Research - Code and Resources

Model-based dynamic off-resonance correction for improved accelerated fMRI in awake behaving non-human primates

Shahdloo M, Schüffelgen U, Papp D, Miller KL, Chiew M. bioRxiv 2021; :2021.09.23.461491 | doi:10.1101/2021.09.23.461491

Here a model-based approach to estimating and correcting for strong, dynamic off-resonance effects in awake behaving non-human primates is presented. Dynamic field fluctuations are modelled as first-order spatial pertubations, and a GRAPPA-operator formalism is used to solve the dynamic estimation problem as a linear encoding shift, based on EPI three-line navigators. This allows for correction of acquired k-space data relative to the reference calibration data used to estimate simultaneous multi-slice reconstruction kernels, reducing residual ghosting/aliasing artefacts and producing more temporally stable time-series data.


Improving robustness of 3D multi-shot EPI by structured low-rank reconstruction of segmented CAIPI sampling for fMRI at 7T

Chen X, Wu W, Chiew M. bioRxiv 2021; :2021.08.19.457024 | doi:10.1101/2021.08.19.457024

This work employs Hankel-structured low-rank matrix recovery for multi-shot 3D EPI at 7T, to improve robustness to physiologically-induced inter-shot phase fluctuations. The low-rank constraint is enforced using a convex nuclear norm penalty, and optimisation is performed with the alternating direction method of multipliers (ADMM).


Uncertainty in denoising of MRSI using low-rank methods

Clarke WT, Chiew M. Magnetic Resonance in Medicine 2022; 87(2):574–588 | doi:10.1002/mrm.29018

Low-rank denoising methods have become increasingly popular for MRSI data, but in this work we demonstrate that uncertainty in metabolite concentrations is often under-estimated after non-linear denoising operations. We propose a more accurate uncertainty estimate based on a boot-strap estimation procedure using an analytic formulation for the covariance of the denoised data, and also explore the use of Stein’s unbiased risk estimate for automatic parameter selection using soft and hard-thresholding spectral non-linearities.


Subspace-constrained approaches to low-rank fMRI acceleration

Mason HT, Graedel NN, Miller KL, Chiew M. NeuroImage 2021; 238:118235 | doi:10.1016/j.neuroimage.2021.118235

This work explores different approaches for subspace constrained, low-rank fMRI reconstruction of accelerated (under-sampled) data. Various L2-constraints are investigated, and comparisons to the original (rank-constrained) k-t FASTER method are performed.


The Set Increment with Limited Views Encoding Ratio (SILVER) Method for Optimizing Radial Sampling of Dynamic MRI

Schauman SS, Okell TW, Chiew M. bioRxiv 2021; :2020.06.25.171017 | doi:10.1101/2020.06.25.171017

In this work, the radial spoke angular increment for dynamic radial MRI is min-max optimised for multiple reconstruction windows. Optimising for a small set of desired windows (temporal resolutions) retains the flexibility of golden-ratio based increments, while achieving more uniform k-space coverage and higher SNR image reconstruction.


Accelerated calibrationless parallel transmit mapping using joint transmit and receive low-rank tensor completion

Hess AT, Dragonu I, Chiew M. Magnetic Resonance in Medicine 2021; 86(5):2454–2467 | doi:10.1002/mrm.28880

This work introduces a method for highly accelerated relative parallel transmit mapping, by leveraging information across receive channels and transmit channels simultaneously through the use of a low-rank tensor representation. This approach enables a considerable speedup of relative transmit mapping acquisitions, by reconstructing with high fidelity under-sampling factors up to 8x.


Highly accelerated vessel-selective arterial spin labeling angiography using sparsity and smoothness constraints

Schauman SS, Chiew M, Okell TW. Magnetic Resonance in Medicine 2020; 83(3):892–905 | doi:10.1002/mrm.27979

In this paper, we developed a reconstruction of highly under-sampled vessel-encoded ASL angiography, leveraging the enhanced spatial sparsity of individual vessel images, and smoothness of the angiographic temporal dynamics. By leveraging the improved spatial sparsity of the vessel-decoded images, the proposed reconstruction was able to recover vessel-selective information for free, with no acquisition time penalty or loss of fidelty compared to non-vessel encoded ASL angiography.


Improved statistical efficiency of simultaneous multi-slice fMRI by reconstruction with spatially adaptive temporal smoothing

Chiew M, Miller KL. NeuroImage 2019; 203:116165 | doi:10.1016/j.neuroimage.2019.116165

Reconstruction of simultaneous multi-slice EPI data using temporal smoothing regularisation to reduce noise amplification (g-factors) at the cost of reduced temporal degrees of freedom, for a net improvement in statistical and tSNR efficiency. The code provided handles arbitrary slice and in-plane acceleration factors, and arbitrary, time-dependent sampling patterns (CAIPI shifts). Calculation of g-factors, effective degrees of freedom, and reconstruction efficiency are also provided.


Recovering task fMRI signals from highly under-sampled data with low-rank and temporal subspace constraints.

Chiew M, Graedel NN, Miller KL. NeuroImage 2018; 174:97–110 | doi:10.1016/j.neuroimage.2018.02.062

This is a method for including task-fMRI design information in the image reconstruction process. It combines low-rank and temporal subspace constraints, or in another sense, performs a PCA/GLM on the under-sampled data.

NB: Recon code depends on J. Fessler’s NUFFT toolbox


Accelerating functional MRI using fixed-rank approximations and radial-cartesian sampling.

Chiew M, Graedel NN, Mcnab JA, Smith SM, Miller KL. Magnetic resonance in medicine 2016; 76(6):1825–1836 | doi:10.1002/mrm.26079

This work generalises the k-t FASTER approach for rank-constrained FMRI data reconstruction by permitting more general linear encoding of the imaging data, including non-Cartesian (e.g. radial) sampling trajectories and coil sensitivity encoding.

NB: Recon code depends on J. Fessler’s NUFFT toolbox


K-t FASTER: Acceleration of functional MRI data acquisition using low rank constraints.

Chiew M, Smith SM, Koopmans PJ, Graedel NN, Blumensath T, Miller KL. Magnetic resonance in medicine 2015; 74(2):353–364 | doi:10.1002/mrm.25395

This project used low-rank matrix completion or matrix recovery techniques to reconstruct under-sampled FMRI data. This was motivated by the low-dimensional structure often enforced by FMRI analysis methods, like in PCA dimensionality reductions or ICA decompositions, but leveraged here to constrain image reconstruction. Here, we use an iterative hard thresholding and matrix shrinkage algorithm to recover Cartesian sampled FMRI data with 4-fold under-sampling.

NB: This work is largely superceded by the above, which generalises this approach to non-Cartesian sampling trajectories with coil sensitivity encoding