1963
Accelerated Carotid 4D flow MRI with Multicontrast HD-PROST Reconstruction1School of Biomedical Engineering & Imaging Sciences, King's College London, London, United Kingdom, 2ISR-Lisboa/LARSyS and Department of Bioengineering, Instituto Superior Técnico – Universidade de Lisboa, Lisbon, Portugal, 3MR Research Collaborations, Siemens Healthcare Limited, Frimely, United Kingdom, 4Hospital da Luz Learning Health, Lisbon, Portugal
Synopsis
4D flow MRI is time consuming, since it requires the acquisition of time-resolved images with three-directional velocity encoding. Undersampled reconstruction techniques have been proposed to accelerate 4D flow carotid imaging, however scan time remains lengthily for high-resolution acquisitions. In this work, we propose to further accelerate 4D flow carotid MR imaging by exploiting patch-based similarities in local, non-local and multi-contrast dimensions with high-dimensional patch-based undersampled reconstruction (HD-PROST). The results show similar velocities for both k-t SENSE and the proposed HD-PROST, however higher precision was obtained with HD-PROST.
Introduction
Four-dimensional (4D) flow MRI is an emerging technique for quantitative flow assessment and visualization of complex blood flow patterns. The noninvasive assessment of carotid flow can provide clinically relevant information about the severity and progression of carotid artery stenosis and cerebral circulation [1]. Unfortunately, 4D flow MRI is time consuming, since it requires the acquisition of time-resolved images with three-directional velocity encoding. Undersampled reconstruction techniques have been proposed to accelerate 4D flow carotid imaging [2], however scan time remains lengthy for high-resolution acquisitions. In this work, we propose to further accelerate 4D flow carotid MR imaging by combining a highly undersampled acquisition with a novel high-order multi-contrast patch-based reconstruction. This is achieved using a recently introduced high-dimensional patch-based undersampled reconstruction (HD-PROST) [3] to jointly reconstruct all the velocity encoded and time-resolved images simultaneously. The feasibility of the proposed method was evaluated in three healthy subjects.Methods
HD-PROST has been recently introduced [3] to jointly reconstruct multi-contrast MR images by exploiting the high redundancy in information, on local (similar patches within a neighborhood) and non-local (between patches) scales, and the strong correlation shared between the multiple contrast images. For 4D flow carotid imaging all the cardiac phases for the different velocity encodings (VE) can be considered as different contrasts, and thus correlation in that direction can be exploited for joint reconstruction. This joint multi-contrast undersampled reconstruction can be formulated as the following joint optimization problem:
$$\mathcal{L}(X,\mathcal{T}):=\underset{X,\mathcal{T}}{\arg\min} \frac{1}{2}||EX-Y ||^2_F + \sum_{p} \lambda_p ||\mathcal{T}_p||_* \quad s.t.\mathcal{T}_p = R_p(X) $$
Where $$$Y$$$ denotes the acquired undersampled data, $$$E$$$ is the encoding operator (including coil sensitivities, Fourier operator, and sampling), and $$$X$$$ denotes the multi-contrast MR images to reconstruct.
The operator $$$R_p(.)$$$ constructs a third order tensor of similar 3D patches from the patch $$$p$$$ centered at pixel $$$p$$$ (see optimization 2). The nuclear norm is used to enforce multi-dimensional low-rank on a multi-contrast patch scale and $$$\lambda_p > 0$$$ controls the strength of sparsity. This optimization problem can be solved via alternating direction method of multipliers (ADMM) by iteratively solving the two following sub-problems:
Optimization 1: A regularized multi-contrast MR reconstruction (optimization on $$$X$$$) is performed using the multi-contrast denoised volume ($$$\mathcal{T}$$$) obtained from optimization 2 as prior knowledge. This optimization is solved using the conjugate gradient (CG) algorithm.
Optimization 2: A high-order denoising on a patch level (optimization on $$$\mathcal{T}_p$$$) is performed by building a 3D tensor from similar 3D+L patches (L being the concatenation of velocity encodings and cardiac phases) and applying a high-order singular value thresholding. The denoised images are then obtained by patch aggregation and used as prior in optimization 1 (Fig.2).
