Qualitative and Quantitative Evaluation of a Deep Learning-Based Reconstruction for Accelerated Cardiac Cine Imaging

Area of Science:

• Cardiovascular Magnetic Resonance Imaging
• Artificial Intelligence in Medical Imaging

Background:

• Two-dimensional (2D) cine imaging is crucial for cardiac MRI (CMR) but requires prolonged breath-holds and consistent heartbeats.
• Patient limitations in breath-holding and irregular heart rhythms pose challenges for traditional cine acquisition.
• Accelerated imaging techniques are needed to improve efficiency and patient comfort in CMR.

Purpose of the Study:

• To systematically evaluate the performance of a deep learning-based reconstruction technique (Sonic DL Cine) for accelerated cardiac cine acquisition.
• To assess image quality, spatiotemporal sharpness, and biventricular cardiac function using Sonic DL Cine at various acceleration factors.
• To compare the deep learning-reconstructed images against fully sampled reference images.

Main Methods:

• Retrospective experiments using a digital cardiac torso phantom and healthy volunteers.
• Evaluation of Sonic DL Cine with accelerations ranging from 4-fold to 12-fold.
• Qualitative and quantitative comparisons of image quality, sharpness, and cardiac function metrics.

Main Results:

• Sonic DL Cine successfully accelerated cine acquisitions up to 12-fold.
• Preserved signal-to-noise ratio (SNR), contrast, and spatiotemporal sharpness comparable to fully sampled images.
• Cardiac function measurements from accelerated scans closely aligned with reference data.

Conclusions:

• Sonic DL Cine enables highly accelerated (up to 12-fold) cardiac cine MRI acquisition.
• The method preserves essential image characteristics (SNR, contrast, sharpness) and quantification accuracy.
• Sonic DL Cine offers a feasible approach for efficient and reliable cardiac function assessment.