260 research outputs found
4D Combined Angiography and Perfusion using Radial Imaging and Arterial Spin Labeling (CAPRIA): Data and Code to Reproduce Statistics and Figures
No full textMatlab code and data to reproduce statistical analysis and figures from the 4D CAPRIA paper pre-print (which can be found here), by Thomas W Okell and Mark Chiew, July 2022. Core Matlab code for CAPRIA signal simulations and image reconstruction can be found on GitHub and Zenodo.Thanks for additional funding support from the Royal Academy of Engineering (RF/132, RF/201617/16/23)
Exploring subspace-constrained approaches to low-rank fMRI acceleration
No full textFunctional magnetic resonance imaging (fMRI) is a medical imaging technique that measures brain activity non-invasively. One of the fundamental quandaries in fMRI is the balance that must be struck between spatial fidelity and temporal resolution. An increase in sampling efficiency could improve either or both of these metrics, allowing images to be created from fewer data points than would otherwise be required. This process is referred to as acceleration.
Some degree of acceleration is already standard in fMRI scans. The most common acceleration methods are parallel imaging methods, which utilise the spatial sensitivity profiles of the receiver coils in order to separate the aliased artefacts that result from using fewer data to create each image. However, there are other mathematical properties which can also be incorporated into the reconstruction process in order to allow a higher degree of acceleration.
One such property is the inherently low-rank nature of fMRI data, which was introduced by Chiew et al. in 2015 as the k-t FASTER method (fMRI Accelerated in Space-Time via Truncation of Effective Rank). The authors also demonstrated in 2016 that the low-rank information could be combined with the coil sensitivity profiles to achieve a higher acceleration factor than either low-rank information or coil profiles could achieve alone.
In this thesis, the k-t FASTER approach is expanded upon by incorporating additional, subspace- specific constraints into the reconstruction process. First, k-t FASTER will be reformulated as an alternating minimisation problem in order to more easily allow subspace-specific regularisation terms. Then, a variety of constraints will be explored in an artificial framework. The constraints being tested are: Tikhonov constraints (which encourage the subspaces to take more minimal energy forms), low-resolution priors (which more greatly weight the oversampled central k-space in radial sampling), and a temporal subspace smoothing constraint (which minimises the variation between frames). These constraints will be applied to real fMRI data acquired with a TURBINE trajectory (Trajectory Using Radially Batched Internal Navigator Echoes), a hybrid radial-Cartesian 3D trajectory with Golden Ratio angle increments in the radial orientation.
The aforementioned subspace-constrained approaches could be seen to achieve better classification of the underlying functional activation over a k-t FASTER reconstruction in real data at R=16, or TRvol=0.5s. Ultimately, Tikhonov constraints are found to provide consistently high-quality reconstructions at a range of acceleration factors and SNR values, but in real data with a slow-paradigm task fMRI experiment at high acceleration (R=26, TRvol=0.3s), the temporal subspace smoothing constraints can outperform Tikhonov constraints
Post-mortem QSM and R2* maps
No full textThis repository contains data associated with the following publication:
Methods for quantitative susceptibility and R2* mapping in whole post-mortem brains at 7T applied to amyotrophic lateral sclerosis
Authors: Chaoyue Wang, Sean Foxley, Olaf Ansorge, Sarah Bangerter-Christensen, Mark Chiew, Anna Leonte, Ricarda A.L. Menke, Jeroen Mollink, Menuka Pallebage-Gamarallage, Martin R. Turner, Karla L. Miller*, Benjamin C. Tendler* (* indicates equal contribution)
The text file dataset_loc_QSM_R2s.txt contains a link to the dataset. Further information about the dataset can be found in dataset_info.txt and the publication
Low Rank Denoising Tools for MRSI
No full textTools for low rank denoising of MRSI
Source code for various low-rank denoising approaches for MRSI. This package contains functions to carry out:
Global and local spatio-temporal low-rank denoising 4
Global and local LORA 4
Linear-predictability denoising 1, 4
SURE optimised local soft thresholding (SURE-SVT) 2
SURE optimised local hard thresholding (SURE-SVHT) 6
Development hosted on the Wellcome Centre for Integrative Neuroimaging GitLab. Python package available from Conda Forge and PyPi.
These tools were developed to accompany the publication Clarke WT, Chiew M. Uncertainty in denoising of MRSI using low-rank methods. Magnetic Resonance in Medicine 2022;87:574–588 doi: 10.1002/mrm.29018. Please cite this work if you use the tools.
References
1: Cadzow JA. Signal enhancement-a composite property mapping algorithm. IEEE Transactions on Acoustics, Speech, and Signal Processing 1988;36:49–62 doi: 10.1109/29.1488.
