Rethinking Intracranial Aneurysm Vessel Segmentation: A Perspective from Computational Fluid Dynamics Applications
Feiyang Xiao, Yichi Zhang, Xigui Li, Yuanye Zhou, Chen Jiang, Xin Guo, Limei Han, Yuxin Li, Fengping Zhu, Yuan Cheng
TL;DR
This paper introduces a new dataset and benchmarks for intracranial aneurysm vessel segmentation, emphasizing CFD applicability, and proposes a simple two-stage segmentation framework as a baseline for future research.
Contribution
The paper presents the first comprehensive multi-center dataset with CFD-related annotations and establishes evaluation benchmarks focused on clinical relevance and CFD model conversion.
Findings
Introduced the IAVS dataset with 641 3D MRA images and detailed annotations.
Developed two evaluation benchmarks for aneurysm localization and segmentation.
Proposed a two-stage segmentation framework serving as a baseline for future studies.
Abstract
The precise segmentation of intracranial aneurysms and their parent vessels (IA-Vessel) is a critical step for hemodynamic analyses, which mainly depends on computational fluid dynamics (CFD). However, current segmentation methods predominantly focus on image-based evaluation metrics, often neglecting their practical effectiveness in subsequent CFD applications. To address this deficiency, we present the Intracranial Aneurysm Vessel Segmentation (IAVS) dataset, the first comprehensive, multi-center collection comprising 641 3D MRA images with 587 annotations of aneurysms and IA-Vessels. In addition to image-mask pairs, IAVS dataset includes detailed hemodynamic analysis outcomes, addressing the limitations of existing datasets that neglect topological integrity and CFD applicability. To facilitate the development and evaluation of clinically relevant techniques, we construct two…
Peer Reviews
Decision·Submitted to ICLR 2026
The paper describes a novel attempt at setting up a complete system of intracranial aneurism detection and evaluation based on 3D MR angiography. The paper is very interesting, well presented and although not perfect, quite sound in its approach. Given the amount of work necessary to collect, annotate and perform CFD simulation on more than 600 MRA volumes, the proposed dataset is quite unique and valuable. The evaluation results are clearly the current state of the art among academic published
Not very many articles tackles the joint segmentation and CFD aspects of vessel segmentation, yet this has been an ongoing research area for a quite time. The article focuses on relatively recent, deep-learning based methods and does not attempt to review previous classical efforts. Previous to deep-learning, vessel segmentation methods used vesselness methods [1], many of which were reviewed and evaluated in [2]. Relatively recent projects have contributed to very similar objectives [3]. Overa
- This paper compiles and curates a large-scale 3D MRA dataset, combining existing datasets with a new in-house collection, resulting in a total of 641 volumes and IAs, which is quite impressive. - The author's promise to open-source data, code, and models is great.
- The overall writing appears somewhat disorganized to the reviewer. If the author intends to claim a contribution to data construction, more detailed information is expected, such as how the data was constructed and which aneurysm-related tasks it supports. From the reviewer's perspective, proposing a new model is unnecessary, particularly since you are introducing new metrics simultaneously. - In its current writing, the reviewer is unclear about how the existing dataset has been enhanced and
1. The proposed IAVS dataset covers multimodal annotations required for the complete workflow from imaging to CFD (3D MRA images, IA/IA-Vessel masks, STL models, centerlines, meshes, CFD analysis results, etc.), which helps promote standardization and reproducibility of end-to-end research. 2. By establishing a CFD usability evaluation system and the CFD-AS metric, the study quantifies the “CFD usability” of segmentation results, bridging the gap between common segmentation metrics (e.g., Dice)
1. Lacks systematic analysis of the coupling between Stage I and Stage II and of the feasibility of end-to-end training; there is no quantitative assessment of error propagation or robustness between the two stages. 2. Although topological constraint losses (e.g., clDice) show improvements in vascular topology, the paper lacks detailed robustness analysis and quantitative results under different topological abnormalities (adhesion, distal branch misalignment, branch disconnection, etc.). Overemp
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsIntracranial Aneurysms: Treatment and Complications · Cerebrospinal fluid and hydrocephalus · Medical Image Segmentation Techniques