ESMRMB Congress (2016) 29 (Suppl 1):S247–S400

S317 [4]

Marques JP, Bowtell R. Evaluation of a new method to correct the effects of motion-induced B0-field variation during fMRI. Proc. 13th Annu. Meet. ISMRM 2005:#510. [5] Lamberton F, Delcroix N, Grenier D, Mazoyer B, Joliot M. A new EPI-based dynamic field mapping method: Application to retrospective geometrical distortion corrections. J. Magn. Reson. Imaging 2007;26:747–755. [6] Dymerska B, Poser BA, Bogner W, Visser E, Eckstein K, Cardoso P, Barth M, Trattnig S, Robinson SD. Correcting dynamic distortions in 7T echo planar imaging using a jittered echo time sequence. Magn. Reson. Med. 2015.

336 Acceleration of functional MRI data acquisition by separation of background and dynamic components L. Weizman1, K. Miller1, Y. Eldar2, M. Chiew1 1 Oxford Centre for Functional MRI of the Brain, Oxford University, Oxford/UNITED KINGDOM, 2Electrical Engineering, Technion, Israel Institute of Technology, Haifa/ISRAEL Purpose/Introduction: In functional MRI (fMRI), faster sampling of data can increase frame rate for improved analysis, and may lead to higher spatial resolution without sacrificing temporal resolution. High spatial and temporal correlations in fMRI data sets enabled the application of compressed sensing (CS) [1] or low-rank matrix completion [2, 3] for fMRI. It is well known that pure low-rank based reconstruction [2] is sensitive to outliers, which may exist due to various noise sources. Therefore, the combination of CS and low-rank [4, 5] is very attractive to increase the acceleration rate. We apply both techniques for fMRI by introducing SELFIE—a method for SparsE and Low-rank based Functional Imaging, using Extended weighting. Subjects and Methods: fMRI data is represented as an NxT matrix, X, where T is the number of time-points and N is a spatial index. We assume that fMRI data can be represented as a combination of low-rank data and sparse temporal component, and aim at finding the low-rank (L) and sparse (S) components. A solution is obtained via reweighted-l1 minimization [6]. To this end, we solve the following minimization problem: min||L||* + k||WS||1 s.t. E(L + S) = D, where E is the acquisition matrix, D is the undersampled k-t data and W is the weighting matrix whose elements are determined iteratively, in order to correctly capture the support of S. The reconstruction algorithm of SELFIE is given in Fig. 1, and the reconstruction result is given by X = L+S. Retrospectively undersampled resting state fMRI data (undersampling factor of 4) simulations were performed using realistic data (same as used [2]) to evaluate the benefits of our approach with knowledge of the ground truth.

References: [1] Bernstein MA, Grgic M, Brosnan TJ, Pelc NJ. Reconstructions of phase contrast, phased array multicoil data. Magn. Reson. Med. 1994;32:330–334. [2] Robinson S, Dymerska B, Trattnig S. Improving the accuracy of 2D phase unwrapping using a triplanar approach. Proc. 22nd Annu. Meet. ISMRM Milan 2014:#3262. [3] Robinson S, Jovicich J. B0 mapping with multi-channel RF coils at high field. Magn. Reson. Med. 2011;66:976–988.

123

S318 SELFIE iterative algorithm for reconstruction of X from undersampled k-space data, D. In each k-th iteration, low-rank (Lk) and sparse (Sk) components are computed, as well as the weighting matrix, Wk, where the reconstruction result is given as Xk. Those are fed in the next iteration and the process continues until convergence. Results: The fidelity of SELFIE was assessed by examining principle angles [7] and resting state networks, obtained with MELODIC [8], vs. ground truth. Figure 2 shows the principle angles of k-t FASTER [2] and SELFIE vs. ground truth. Figure 3 shows a visual and default mode network (DMN) maps of k-t FASTER and SELFIE vs. ground truth. It can be seen that SELFIE provides subspaces with smaller angles with the ground truth, leading to better accuracy of the spatial subset, as well as better estimation of the resting state maps.