Experiments:
4D flow MRI acquisition was performed free-running (with retrospective ECG synchronization) in three healthy subjects. Acquisitions were performed on a 1.5T scanner (Siemens Magnetom Aera). VE with minimal TE was applied to obtain a reference image and encoding in three directions (x,y,z) [4]. A free-running variable density Cartesian sampling with spiral-like ordering and golden angle step was employed (Fig.1) [3,5]. Acquisitions were performed with a 3D gradient echo sequence using undersampling factors of 2 (17:43 min), 4 (8:52 min), and 8 (4:35 min). Acquisition parameters included: FOV = 200 × 214 × 45 mm3; resolution 0.9 × 0.9 ×1.5 mm3; TR/TE =9.1/4.55ms; FA=15º, and maximum VE = 100cm/s.
Two frameworks were compared to reconstruct 10 cardiac phases: iterative kt-SENSE with TV constraint in the time direction [6], and the proposed HD-PROST. In the first approach each of the three VE and reference are reconstructed separately, while HD-PROST takes advantage of the redundant information across cardiac phases and VE contrasts to reconstruct them simultaneously.
Results
Magnitude and phase images reconstructed with kt-SENSE and HD-PROST, for an undersampling factor of 4 and 8, are shown in Fig. 3 and 4 for a representative subject. HD-PROST reduced noise amplification and noise-like artefacts when compared with kt-SENSE. The phase images were combined to calculate velocities in x, y and z directions, which are presented in Fig. 5 for both kt-SENSE and HD-PROST. The velocities over the cardiac cycle are also included in Fig. 5. HD-PROST achieves lower standard deviation compared to kt-SENSE, indicating a higher precision of velocity measurements with HD-PROST.Conclusion
The present work shows the feasibility of HD-PROST to accelerate carotid 4D flow imaging. This method takes advantage of the redundancy present between velocity encoding contrasts and the different cardiac phases to improve the reconstruction and reduce noise-like artefacts. The results show similar velocities for both kt-SENSE and the proposed HD-PROST, however higher precision was obtained with HD-PROST. Further evaluations with higher undersampling factors will be performed in order to evaluate the robustness of the proposed method.Acknowledgements
This work was supported by EPSRC (EP/L015226/1, EP/P001009, EP/P007619, EP/P032311/1), Wellcome EPSRC Centre for Medical Engineering (NS/A000049/1), and Portuguese Foundation for Science and Technology (FCT - IF/00364/2013, UID/EEA/50009/2013, SFRH/BD/120006/2016).References
[1] Makris GC, Teng Z, Patterson AJ, et al. Advances in MRI for the evaluation of carotid atherosclerosis. Br J Radiol. 2015;88:20140282.
[2] Peper E, Gottwald L, Zhang Q, Coolen B, van Ooij P, Strijkers G, Nederveen A. 30 times accelerated 4D flow MRI in the carotids using a Pseudo Spiral Cartesian acquisition and a Total Variation constrained Compressed Sensing reconstruction. Proc Int Soc Magn Reson Med. 2017.
[3] Bustin A, Ginami G, Cruz G, et al. Five-minute whole-heart coronary MRA with sub-millimeter isotropic resolution, 100% respiratory scan efficiency, and 3D-PROST reconstruction. Magn Reson Med. 2018;00:1–14.
[4] Bernstein MA., Shimakawa A, Pelc NJ. Minimizing TE in moment-nulled or flow-encoded two-and three-dimensional gradient-echo imaging. Magn Reson Med, 1992; 2: 583–588.
[5] Prieto C, Doneva M, Usman M, et al. Highly efficient respiratory motion compensated free-breathing coronary mra using golden-step Cartesian acquisition. J Magn Reson Imaging. 2015;41:738-746.
[6] Tsao J, Boesiger P, and Pruessmann KP. k‐t BLAST and k‐t SENSE: Dynamic MRI with high frame rate exploiting spatiotemporal correlations. Magn Reson. Med 2013; 50: 1031-1042.
Figures
