Neural Fields for Highly Accelerated 2D Cine Phase Contrast MRI.

Area of Science:

• Medical Imaging
• Biophysics
• Computational Science

Background:

• 2D cine phase contrast (CPC) MRI is crucial for quantifying blood flow but suffers from long acquisition times.
• Accelerating MRI acquisition is essential for clinical feasibility and patient comfort.
• Current reconstruction methods for undersampled MRI data often lead to oversmoothing and reduced accuracy.

Purpose of the Study:

• To develop a novel method using neural fields for reconstructing velocity fields from undersampled 2D CPC MRI data.
• To enable accurate blood flow quantification and anatomical depiction with significantly reduced scan times.
• To address the oversmoothing issue in neural field reconstructions under severe undersampling.

Main Methods:

• Proposed neural fields as a continuous spatiotemporal parametrization for complex-valued MRI images.
• Jointly modeled magnitude and phase across multiple echoes for velocity estimation.
• Introduced a voxel-based postprocessing step to mitigate oversmoothing.
• Validated the method numerically using Cartesian and radial k-space data with varying temporal resolutions.

Main Results:

• Achieved accurate velocity field reconstructions at high acceleration factors (e.g., 32x, 64x undersampling for high temporal resolution).
• Demonstrated low reconstruction errors even with severe undersampling.
• Consistently outperformed classical locally low-rank regularized voxel-based methods.
• Showcased superior performance in both flow estimates and anatomical depiction.

Conclusions:

• Neural fields offer a promising approach for accelerated 2D CPC MRI.
• The proposed method effectively reconstructs blood flow data, improving efficiency and accuracy.
• This technique has the potential to enhance clinical applications of quantitative flow MRI.