Super-resolution 4D flow MRI to quantify aortic regurgitation using computational fluid dynamics and deep learning.

Area of Science:

• Cardiovascular Imaging and Hemodynamics
• Artificial Intelligence in Medical Diagnostics
• Biomedical Engineering

Background:

• Aortic regurgitation (AR) diagnosis and severity assessment rely on cardiovascular hemodynamics.
• Four-dimensional (4D) flow magnetic resonance imaging (MRI) offers non-invasive hemodynamic metrics.
• Current 4D flow MRI is limited by insufficient spatial resolution, hindering accurate AR analysis.

Purpose of the Study:

• To develop an advanced imaging technique for improved assessment of aortic regurgitation.
• To enhance the spatial resolution of 4D flow MRI data for more accurate hemodynamic analysis.
• To leverage computational fluid dynamics and neural networks for super-resolution imaging.

Main Methods:

• Computational fluid dynamics (CFD) simulations generated synthetic 4D flow MRI data.
• Neural networks were trained on synthetic data to achieve super-resolution imaging (upsample factor of 4).
• The developed method was validated using in vivo 4D flow MRI datasets.

Main Results:

• The super-resolution networks significantly reduced velocity errors in flow images.
• High structural similarity scores indicated excellent preservation of image integrity.
• Validation demonstrated successful de-noising and improved image quality for in vivo data.

Conclusions:

• This AI-driven super-resolution approach enhances 4D flow MRI capabilities for AR analysis.
• The method offers a pathway for more comprehensive and non-invasive evaluation of AR hemodynamics.
• Improved imaging resolution facilitates better understanding and management of valvular heart disease.