Abstract
Evaluation of cardiac function is clinically performed using multi-slice breath-hold (BH) acquisition, in which patient breathes for a period of 40-60 seconds between a BH of 10-12 seconds. This results in suboptimal data acquisition efficiency. In this study, we propose to take advantage of the time between multiple BHs to acquire additional free-breathing (FB) data that can be used in the reconstruction of undersampled BH acquisition without increasing the total scan time.
Introduction
Evaluating cardiac function via MRI typically requires patients to hold their breath (BH) for 10–12 seconds to avoid motion blur, followed by a 40–60 second recovery period. This cycle is repeated for multiple "slices" of the heart. This traditional approach is highly inefficient; the scanner sits idle for roughly 80% of the total exam time while the patient rests. To speed up the process, clinicians often "undersample" the data (taking fewer measurements), but this usually results in "streaking" artifacts and blurry images. This paper proposes a novel strategy to utilize that idle resting time to improve the quality of fast, undersampled scans.
Proposed Methodology
The authors introduced a novel data acquisition and reconstruction strategy that turns "waste" time into valuable prior knowledge. During the standard breath-hold, undersampled radial cine data is acquired. During the subsequent resting period, instead of stopping the scanner, fully-sampled "free-breathing" (FB) data is acquired while the patient breathes normally.
The FB data is reconstructed using a standard gridding algorithm to create a reference image ($m_{FB,i}$) for each phase of the cardiac cycle. This reference is then used in a Compressed Sensing (CS) algorithm that minimizes a specific objective function. The key innovation is the inclusion of a "difference" term: $\|m_i - m_{FB,i}\|_1$. While standard CS relies on the image being sparse in a transform domain, this method assumes that the difference between the breath-hold image and the free-breathing image is even sparser than the image itself. By subtracting the FB reference, the algorithm can more effectively "fill in the blanks" of the undersampled BH data.
Results and Performance
The proposed method was compared against conventional gridding and standard CS algorithms (utilizing x-f space sparsity) using data from a 1.5T Philips scanner. BH data was retrospectively undersampled to 33% and 20% of the full radial projections to test performance under high acceleration.
Visual analysis demonstrated that the proposed method showed significantly fewer streaking artifacts compared to conventional CS. There were noticeably sharper borders between the blood and the myocardium. Notably, the method remained robust even when the breath-hold data was undersampled to just 20% (a 5x acceleration), suggesting that breath-hold durations could be significantly shortened or spatial resolution could be increased without sacrificing image quality.
Discussion and Future Directions
The most significant insight of this work is that motion-corrupted free-breathing data contains enough structural information to serve as a powerful "prior" for reconstruction. Because the FB data is collected during the time the patient must rest anyway, this improvement in image quality comes with zero cost in terms of total scan duration.
However, the authors note that further studies are needed to evaluate global cardiac function (like ejection fraction) in a larger patient cohort to ensure clinical accuracy. Additionally, while the FB data provides a good reference, significant changes in heart rate or position between the BH and FB states could potentially introduce errors in the subtraction process. Future work will focus on optimizing the amount of FB data needed and further refining the motion robustness of the workflow.
Summary updated