TL;DR
This paper introduces Motion-Corrected CINE (MC-CINE), a novel reconstruction technique that enables high-quality cardiac cine imaging from a single heartbeat. By incorporating nonrigid cardiac motion correction directly into the reconstruction process and using a motion-aligned patch-based regularization, MC-CINE significantly improves image sharpness and quality compared to existing methods, paving the way for faster, more comfortable cardiac MRI scans.
What Problem Does This Paper Address?
Traditional cardiac cine MRI, a crucial tool for assessing heart function and morphology, typically requires acquiring data over multiple heartbeats. This is because capturing enough spatial and temporal information to reconstruct detailed images of the beating heart at the necessary resolution demands a significant amount of data. While retrospective electrocardiogram (ECG) gating allows for the grouping of data into different cardiac phases, this multi-heartbeat approach leads to longer scan times, which can be challenging for patients, especially those who have difficulty holding their breath. Furthermore, the assumption of cardiac motion periodicity can be problematic for patients with irregular heartbeats. This paper aims to overcome these limitations by developing a method that can produce high-quality cardiac cine images from data acquired within a single heartbeat.
Methods
The core innovation of MC-CINE lies in its reconstruction framework, which integrates cardiac motion correction directly into the image reconstruction process. This is achieved through a four-step pipeline:
- ECG Binning: Acquired data (using a tiny golden radial trajectory) are retrospectively binned into distinct cardiac phases, guided by ECG signals.
- Auxiliary Reconstruction: An initial, lower-quality cardiac-resolved reconstruction is performed using a standard method like XD-GRASP. This serves as a basis for motion estimation.
- Cardiac Motion Estimation: Image registration techniques are applied to the auxiliary reconstructions to estimate nonrigid motion fields between different cardiac phases.
- MC-CINE Reconstruction: The estimated motion fields are incorporated into the forward model of the reconstruction. This allows the reconstruction of each cardiac phase to utilize all acquired data, effectively correcting for motion. Additionally, a novel regularization strategy, a motion-aligned patch-based denoiser, is employed to further enhance image quality by exploiting structural similarities in motion-warped image patches. This approach is formulated as a jointly regularized optimization problem solved using the Alternating Direction Method of Multipliers (ADMM).
Datasets
The study evaluated the MC-CINE approach on data acquired from 14 healthy human subjects. The paper does not introduce a new benchmark dataset but rather compares its method against existing techniques using data from these subjects.
Key Insights and Ideas
- Motion Correction within Reconstruction: The most significant insight is the successful integration of nonrigid cardiac motion correction directly into the reconstruction process, rather than treating it as a post-processing step or solely as a regularization constraint. This fundamentally changes how the problem is posed, making it better conditioned.
- Single-Heartbeat Feasibility: The paper demonstrates that high-quality cardiac cine imaging is achievable from a single heartbeat, which is a substantial leap towards faster and more patient-friendly MRI.
- Motion-Aligned Regularization: The use of a patch-based regularizer that is aligned with the estimated cardiac motion is crucial. This allows the regularization to exploit redundancies more effectively by comparing similar structures across cardiac phases that have been "warped" to a common motion state.
- Robustness to Irregular Heartbeats: By resolving each cardiac phase temporally, MC-CINE is inherently less susceptible to variations in heart rate or arrhythmias, which can be a significant issue for multi-heartbeat methods.
Key Results
- Superior Image Quality: MC-CINE consistently outperformed both iterative SENSE (itSENSE) and XD-GRASP in terms of image sharpness, reader scores, and reader rankings, even at high acceleration factors (equivalent to one heartbeat).
- High Acceleration Performance: At an acceleration factor of R=26.9 (approximately one heartbeat), MC-CINE achieved image quality comparable to itSENSE acquired with a fully sampled dataset (R=1.3, ~20 heartbeats).
- Sharpness Improvement: MC-CINE achieved an average LV sharpness of 82.5% at R=1.3, 83.2% at R=13.5, and 82.1% at R=26.9, significantly outperforming itSENSE (74.6%, 57.0%, 53.2%) and XD-GRASP (74.2%, 72.0%, 65.8%) at higher accelerations.
- Reader Scores: Reader scores for MC-CINE at R=26.9 were 3.9 (out of 5), comparable to itSENSE at R=1.3 (4.0). Reader rankings also showed MC-CINE's superiority, with R=26.9 achieving a rank of 5.4, similar to itSENSE at R=1.3 (5.3).
- Quantitative Metrics: While there were small, nonsignificant biases in end-systolic volume (ESV), end-diastolic volume (EDV), and ejection fraction (EF) compared to itSENSE, MC-CINE demonstrated comparable accuracy to XD-GRASP and no significant differences were observed for these metrics across methods.
Limitations and Future Work
- Computational Cost: The current implementation of MC-CINE is computationally intensive, with motion estimation and reconstruction being the most time-consuming steps. The authors suggest that recent advancements in deep learning for motion estimation and reconstruction could significantly reduce this burden.
- Healthy Subject Cohort: The study was conducted on young, healthy subjects. Future work should validate MC-CINE in patient populations, who may exhibit more complex motion patterns and lower image quality.
- Slice Orientation: The study focused on midventricular short-axis slices. Validation in other orientations (e.g., two-chamber, four-chamber) is needed.
- Prospective Acquisition: The study used retrospective undersampling. Prospective implementation, especially considering the initial transient state of magnetization, requires further investigation.
- Through-Plane Motion: The current 2D framework does not account for through-plane motion, which could be addressed by extending to a 3D framework.
- ECG-less Acquisition: While MC-CINE can resolve cardiac phases temporally, enabling ECG-less acquisition, this was not explicitly tested.
- Free-Breathing: The framework could be extended to free-breathing acquisitions by incorporating respiratory motion correction.
Summary updated