Accelerated Cardiac MRI with Deep Learning-based Image Reconstruction for Cine Imaging

Area of Science:

• Cardiovascular Imaging
• Medical Physics
• Artificial Intelligence in Medicine

Background:

• Cardiac MRI (magnetic resonance imaging) is crucial for assessing heart function.
• Traditional cine sequences can be lengthy, impacting patient comfort and throughput.
• Deep learning (DL) offers potential for accelerating MRI acquisition and reconstruction.

Purpose of the Study:

• To evaluate the impact of DL-based reconstruction on cardiac MRI cine sequences.
• To compare acquisition time, image quality, and volumetric accuracy of DL cine sequences against standard methods.
• To determine the optimal DL cine acquisition strategy (1, 3, or 6 cardiac cycles).

Main Methods:

• Prospective study involving 55 healthy volunteers undergoing 1.5 T cardiac MRI.
• DL cine sequences (1, 3, 6 cardiac cycles) and a standard cine sequence (10-12 cycles) were acquired.
• Evaluated parameters included acquisition time, subjective image quality, edge sharpness, and left ventricular ejection fraction (LVEF).

Main Results:

• DL cine sequences markedly reduced acquisition time: 47s (1RR), 108s (3RR), 184s (6RR) vs. 227s (standard).
• LVEF was comparable for standard, 3RR, and 6RR DL cine sequences (63%, 61%, 62% median, respectively).
• 1RR DL cine underestimated LVEF (57%); 3RR DL cine showed similar image quality and sharpness to the standard sequence.

Conclusions:

• Deep learning-based cine sequences significantly shorten cardiac MRI acquisition times.
• A three-cardiac-cycle DL acquisition provides an optimal balance, matching image quality and volumetric accuracy of standard sequences.
• This approach offers over 50% time savings compared to conventional methods without compromising diagnostic utility.