Real-Time Phase-Contrast Cardiovascular MRI Using a Deep Learning Reconstruction Network With Combined Dictionary...

Area of Science:

• Cardiovascular Imaging
• Medical Physics
• Artificial Intelligence in Medicine

Background:

• Real-time cardiovascular MRI is crucial for dynamic blood flow assessment.
• Current methods often face limitations in speed and latency, hindering clinical application.
• Deep learning offers potential for accelerated image reconstruction in MRI.

Purpose of the Study:

• To develop a real-time phase-contrast (PC) cardiovascular MRI framework with low latency.
• To enhance image reconstruction quality and flow quantification accuracy.
• To demonstrate the feasibility of real-time cardiovascular MRI for clinical use.

Main Methods:

• A golden-angle radial sequence combined with a deep-learning reconstruction network (DLCNet) was employed.
• DLCNet integrates dictionary learning and CNNs to capture spatiotemporal features.
• The framework was trained and tested on 15 healthy subjects using the Gadgetron platform.

Main Results:

• The proposed DLCNet significantly outperformed other reconstruction algorithms in image quality and flow quantification.
• The system achieved a high imaging speed of 14.6 frames per second with <60ms display latency.
• Real-time flow quantification demonstrated strong agreement with conventional ECG-gated, breath-hold PC-MRI.

Conclusions:

• The developed framework successfully enables real-time PC cardiovascular MRI.
• High-quality image reconstruction and low-latency display were achieved.
• This technology holds promise for improved cardiovascular diagnostics.