What question did this study set out to answer?

This research aims to develop a physics-informed graph neural network for predicting wall shear stress in coronary arteries affected by stenosis.

April 8, 2026Scientific ReportsOpen Access

Physics-informed graph neural networks for real-time prediction of wall shear stress in stenotic coronary arteries

View Full Paper

Ask AI

Bookmark

Authors

Research demonstrates real-time prediction of wall shear stress in stenotic coronary arteries, implying improved clinical decision-making.

This research aims to develop a physics-informed graph neural network for predicting wall shear stress in coronary arteries affected by stenosis.
Developed a physics-informed graph neural network (PI-GNN) for prediction.
Utilized 40 subject-specific coronary artery geometries from CT angiography.
Created 1000 synthetic models using statistical shape modeling.
Performed full computational fluid dynamics (CFD) simulations for ground-truth data.
Conducted node-wise Bland–Altman analysis for local agreement with CFD.
PI-GNN achieved superior global performance with an R of 0.94 and MAE of 1.05 Pa.
Outperformed U-Net with an R of 0.85 and multilayer perceptron with an R of 0.24.
Demonstrated negligible mean bias of |bias| < 2 Pa in node-wise analysis.
Inference times were reduced to seconds for real-time applications.

Luo et al. (2026) studied this question.

View Full Paper

Ask AI

Bookmark

April 8, 2026Scientific ReportsOpen Access

View Full Paper

Ask AI

Bookmark

Authors

Research demonstrates real-time prediction of wall shear stress in stenotic coronary arteries, implying improved clinical decision-making.

This research aims to develop a physics-informed graph neural network for predicting wall shear stress in coronary arteries affected by stenosis.
Developed a physics-informed graph neural network (PI-GNN) for prediction.
Utilized 40 subject-specific coronary artery geometries from CT angiography.
Created 1000 synthetic models using statistical shape modeling.
Performed full computational fluid dynamics (CFD) simulations for ground-truth data.
Conducted node-wise Bland–Altman analysis for local agreement with CFD.
PI-GNN achieved superior global performance with an R of 0.94 and MAE of 1.05 Pa.
Outperformed U-Net with an R of 0.85 and multilayer perceptron with an R of 0.24.
Demonstrated negligible mean bias of |bias| < 2 Pa in node-wise analysis.
Inference times were reduced to seconds for real-time applications.

Luo et al. (2026) studied this question.

View Full Paper

Ask AI

Bookmark