Synopsis

Free-breathing continuous acquisitions, so called free-running, enable 3D whole-heart coverage for motion-resolved functional cardiac MRI. In prior work approaches based on 3D radial imaging were proposed for this task. However, free-running 3D radial image reconstruction is computationally demanding. In this work, we propose a novel 3D Cartesian free-running bSSFP sequence. Data is acquired continuously with a variable-density Cartesian trajectory with spiral profile order and retrospectively binned based on respiratory self-navigation and cardiac ECG signal synchronization. A multi-bin-PROST reconstruction is proposed to exploit spatial, cardiac and respiratory temporal redundancies to reconstruct high-resolution images.

Introduction

3D radial^1-4 free-breathing continuous acquisitions, so called free-running, have been recently proposed to reconstruct 3D whole-heart images at different cardiac and respiratory phases. These approaches retrospectively assign the data into different cardiac and respiratory phases using ECG or self-extracted cardiac and/or respiratory signals. Cardiac and respiratory resolved images are then reconstructed exploiting temporal redundancy in both cardiac and respiratory directions⁵. These approaches have shown promising image quality in reasonable scan times ~10-12min. Those methods^1-7 however exhibit several drawbacks including long reconstruction times due to the non-Cartesian nature of the acquisition.

Here, we propose a 3D Cartesian free-running bSSFP acquisition which enables cardiac and respiratory resolved images of the whole-heart without contrast administration in a clinically feasible scan time and with adequate reconstruction times. A variable-density Cartesian sampling with spiral order (VD-CASPR)^8,9 is extended to enable uniform retrospective sample-to-motion state binning. An embedded self-navigation signal captures the respiratory motion. A motion-resolved low-rank patch-based multi-bin-PROST reconstruction is proposed to exploit spatial, cardiac and respiratory temporal redundancies and to reconstruct high quality images.

Methods

Acquisition: The proposed acquisition and reconstruction framework is depicted in Fig.1. In order to minimize eddy current artefacts, a VD-CASPR sampling with linear angle increment $$$\varphi$$$ and an outward-in trajectory between spirals continuously subsamples the Cartesian k_y/k_z plane by an acceleration factor L. Depending on L, matrix size and number of spiral rings, after S spirals a fully sampled k-space is reached, i.e. all high-frequency points are sampled at least once whilst the low-frequency range is oversampled. This property originates from the variable-density sampling and is desirable to enable low-frequency high-contrast sampling more frequently. The trajectory provides incoherent subsampling artifacts for the different motion-resolved images. Data is continuously acquired for a given desired acquisition time. The center line of k-space is periodically sampled serving as a 1D respiratory self-navigator. An α/2 preparation RF pulse and dummy pre-pulses bring the magnetization into bSSFP steady-state before data acquisition.

Reconstruction: A bandpass-filtered PCA extracts the respiratory-induced displacement of the heart from the 1D self-navigator. A retrospective dual gating bins the data into respiratory and cardiac states based on the respiratory self-navigation signal and ECG. Cardiac bins are adaptively adjusted per RR-interval to the current heart-rate to increase motion state consistency over all cardiac cycles. A parametrizable view-sharing amongst neighboring gates and a low-frequency view-sharing ensure good initialization for the subsequent reconstruction. The subsampled 6D k-space (3D spatial + 1 cardiac + 1 respiratory + Rx channels) is reconstructed extending PROST⁸ to a multi-bin-PROST for cardiac and respiratory motion-resolved reconstruction (Fig.2). PROST has been recently introduced to reconstruct subsampled MR images by exploiting the highly redundant information, on a local (similar patches within a neighborhood) and non-local (low rankness of all patches in the image) scale. Here the strong correlation shared between the multiple cardiac and respiratory images is also exploited by extending the search for similar patches to a spatial/cardiac-motion/respiratory-motion neighborhood. Multi-bin-PROST reconstruction iterates between an MR reconstruction optimization and an efficient patch-based denoising using an alternating direction method of multipliers.

Data was acquired on a 1.5T MR (MAGNETOM Aera, Siemens Healthineers, Erlangen, Germany) in five healthy subjects in coronal orientation with 1.4mm³ isotropic resolution, FOV=320x320x260mm, TE/TR=1.6/3.2ms, flip angle=60°, spiral rings/segments=30, acceleration L=3, bandwidth=880Hz/px, 200 pre-dummy pulses, self-navigation period = 96ms, acquisition time = 7min52s. For the multi-bin-PROST a patch size of 5x5x5px, search window = 60px, patch offset = 4px with 20 simultaneously selected patches was used.

Results and Discussion

Conclusion

Acknowledgements

References