CS-MRI reconstruction based on analysis dictionary learning and manifold structure regularization

Compressed sensing (CS) significantly accelerates magnetic resonance imaging (MRI) by allowing the exact reconstruction of image from highly undersampling k-space data. In this process, the high sparsity obtained by the learned dictionary and exploitation of correlation among patches are essential to the reconstructed image quality. In this paper, by a use of these two aspects, we propose a novel CS-MRI model based on analysis dictionary learning and manifold structure regularization (ADMS). Furthermore, a proper tight frame constraint is used to obtain an effective overcomplete analysis dictionary with a high sparsifying capacity. The constructed manifold structure regularization nonuniformly enforces the correlation of each group formed by similar patches, which is more consistent with the diverse nonlocal similarity in realistic images. The proposed model is efficiently solved by the alternating direction method of multipliers (ADMM), in which the fast algorithm for each sub-problem is separately developed. The experimental results demonstrate that main components in the proposed method contribute to the final reconstruction performance and the effectiveness of the proposed model.

A novel temporal filtering strategy for functional MRI using UNFOLD

A major challenge for fMRI at high spatial resolution is the limited temporal resolution. The UNFOLD method increases image acquisition speed and potentially enables high acceleration factors in fMRI. Spatial aliasing artifacts due to interleaved k-space sampling are to be removed from the image time series by temporal filtering before statistical mapping in the time domain can be carried out. So far, low-pass filtering and multi-band filtering have been proposed. Particularly at high UNFOLD factors both methods are non-optimal. Low-pass filtering severely degrades temporal resolution and multi-band filtering leads to temporal autocorrelations affecting statistical modelling of activation. In this work, we present a novel temporal filtering strategy that significantly reduces temporal autocorrelations compared to multi-band filtering. Two datasets (finger-tapping and resting state) were post-processed using the proposed and the multi-band filter with varying set-ups (i.e. transition bands). When the proposed filtering strategy was used, a linear regression analysis revealed that the number of false positives was significantly decreased up to 34% whereas the number of activated voxels was not significantly affected for most filter parameters. In total, this led to an effective increase in the number of activated voxels per false positive for each filter set-up. At a significance level of 5%, the number of activated voxels was increased up to 41% by using the proposed filtering strategy.

Spiral imaging in fMRI

T2*-weighted Blood Oxygen Level Dependent (BOLD) functional magnetic resonance imaging (fMRI) requires efficient acquisition methods in order to fully sample the brain in a several second time period. The most widely used approach is Echo Planar Imaging (EPI), which utilizes a Cartesian trajectory to cover k-space. This trajectory is subject to ghosts from off-resonance and gradient imperfections and is intrinsically sensitive to cardiac-induced pulsatile motion from substantial first- and higher order moments of the gradient waveform near the k-space origin. In addition, only the readout direction gradient contributes significant energy to the trajectory. By contrast, the spiral method samples k-space with an Archimedean or similar trajectory that begins at the k-space center and spirals to the edge (spiral-out), or its reverse, ending at the origin (spiral-in). Spiral methods have reduced sensitivity to motion, shorter readout times, improved signal recovery in most frontal and parietal brain regions, and exhibit blurring artifacts instead of ghosts or geometric distortion. Methods combining spiral-in and spiral-out trajectories have further advantages in terms of diminished susceptibility-induced signal dropout and increased BOLD signal. In measurements of temporal signal to noise ratio measured in 8 subjects, spiral-in/out exhibited significant increases over EPI in voxel volumes recovered in frontal and whole brain regions (18% and 10%, respectively).

Rapid 3D radial multi-echo functional magnetic resonance imaging

Functional magnetic resonance imaging with readouts at multiple echo times is useful for optimizing sensitivity across a range of tissue T2* values as well as for quantifying T2*. With single-shot acquisitions, both the minimum TE value and the number of TEs which it is possible to collect within a single TR are limited by the long echo-planar imaging readout duration (20-40 ms). In the present work, a multi-shot 3D radial acquisition which allows rapid whole-brain imaging at a range of echo times is proposed. The proposed 3D k-space coverage is implemented via a series of rotations of a single 2D interleaf. Data can be reconstructed at a variety of temporal resolutions from a single dataset, allowing for a flexible tradeoff between temporal resolution and BOLD contrast to noise ratio. It is demonstrated that whole-brain images at 5 echo times (TEs from 10 to 46 ms) can be acquired at a temporal rate as rapid as 400 ms/volume (3.75 mm isotropic resolution). Activation maps for a simultaneous motor/visual task consistent across multiple acceleration factors are obtained. Weighted combination of the echoes results in Z-scores that are significantly (p=0.016) higher than those resulting from any of the individual echo time images.