IAVS: A Multi-Center Dataset and Applicability Evaluation System for Computational Fluid Dynamics-Oriented Intracranial Aneurysm Segmentation

Feiyang Xiao, Yichi Zhang, Xigui Li, Yuanye Zhou, Chen Jiang, Xin Guo, Limei Han, Yuxin Li, Fengping Zhu, and Yuan Cheng. MICCAI 2026.

[Paper] [Project page] [Data access]

IAVS is a multi-center benchmark for intracranial aneurysm (IA) and parent-vessel segmentation designed around a practical question: can a segmentation be converted into a simulation-ready computational fluid dynamics (CFD) model? Alongside conventional image-based metrics, IAVS introduces the CFD-Applicability Score (CFD-AS) to assess whether a correctly localized aneurysm prediction passes vascular topology inspection, mesh generation, and blood-flow computation.

Dataset at a glance

IAVS contains 641 three-dimensional MRA examinations, 587 expert-verified aneurysm and IA-vessel annotations, and 124 aneurysm-negative examinations. Cases are annotated and quality-controlled for downstream CFD use. The complete IAVS data model comprises seven asset types:

1. Whole-brain MRA image
2. Intracranial-aneurysm mask
3. IA-vessel mask
4. STL surface model with cut inlet and outlet faces
5. Vascular centerlines
6. Mesh files with boundary annotations
7. CFD analysis outputs

Partition	Images	Public-source cases	Private cases	Intended use
Train/validation	467	175	292	Method development
Set A	76	76	0	Public-source internal evaluation
Set B	98	0	98	Private clinical-scenario evaluation
Total	641	251	390	Benchmark

The data were integrated from ADAM, INSTED, the Royal Brisbane TOF-MRA Intracranial Aneurysm Database, and an in-house cohort. IAVS-derived IA-vessel and CFD assets are not interchangeable with the original source annotations.

Data access and redistribution

No patient-level image, annotation, mesh, or CFD-result file is stored in this Git repository. This is intentional: public access to a source dataset does not necessarily grant permission to redistribute it or its derivatives.

The data card records the exact release boundary, source links, attribution requirements, and access restrictions. In particular, ADAM data and all data derived from it must be obtained directly from the challenge organizers; private IAVS cases are not released. The Royal Brisbane source dataset is CC0 and can be downloaded from OpenNeuro ds005096. Any IAVS-derived public assets will be versioned and deposited separately rather than committed to Git.

CFD applicability evaluation

For a predicted segmentation, IAVS defines three binary checks:

• VTA — vascular topology availability;
• MGA — mesh generation availability;
• BFA — blood-flow availability.

A true-positive prediction is CFD-applicable only when all three checks succeed. With TdP denoting the number of such predictions, CFD-AS is

CFD-AS = TdP / (TP + FP + FN).

VTA, MGA, and BFA are diagnostic stages rather than stand-alone quality metrics. They identify whether a failure arose from a topological defect, unsuccessful meshing, or failure in the subsequent flow simulation. The reference pipeline comprises topology inspection, morphological preprocessing, surface and centerline extraction, inlet/outlet cutting, mesh enhancement and fitting, boundary assignment, meshing, and CFD computation.

The runnable CFD-AS reference implementation, its environment specification, and a scored example are available in cfd_as/. The geometry workflow retains its original 3D Slicer, SlicerVMTK, Geomagic Wrap, and Ansys SpaceClaim integrations; commercial applications and patient-level inputs are not bundled.

Repository status

Component	Status
Paper, citation, and dataset documentation	Available
Source-data access and redistribution policy	Available
Public IAVS derivative-data release	In preparation; see the data card
CFD-AS implementation	Available in cfd_as/
Baseline training/inference code and weights	In preparation

Citation

@InProceedings{Xiao2026IAVS,
  author    = {Feiyang Xiao and Yichi Zhang and Xigui Li and Yuanye Zhou and
               Chen Jiang and Xin Guo and Limei Han and Yuxin Li and
               Fengping Zhu and Yuan Cheng},
  title     = {IAVS: A Multi-Center Dataset and Applicability Evaluation System
               for Computational Fluid Dynamics-Oriented Intracranial Aneurysm Segmentation},
  booktitle = {Medical Image Computing and Computer Assisted Intervention (MICCAI)},
  year      = {2026}
}

Contact

For questions about IAVS, data access, or the CFD-AS release, please open a GitHub issue after the issue tracker is enabled, or contact the corresponding authors listed in the paper.