Rapid wall shear stress prediction for aortic aneurysms using deep learning: a fast alternative to CFD

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Computational Fluid Dynamics

Background:

• Aortic aneurysms carry a high risk of rupture, often associated with low wall shear stress (WSS) regions.
• Accurate WSS assessment is vital for predicting abdominal aortic aneurysm (AAA) rupture.
• Computational fluid dynamics (CFD) methods for WSS calculation are accurate but computationally demanding, limiting clinical application.

Purpose of the Study:

• To develop a rapid and accurate deep learning (DL) surrogate model, MultiViewUNet, for predicting time-averaged WSS (TAWSS) distributions in AAAs.
• To overcome the computational intensity of traditional CFD methods for hemodynamic analysis in AAAs.
• To enable faster and more efficient WSS quantification for clinical decision-making.

Main Methods:

• A deep learning surrogate model, MultiViewUNet, was developed for TAWSS prediction.
• A domain transformation technique was utilized to adapt complex aortic geometries for neural network input.
• The MultiViewUNet model was trained using a dataset comprising real and synthetic AAA geometries.

Main Results:

• The MultiViewUNet model achieved a low average normalized mean absolute error (NMAE) of just 0.045 in TAWSS prediction.
• The DL surrogate demonstrated high accuracy in predicting TAWSS distributions on AAA geometries.
• The model successfully translated complex 3D aortic shapes into a format suitable for neural network analysis.

Conclusions:

• The MultiViewUNet framework offers a computationally efficient alternative to traditional CFD for hemodynamic analysis in AAAs.
• This DL approach has the potential to significantly streamline WSS quantification for clinical risk assessment.
• The developed method could be applicable to other clinical scenarios requiring rapid and precise stress analysis.