2: Candès EJ, Sing-Long CA, Trzasko JD. Unbiased Risk Estimates for Singular Value Thresholding and Spectral Estimators. IEEE Transactions on Signal Processing 2013;61:4643–4657 doi: 10.1109/TSP.2013.2270464.
3: Chen Y, Fan J, Ma C, Yan Y. Inference and uncertainty quantification for noisy matrix completion. PNAS 2019;116:22931–22937 doi: 10.1073/pnas.1910053116.
4: Nguyen HM, Peng X, Do MN, Liang Z. Denoising MR Spectroscopic Imaging Data With Low-Rank Approximations. IEEE Transactions on Biomedical Engineering 2013;60:78–89 doi: 10.1109/TBME.2012.2223466.
5: Song J, Xia S, Wang J, Patel M, Chen D. Uncertainty Quantification for Hyperspectral Image Denoising Frameworks based on Low-rank Matrix Approximation. arXiv:2004.10959 [cs, eess] 2021.
6: Ulfarsson MO, Solo V. Selecting the Number of Principal Components with SURE. IEEE Signal Processing Letters 2015;22:239–243 doi: 10.1109/LSP.2014.2337276
k-t FASTER: a novel method for accelerating FMRI data acquisition using low rank constraints
No full textPurposeIn functional MRI (fMRI), faster sampling of data can provide richer temporal information and increase temporal degrees of freedom. However, acceleration is generally performed on a volume‐by‐volume basis, without consideration of the intrinsic spatio‐temporal data structure. We present a novel method for accelerating fMRI data acquisition, k‐t FASTER (FMRI Accelerated in Space‐time via Truncation of Effective Rank), which exploits the low‐rank structure of fMRI data.Theory and MethodsUsing matrix completion, 4.27× retrospectively and prospectively under‐sampled data were reconstructed (coil‐independently) using an iterative nonlinear algorithm, and compared with several different reconstruction strategies. Matrix reconstruction error was evaluated; a dual regression analysis was performed to determine fidelity of recovered fMRI resting state networks (RSNs).ResultsThe retrospective sampling data showed that k‐t FASTER produced the lowest error, approximately 3–4%, and the highest quality RSNs. These results were validated in prospectively under‐sampled experiments, with k‐t FASTER producing better identification of RSNs than fully sampled acquisitions of the same duration.ConclusionWith k‐t FASTER, incoherently under‐sampled fMRI data can be robustly recovered using only rank constraints. This technique can be used to improve the speed of fMRI sampling, particularly for multivariate analyses such as temporal independent component analysis. Magn Reson Med 74:353–364, 2015. © 2014 The Authors Magnetic Resonance in Medicine published by Wiley Periodicals, Inc. on behalf of International Society for Magnetic Resonance in Medicine. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cite
Provenance-based trust for grid computing: Position Paper
No full textCurrent evolutions of Internet technology such as Web Services, ebXML, peer-to-peer and Grid computing all point to the development of large-scale open networks of diverse computing systems interacting with one another to perform tasks. Grid systems (and Web Services) are exemplary in this respect and are perhaps some of the first large-scale open computing systems to see widespread use - making them an important testing ground for problems in trust management which are likely to arise. From this perspective, today's grid architectures suffer from limitations, such as lack of a mechanism to trace results and lack of infrastructure to build up trust networks. These are important concerns in open grids, in which "community resources" are owned and managed by multiple stakeholders, and are dynamically organised in virtual organisations. Provenance enables users to trace how a particular result has been arrived at by identifying the individual services and the aggregation of services that produced such a particular output. Against this background, we present a research agenda to design, conceive and implement an industrial-strength open provenance architecture for grid systems. We motivate its use with three complex grid applications, namely aerospace engineering, organ transplant management and bioinformatics. Industrial-strength provenance support includes a scalable and secure architecture, an open proposal for standardising the protocols and data structures, a set of tools for configuring and using the provenance architecture, an open source reference implementation, and a deployment and validation in industrial context. The provision of such facilities will enrich grid capabilities by including new functionalities required for solving complex problems such as provenance data to provide complete audit trails of process execution and third-party analysis and auditing. As a result, we anticipate that a larger uptake of grid technology is likely to occur, since unprecedented possibilities will be offered to users and will give them a competitive edge
Structured low-rank methods for robust 3D multi-shot EPI
No full textMagnetic resonance imaging (MRI) has inherently slow acquisition speed, and Echo-Planar Imaging (EPI), as an efficient acquisition scheme, has been widely used in functional magnetic resonance imaging (fMRI) where an image series with high temporal resolution is needed to measure neuronal activity. Recently, 3D multi-shot EPI which samples data from an entire 3D volume with repeated shots has been drawing growing interest for fMRI with its high isotropic spatial resolution, particularly at ultra-high fields. However, compared to single-shot EPI, multi-shot EPI is sensitive to any inter-shot instabilities, e.g., subject movement and even physiologically induced field fluctuations. These inter-shot inconsistencies can greatly negate the theoretical benefits of 3D multi-shot EPI over conventional 2D multi-slice acquisitions.
