Enhancement of cerebrovascular 4D flow MRI velocity fields using machine learning and computational fluid dynamics...

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Fluid Dynamics

Background:

• Four-dimensional (4D) flow phase contrast (PC) magnetic resonance imaging (MRI) is valuable for cerebrovascular analysis.
• PC MRI is prone to inherent errors, limiting quantitative and qualitative analysis accuracy.
• Computational fluid dynamics (CFD) generates accurate, physics-based velocity fields.

Purpose of the Study:

• To investigate the potential of augmenting cerebral 4D flow MRI data with CFD-informed neural networks.
• To develop a method for producing highly accurate physiological flow fields in cerebral blood vessels.
• To enhance MRI-derived velocity fields using CFD-trained convolutional neural networks.

Main Methods:

• Trained a convolutional neural network using high-resolution, patient-specific CFD data.
• Utilized the trained network to enhance MRI-derived velocity fields in cerebral blood vessel datasets.
• Tested the method on simulated images, phantom data, and 4D flow data from 20 patients.

Main Results:

• The trained neural network successfully de-noised 4D flow MRI images.
• Demonstrated a decrease in velocity error in the enhanced flow fields.
• Improved near-vessel-wall velocity quantification and visualization.

Conclusions:

• CFD-informed neural networks show significant potential for enhancing cerebrovascular 4D flow MRI.
• This approach can improve both qualitative and quantitative analysis in clinical and experimental settings.
• The method offers a pathway to more accurate physiological flow field assessment.