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

Cardiac cine MRI — the clinical standard for measuring ventricular volumes and function — typically requires multiple sustained breath-holds. This study evaluated whether deep learning-based reconstruction could shorten acquisition to three or six heartbeats while maintaining diagnostic accuracy.

In 55 patients, three-heartbeat (3RR) and six-heartbeat (6RR) sequences matched the standard protocol for volumetric accuracy and image quality despite significantly shorter acquisition times. The deep learning model filled in undersampled k-space data to produce diagnostically equivalent images.

Faster cardiac MRI matters clinically: it reduces motion artifact, improves patient tolerance, and makes the technique more accessible for patients who cannot sustain reliable breath-holds — including those with heart failure or advanced age. The 3RR protocol emerges as the optimal balance between speed and image quality.

The limitation is sample size and single-center design. Fifty-five patients cannot validate a technique for broad clinical adoption, and performance may vary with different scanner hardware and patient populations.

We rate the evidence limited. A small but technically sound feasibility study demonstrating that DL-accelerated cine MRI can match standard acquisitions; multicenter validation at scale is needed before broad implementation.