RHSIA: Real-time Hemodynamics Surrogation for Non-idealized Intracranial Aneurysms

arXiv ID: 2601.19876v1

Published: 2026-01-27

Authors: Yiying Sheng, Wenhao Ding, Dylan Roi, Leonard Leong Litt Yeo, Hwa Liang Leo, Choon Hwai Yap

Categories: cs.LG

Relevance Score: 0.95 / 1.00

View on arXiv Download PDF

Summary

This paper introduces RHSIA, a Graph Transformer deep learning model that enables real-time prediction of temporal Wall Shear Stress (WSS) patterns for intracranial aneurysms (IAs) directly from surface meshes. This addresses the current clinical translation barrier of Computational Fluid Dynamics (CFD) due to its high cost and time. The model achieves high accuracy and effectively captures the pulsatile nature of hemodynamics, further enhanced by a novel data augmentation strategy using low-cost steady-state CFD data.

Medical Relevance

This research provides a pathway to obtain critical biomechanical markers like WSS in real time, accelerating the assessment of intracranial aneurysm progression risk. This enables faster clinical decision-making, improved patient stratification, and supports high-throughput analysis for clinical trials, overcoming a significant hurdle in translating fluid mechanics research into clinical practice.

AI Health Application

The AI application is a deep learning model (Graph Transformer) that enables real-time prediction of crucial cardiovascular fluid mechanical parameters (like Wall Shear Stress) from intracranial aneurysm morphology. This can assist physicians in obtaining biomechanical markers instantly, without complex Computational Fluid Dynamics (CFD) simulations, thereby improving diagnosis, risk assessment for aneurysm rupture, and treatment planning for patients with intracranial aneurysms. It also has potential applications in accelerating clinical trials and research by providing high-throughput analysis of biomechanical markers.

Key Points

Fluid mechanical markers from CFD are crucial for intracranial aneurysm (IA) progression risk assessment but are not clinically translated due to CFD's time-consuming nature, specialized expertise requirement, and low throughput.
The proposed solution is a Graph Transformer deep learning model, named RHSIA, that incorporates temporal information to predict Wall Shear Stress (WSS) across the cardiac cycle from IA surface meshes.
The model demonstrates high accuracy in predicting WSS, achieving a Structural Similarity Index (SSIM) of up to 0.981 and a maximum-based relative L2 error of 2.8%.
It effectively captures the complex temporal variations of the WSS pattern, crucial for understanding pulsatile blood flow dynamics in aneurysms.
A novel data augmentation strategy was employed, injecting a large amount of low-cost steady-state CFD data to substantially enhance network performance, especially when the sample size of computationally expensive pulsatile CFD data is limited.
Ablation studies and comparison with state-of-the-art methods confirmed the optimality and effectiveness of the Graph Transformer model.
The study provides a proof of concept for real-time computation of temporal sequences of cardiovascular fluid mechanical parameters using deep learning from geometric meshes, even with small pulsatile CFD datasets.

Methodology

The study utilized a Graph Transformer deep learning model, supervised by a large dataset of Computational Fluid Dynamics (CFD) simulations, to map IA surface meshes to pulsatile Wall Shear Stress (WSS) patterns. The model was specifically designed to incorporate temporal information to capture WSS variations across the cardiac cycle. To address the scarcity of expensive pulsatile CFD data, a data augmentation strategy was implemented by incorporating a large volume of low-cost steady-state CFD data. Model performance and optimality were validated through ablation studies and comparison with existing state-of-the-art techniques.

Key Findings

The Graph Transformer model accurately predicted temporal WSS patterns across the cardiac cycle for IAs, achieving an SSIM of up to 0.981 and a maximum-based relative L2 error of 2.8%. The model effectively captured the dynamic temporal variations of WSS patterns. Crucially, the innovative strategy of augmenting pulsatile CFD data with low-cost steady-state CFD data significantly enhanced the network's performance, particularly when the sample size of pulsatile data was small.

Clinical Impact

This technology can provide physicians with real-time access to crucial fluid mechanical markers (e.g., WSS) indicating intracranial aneurysm progression risk, eliminating the need for time-consuming and expertise-intensive CFD simulations. This real-time capability will enable faster diagnosis, more precise risk stratification, and personalized treatment planning. It also facilitates the feasibility of large-scale clinical trials by allowing high-throughput biomechanical assessment, potentially leading to improved patient outcomes and more effective management strategies for IAs.

Limitations

While the abstract highlights the limitations of traditional CFD, it does not explicitly state limitations for the proposed deep learning method. However, as the model is supervised by CFD data, its accuracy is inherently constrained by the fidelity and underlying assumptions of the CFD simulations used for training. Further, as a proof-of-concept, broader validation across diverse and heterogeneous clinical datasets would be necessary.

Future Directions

The authors suggest that their approach is likely applicable to other cardiovascular scenarios beyond intracranial aneurysms. Future research could involve extending the application of this real-time hemodynamics surrogation model to other vascular pathologies (e.g., carotid atherosclerosis, aortic aneurysms) and integrating it into clinical workflows for broader utility.

Medical Domains

Neurosurgery Cardiology Vascular Surgery Diagnostic Radiology Biomedical Engineering

Keywords

Intracranial Aneurysm Wall Shear Stress Computational Fluid Dynamics Deep Learning Graph Transformer Real-time Hemodynamics Data Augmentation Cardiovascular Fluid Mechanics

Abstract

Extensive studies suggested that fluid mechanical markers of intracranial aneurysms (IAs) derived from Computational Fluid Dynamics (CFD) can indicate disease progression risks, but to date this has not been translated clinically. This is because CFD requires specialized expertise and is time-consuming and low throughput, making it difficult to support clinical trials. A deep learning model that maps IA morphology to biomechanical markers can address this, enabling physicians to obtain these markers in real time without performing CFD. Here, we show that a Graph Transformer model that incorporates temporal information, which is supervised by large CFD data, can accurately predict Wall Shear Stress (WSS) across the cardiac cycle from IA surface meshes. The model effectively captures the temporal variations of the WSS pattern, achieving a Structural Similarity Index (SSIM) of up to 0.981 and a maximum-based relative L2 error of 2.8%. Ablation studies and SOTA comparison confirmed its optimality. Further, as pulsatile CFD data is computationally expensive to generate and sample sizes are limited, we engaged a strategy of injecting a large amount of steady-state CFD data, which are extremely low-cost to generate, as augmentation. This approach enhances network performance substantially when pulsatile CFD data sample size is small. Our study provides a proof of concept that temporal sequences cardiovascular fluid mechanical parameters can be computed in real time using a deep learning model from the geometric mesh, and this is achievable even with small pulsatile CFD sample size. Our approach is likely applicable to other cardiovascular scenarios.