- Poster presentation
- Open Access
2D bSSFP real-time cardiac CINE-MRI: compressed sensing featuring weighted redundant Haar Wavelet regularization in space and time
© Liu et al; licensee BioMed Central Ltd. 2013
- Published: 30 January 2013
- Proximal Operator
- Residual Artifact
- Comprehensive Clinical Evaluation
- Sparse Sense
- Wavelet Regularization
Compressed sensing promises to achieve a significant acceleration of the data acquisition in the case of real-time CINE-MRI. The major challenge is to achieve both high spatial and temporal resolution. In the current work, an alternative solution to k-t SPARSE SENSE is proposed which exploits spatio-temporal correlations beneficially.
Sparse, incoherent sampling was implemented in a conventionally triggered 2D bSSFP real-time cardiac CINE-MRI sequence. The sampling pattern features an increasing acceleration factor towards the k-space periphery while all phase-encoding steps outside a central region were skipped on one side. Pairing of successive readouts was used to avoid eddy currents. In addition, a pseudo-random offset was applied for each time frame.
Image reconstruction was performed representing the time series of images as a 3-dimensional tensor x=[x1, x2,…, xt] and minimizing ∑_i ∑_j 1/2 \|\|D_i F(s^j⊙x_i)-y_i^j \|\|_2^2 +λ\|\|d⊙(Wx)\|\|_1.
Here, Di is the undersampling operator, F is the 2-dimensional Fourier transformation, yij are the k-space data acquired by the j-th coil element. ⊙ is the component-wise multiplication operator. Wx represent the coefficients after redundant Haar Wavelet transformation. These are compressible due to spatial and temporal correlations. d is a weighting operator that imposes strong weights to the temporal correlations. sj is the coil profile of the j-th coil which was estimated following the Eigen-vector approach applied to the temporal average of all input data. The optimization problem was solved using the Nesterov's approach together with an extension of the Dykstra's algorithm for solving the associated proximal operator.
Data acquisition was performed on a 3.0T MR scanner (MAGNETOM Skyra, Siemens AG, Erlangen, Germany) in 5 volunteers: FOV 380×255 mm2, matrix 224×168, TR/TE 2.9/1.26 ms, bandwidth 1395 Hz/px, flip angle 68, temporal resolution 33 ms, 38 temporal phases, 30 coils, and 10 frequency encodings were acquired for each temporal phase.
The current work demonstrates a significant improvement in image quality if spatio-temporal correlations are taken into account in case of real-time cardiac CINE-MRI. Since the initial experience in volunteers is promising, a comprehensive clinical evaluation is pending.
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.