Quantifying physiological variability and improving reproducibility in 4D-flow MRI cerebrovascular measurements with...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• 4D-Flow MRI is crucial for assessing cerebrovascular hemodynamics.
• Measurement variability in 4D-Flow MRI can stem from technical noise and physiological variations.
• Reducing variability is key to improving the reliability of hemodynamic assessments.

Purpose of the Study:

• To evaluate the effectiveness of self-supervised deep learning (DL) denoising in minimizing measurement variability in 4D-Flow MRI.
• To differentiate the impact of technical noise versus physiological variations on cerebrovascular hemodynamics.

Main Methods:

• A self-supervised DL framework was developed and trained using 3D radially sampled 4D-Flow MRI data.
• A prospective test-retest study with 10 participants was conducted, including back-to-back and interleaved scans.
• DL denoising efficacy was assessed by comparing velocity and hemodynamic metrics pre- and post-denoising.

Main Results:

• DL denoising significantly improved the reproducibility of 4D-Flow MRI measurements, reducing confidence intervals for velocity.
• Noise-sensitive measures like pulsatility index showed significant improvement post-DL denoising.
• While DL reduced technical noise, it did not eliminate variations arising from physiology or post-processing.

Conclusions:

• Self-supervised DL denoising enhances the quantitative repeatability of 4D-Flow MRI by mitigating technical noise.
• Physiological variations and post-processing artifacts are not eliminated by DL denoising.
• Accurate neurovascular flow imaging requires consideration of both technical and physiological variability, especially for biomarker development.