ESMRMB Congress (2016) 29 (Suppl 1):S247–S400 References: [1] Holland, D. J., et al. ‘‘Compressed sensing reconstruction improves sensitivity of variable density spiral fMRI.’’ Magnetic Resonance in Medicine 70.6 (2013): 1634–1643. [2] Chiew, Mark, et al. ‘‘k-t FASTER: Acceleration of functional MRI data acquisition using low rank constraints.’’ Magnetic resonance in medicine 74.2 (2015): 353–364. [3] Chiew, Mark, et al. ‘‘Accelerating functional MRI using fixedrank approximations and radial-cartesian sampling.’’ Magnetic resonance in medicine (2016). [4] Cande`s, Emmanuel J., et al. ‘‘Robust principal component analysis?.’’Journal of the ACM (JACM) 58.3 (2011): 11. [5] Otazo, Ricardo, et al. ‘‘Low-rank plus sparse matrix decomposition for accelerated dynamic MRI with separation of background and dynamic components.’’ Magnetic Resonance in Medicine 73.3 (2015): 1125–1136. [6] Weizman, Lior, et al. ‘‘Compressed sensing for longitudinal MRI: An adaptive-weighted approach.’’ Medical physics 42.9 (2015): 5195–5208. [7] Knyazev, Andrew V., et al. ‘‘Principal angles between subspaces in an A-based scalar product: algorithms and perturbation estimates.’’ SIAM Journal on Scientific Computing 23.6 (2002): 2008–2040. [8] Beckmann, Christian F., and Stephen M. Smith. ‘‘Probabilistic independent component analysis for functional magnetic resonance imaging.’’ Medical Imaging, IEEE Transactions on 23.2 (2004): 137–152. [9] Smith, Stephen M., et al. ‘‘Temporally-independent functional modes of spontaneous brain activity.’’ Proceedings of the National Academy of Sciences 109.8 (2012): 3131–3136.

337 BOLD fMRI in the basal ganglia at 7T using simultaneous multislice (SMS) multi-echo EPI S. Bollmann1, A. Pucket2, B. A. Poser3, R. Cunnington2, M. Barth1 1 The University of Queensland, Centre for Advanced Imaging, Brisbane/QLD/AUSTRALIA, 2The University of Queensland, Queensland Brain Institute, Brisbane/QLD/AUSTRALIA, 3, Maastricht University, Department of Cognitive Neuroscience, MR Physics, Maastricht/NETHERLANDS

Discussion/Conclusion: The application of SELFIE for fMRI allows better modeling of the data and thus improves reconstruction performance and statistical maps estimation.Our results strongly suggest that SELFIE can be used to improve the speed of fMRI sampling, particularly for multivariate analyses such as temporal independent component analysis [9].

123

Purpose/Introduction: Basal ganglia nuclei are involved in a wide range of functions, however their depiction in functional BOLD MRI has been notoriously difficult due to their small size and high iron content, and hence, shortened T2*. To avert this signal loss but pertain high BOLD sensitivity, multi-echo fMRI has been developed acquiring multiple echoes at different echo times [1, 2]. Here, we compare temporal SNR and t-values of multi-echo and single-echo sequences with simultaneous multi-slice (SMS) acquisition [3, 4], with emphasis on the function of the basal ganglia at 7T. ME-SMSEPI has previously been shown at 3T [5] and 7T [6] to enhance sensitivity compared to single-band EPI sequences. Subjects and Methods: We acquired MRI data on a MAGNETOM 7T whole-body scanner (Siemens Healthcare, Erlangen, Germany) with a 32-channel head coil (Nova Medical,Wilmington,US). We compared SMS implementations of both multi-echo and single-echo sequences with a 2.5 mm isotropic voxel size and a TR of 830 and 589 ms, respectively (Fig. 1). 6 subjects performed two runs for each sequence of a 6-min rhythmic finger-tapping task (block length: 18 s) known to elicit strong basal ganglia activation [7]. Preprocessing and statistical analysis were performed in Matlab [8] using SPM12 [9], including realignment,