Highly accelerated free-breathing real-time phase contrast cardiovascular MRI via complex-difference deep learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Cardiovascular Imaging

Background:

• Real-time phase contrast (PC) MRI is crucial for assessing cardiovascular hemodynamics.
• Current methods face limitations in speed and artifact reduction, hindering clinical application.
• Deep learning (DL) offers potential for improving MRI acquisition and reconstruction.

Purpose of the Study:

• To develop and evaluate a novel real-time PC-MRI protocol using a complex-difference deep learning (DL) framework.
• To assess the framework's ability to filter artifacts and accelerate image acquisition.
• To compare the DL-based method with existing reconstruction techniques for hemodynamic parameter quantification.

Main Methods:

• A DL framework with two 3D U-nets was employed to filter aliasing artifacts from real-time PC images.
• The U-nets were trained using synthetic real-time PC data derived from ECG-gated, segmented PC acquisitions of 510 patients.
• Prospective real-time PC data (acceleration rate = 28.8) and ECG-gated segmented PC data (acceleration rate = 2) were acquired in 21 patients for comparison.

Main Results:

• The DL-filtered synthetic real-time PC demonstrated strong correlation (R > 0.98) and good agreement with ECG-gated segmented PC for hemodynamic parameters (CO, SV, peak mean velocity).
• The DL framework achieved a filtering speed of 0.39 s/frame, which was 4.6 times faster than compressed sensing (CS).
• DL showed superior correlation and tighter limits of agreement compared to CS and gridding reconstruction, with specific advantages in peak mean velocity accuracy.

Conclusions:

• The complex-difference DL framework significantly accelerates real-time PC-MRI, achieving an acceleration factor of nearly 28-fold.
• This advancement enables rapid, free-running real-time assessment of flow hemodynamics.
• The DL approach shows promise for improving the efficiency and accuracy of cardiovascular flow imaging.