T1CE-based neurosurgical vessel segmentation (Vayeghan et al., 23 Nov 2025), multi-domain adaptation across MRA, CTA, and SWI (Galati et al., 1 Oct 2025), and generative vascular-tree synthesis (Feldman et al., 17 Jun 2025). This suggests that a mature neurovascular platform could eventually connect pedagogical 3D visualization, patient-specific segmentation, and synthetic vascular modeling within a unified environment, although that integrated platform is not yet the object of a single published system.

NeuroVase therefore occupies a dual position. In its strict and primary sense, it is a clinician-informed, stroke-focused mobile AR learning system with a dual-mode cue-card design and high reported usability (Jahani et al., 31 Mar 2026). In a broader and more speculative sense, the name has become associated with a family of neurovascular technologies concerned with making cerebrovascular structure computable, explorable, and quantitatively actionable across education, reconstruction, and clinical imaging (2002.01568, Vayeghan et al., 23 Nov 2025, Feldman et al., 17 Jun 2025).