Rapid whole-heart CMR with single volume super-resolution

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Image Reconstruction

Background:

• Whole heart balanced steady state free precession (WH-bSSFP) sequences are crucial for cardiac and vascular anatomy visualization.
• Long acquisition times limit the clinical utility of WH-bSSFP sequences.
• Deep learning offers a potential solution for accelerating image acquisition.

Purpose of the Study:

• To develop and validate a deep-learning-based super-resolution technique for rapid acquisition of WH-bSSFP images.
• To assess the impact of super-resolution on image quality, quantitative measurements, and diagnostic confidence.
• To evaluate the clinical feasibility of this accelerated imaging approach.

Main Methods:

• A 3D residual U-Net was trained using synthetic low-resolution WH-bSSFP data derived from high-resolution images.
• The network was validated on synthetic datasets and prospectively acquired data from 40 patients.
• Image quality, vessel diameter measurements, and diagnostic scoring were compared between low-resolution, super-resolution, and high-resolution images.

Main Results:

• Super-resolution significantly improved image quality, edge sharpness, and signal-to-noise ratio compared to low-resolution images.
• Acquisition time was reduced by approximately 3-fold with super-resolution reconstruction (<1 second per volume).
• Super-resolution measurements of great vessel diameters were accurate, with minor underestimation in the proximal left coronary artery.

Conclusions:

• Deep learning super-resolution can effectively reconstruct high-quality WH-bSSFP images from rapid acquisitions.
• The technique enhances image quality and diagnostic confidence, offering a promising solution for faster cardiac MRI.
• This approach has the potential to significantly expedite whole-heart cardiovascular magnetic resonance imaging in clinical settings.