Structured low-rank image reconstruction which regularises under-sampled image reconstruction by exploiting the linear dependencies in MRI data has been successfully demonstrated in a variety of applications. In this thesis, a structured low-rank reconstruction method is optimised for 3D multi-shot EPI imaging together with a dedicated sampling pattern termed seg-CAIPI, in order to enhance the robustness to physiological fluctuations and improve the temporal stability of 3D multi-shot EPI for fMRI at 7T. Moreover, a motion compensated structured low-rank reconstruction framework is also presented for robust 3D multi-shot EPI which further takes into account inter-shot instabilities due to bulk motion. Lastly, this thesis also investigates into the improvement of structured low-rank reconstruction from an algorithmic perspective and presents the locally structured low-rank reconstruction scheme
Optimized Self-Supervised Learning for MRI Reconstruction
No full textSupervised deep learning MRI reconstruction recovers images from accelerated MRI scans, but requires fully-sampled k-space for training. Self-supervised learning for MRI reconstruction bypasses the need for fully sampled k-space by partitioning under-sampled k-space into two disjoint sets, and training a neural network to predict one set from the other. Remarkably, self-supervised learning achieves performance similar to supervised learning without requiring fully sampled data. In this thesis, I enhance self-supervised learning methods through two key approaches: (1) learning an optimal k-space partitioning strategy, and (2) incorporating multi-contrast information into the reconstruction process. In the first approach, I trained a network to learn an optimal k-space partitioning probability distribution for self-supervised learning, outperforming previous heuristic-based methods. In the second approach, I demonstrate that integrating multiple MRI contrasts improves self-supervised reconstruction performance by leveraging correlated information across contrasts. I further improve the second approach by extending multi-contrast self-supervised learning to jointly learn an optimal k-space partitioning for each contrast. These proposed enhancements improve self-supervised reconstruction fidelity compared to previous single-contrast self-supervised learning methods.M.Sc
Three-dimensional hybrid radial Cartesian echo planar imaging for functional MRI
No full textFunctional magnetic resonance imaging (fMRI) has provided neuroscientists with a powerful tool to non-invasively study brain function. Typically, fMRI data acquisition is performed using the well-established multi-slice two-dimensional echo planar imaging (2D EPI) technique. While 2D EPI has the considerable advantage of robustness, it is relatively SNR inefficient, particularly at high spatial resolution. Three dimensional (3D) sampling approaches, such as multi-shot 3D EPI provide a theoretical SNR gain compared to 2D EPI and can utilize parallel imaging acceleration along multiple dimensions, leading to the potential for higher spatial and temporal resolution. However, these multi-shot acquisitions span several seconds, making them susceptible to physiological fluctuations. In particular, subject motion is a major source of image degradation.
This thesis aims to characterise and improve fMRI acquisition techniques based on 3D EPI approaches. We explored the temporal SNR characteristics of standard segmented 3D EPI for different spatial resolutions and acceleration factors. Specifically, we studied how physiological noise affects the optimal choice of imaging parameters, such as the amount of acceleration.
To address some of the shortcomings of conventional 3D EPI, we implemented a hybrid radial-Cartesian 3D EPI trajectory, called TURBINE. This scheme collects EPI "blades" which are rotated about the phase-encoding axis using a golden angle rotation increment, allowing reconstruction at flexible temporal resolution. The self-navigating properties of the sequence are used to determine motion estimates from high temporal resolution navigator images and correct for subject motion as part of the image reconstruction process. We demonstrated that this scheme reduces the impact of motion on fMRI data in the presence of subtle and large subject motions. The techniques developed in this thesis aim to increase the flexibility and robustness of fMRI acquisitions. Ultimately, this research may help increase the utility of fMRI in difficult subjects or patient populations.</p
Development of a strategy to improve the accuracy and efficiency of computerized E-codes classification: Narrative coding assignments between human and computer
No full textExternal cause of injury codes (E-codes) and the Occupational Injury and Illness Classification system codes (OIICS) are useful for the purpose of accident prevention analysis. However, the coding task has become burdensome as the trained coders need to code a huge amount of text narratives. This study presents the use of Naïve Bayes machine learning tool to classify large amounts of narrative texts, including the strategic assignment of the tasks between human and computer in order to reduce erroneous decisions. Receiver Operating Characteristic (ROC) curves were used to identify the optimal region for different categories to achieve optimal results by effectively minimizing resources necessary for manual coding. The results showed that by utilizing the ROC based reject rule to assign difficult tasks for manual coding, it was possible to obtain an final accuracy of the classification of 82 percent, 89 percent and 94 percent respectively for filtering strategies in which 20 percent, 35 percent and 49 percent of the narratives were manually coded
- …
