Focus Session
SCMR 22nd Annual Scientific Sessions
Rui Li, PhD
Associate Professor
Tsinghua University
Aiqi Sun, PhD
Student
Tsinghua University
Bo Zhao, PhD
Postdoc
Massachusetts General Hospital
Yichen Zheng, MSc
Student
Tsinghua University
Chun Yuan, PhD
Professor
University of Washington
Background: Real-time flow MRI is a promising technique that overcomes many limitations of conventional cine flow imaging. However, in real-time imaging, (k, t)-space is highly undersampled, which often leads to a challenging image reconstruction problem. The feasibility of real-time imaging has only been demonstrated in two-dimensional (2D) flow imaging applications (i.e., with 2D spatial encoding and one-directional flow encoding) [1]. In this work, we develop a subspace imaging [2] approach to enable real-time 4D flow MRI[3], which measures blood flow velocities with three-directional flow encoding and 3D anatomical coverage without electrocardiogram (ECG) synchronization and respiration control.
Methods: The proposed approach formulates the flow imaging reconstruction problem as a low-rank matrix recovery problem, exploiting strong correlation of the data in the spatial, temporal, and flow encoding dimensions. It then uses the low-rank/subspace model to dictate both data acquisition and image reconstruction for real-time flow imaging. For data acquisition, it acquires two sets of (k, t)-space data, one with high temporal resolution and the other with high spatial resolution. For image reconstruction, it first determines the temporal subspace from the data with high temporal resolution, and then determines the spatial subspace from the remaining data. Together, it reconstructs high-resolution time-series images, from which the velocity information is estimated. The performance of the proposed approach was evaluated and validated in real-time 4D flow experiments of the aorta, in which ten healthy subjects and one patient with cardiac arrhythmia were recruited.
Results:
The proposed approach enables real-time 4D flow imaging with an isotropic spatial resolution of 2.4mm and a temporal resolution of 35.2ms in an acquisition time under 17 minutes. It reconstructs velocity maps with comparable velocity-to-noise ratio to the conventional cine flow imaging (figure 1). For the experiments on healthy subjects, it provides flow measurements that are in good agreement with those from the conventional approach (figure 2). For the experiments on the arrhythmic patient, it well resolves beat-by-beat pathological variations (figure 3) that cannot be obtained from the conventional approach.
Conclusion: We demonstrate, for the first time, the feasibility of real-time 4D flow MRI. The proposed approach provides high-resolution velocity measurements, and well resolves beat-by-beat flow variations.