Simulating how tissue microstructure affects MRI
Poster No:
2414
Submission Type:
Abstract Submission
Authors:
Michiel Cottaar1, Zhiyu Zheng1, Benjamin Tendler1, Karla Miller1, Saad Jbabdi1
Institutions:
1Oxford University, Oxford, United Kingdom
First Author:
Co-Author(s):
Introduction:
MRI has many different modalities that are sensitive to tissue microstructure, including diffusion-weighted MRI, susceptibility-weighted MRI, magnetisation transfer, and quantitative relaxometry. Because many of these modalities are sensitive to different aspects of the same microstructural components (e.g., myelin), combining these modalities may provide a more comprehensive view of those microstructural components. However, these different modalities are usually analysed in isolation with each modality coming with its own set of models and assumptions. Here we present a new Monte Carlo MR (MCMR) simulator1 that aims to capture different ways that tissue microstructure affects the MRI signal evolution across a wide range of MR modalities, which are presented in Figure 1.
·Figure 1
Methods:
MCMR simulator has been implemented in the Julia programming language2. It has both a Julia and command line interface, with comprehensive documentation and tutorials available for both.
An overview of the simulator methodology is shown in Figure 2. Briefly, the user synthesises a tissue geometry (consisting of any combination of infinite walls, infinite cylinders, spheres, and arbitrary meshes), and defines one or more MR sequences for which the MR signal will be computed in parallel. These sequences can include finite or instantaneous radiofrequency (RF) pulses and gradients and can be flexibly defined by the user or directly read from pulseq files3. For each sequence the simulator will predict the MRI signal for a single voxel.
The simulator uses a Monte Carlo approach, which follows the random walk of individual spins, which is hindered by the user-provided tissue geometry. The spin magnetisation is updated using the Bloch equations, which includes the off-resonance field due to the tissue magnetic susceptibility of myelinated, infinite cylinders, or of arbitrary meshes.
At each collision with the tissue, there is also user-defined probability of spins to pass through the surface, which enables the modelling of exchange. Otherwise, the spin will bounce back.
Magnetisation transfer is implemented by allowing the spin magnetisation to transfer to a stationary "bound" pool during a collision with the tissue. The magnetisation transfer is controlled by a user-defined surface density and dwell time.
The permeability, bound pool properties, and the magnetic susceptibility can be set on a global level or set to a different values for different simulated objects (down to individual triangles in any meshes), which would allow one to vary the permeability or myelination between cells or even along individual axons. Similarly, the relaxation rates of the free water are usually set globally, but can be updated within specific compartments.
While the simulator does provide some support for generating geometries consisting of randomly distribution cylinders or spheres, for more complicated geometries we recommend generating a mesh using an existing tool4-7 (e.g., Config5 which was used to generate the mesh in Figure 2).
An overview of the simulator methodology is shown in Figure 2. Briefly, the user synthesises a tissue geometry (consisting of any combination of infinite walls, infinite cylinders, spheres, and arbitrary meshes), and defines one or more MR sequences for which the MR signal will be computed in parallel. These sequences can include finite or instantaneous radiofrequency (RF) pulses and gradients and can be flexibly defined by the user or directly read from pulseq files3. For each sequence the simulator will predict the MRI signal for a single voxel.
The simulator uses a Monte Carlo approach, which follows the random walk of individual spins, which is hindered by the user-provided tissue geometry. The spin magnetisation is updated using the Bloch equations, which includes the off-resonance field due to the tissue magnetic susceptibility of myelinated, infinite cylinders, or of arbitrary meshes.
At each collision with the tissue, there is also user-defined probability of spins to pass through the surface, which enables the modelling of exchange. Otherwise, the spin will bounce back.
Magnetisation transfer is implemented by allowing the spin magnetisation to transfer to a stationary "bound" pool during a collision with the tissue. The magnetisation transfer is controlled by a user-defined surface density and dwell time.
The permeability, bound pool properties, and the magnetic susceptibility can be set on a global level or set to a different values for different simulated objects (down to individual triangles in any meshes), which would allow one to vary the permeability or myelination between cells or even along individual axons. Similarly, the relaxation rates of the free water are usually set globally, but can be updated within specific compartments.
While the simulator does provide some support for generating geometries consisting of randomly distribution cylinders or spheres, for more complicated geometries we recommend generating a mesh using an existing tool4-7 (e.g., Config5 which was used to generate the mesh in Figure 2).
·Figure 2
Results:
By combining the effects of diffusion, permeability (exchange), magnetic susceptibility, and magnetisation transfer in the MCMR simulator, we allow for a more coherent analysis of these different ways microstructure affects the MRI signal. One could change one aspect of the tissue microstructure (e.g., reduce myelin thickness) and investigate its effect on a wide range of different MRI acquisitions. This could be used to help interpret changes seen in multi-modal MRI acquisitions or for the development of new acquisition protocols (or sequences) sensitive to specific aspects of the tissue microstructure.
Conclusions:
Documentation including installation instructions and tutorials for both the Julia and command line interface is available at https://open.win.ox.ac.uk/pages/ndcn0236/mcmrsimulator.jl/stable/. The code is available at https://git.fmrib.ox.ac.uk/ndcn0236/mcmrsimulator.jl.
Modeling and Analysis Methods:
Diffusion MRI Modeling and Analysis
Methods Development 2
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
Cortical Cyto- and Myeloarchitecture
White Matter Anatomy, Fiber Pathways and Connectivity
Novel Imaging Acquisition Methods:
Multi-Modal Imaging 1
Keywords:
Acquisition
Cellular
Modeling
MRI PHYSICS
Myelin
Neuron
Open-Source Software
White Matter
1|2Indicates the priority used for review
Provide references using author date format
2. Bezanson, J. et al. (2017) ‘Julia: A fresh approach to numerical computing’, SIAM review, 59(1), pp. 65–98.
3. Layton, K.J. et al. (2017) ‘Pulseq: A rapid and hardware-independent pulse sequence prototyping framework’, Magnetic Resonance in Medicine, 77(4), pp. 1544–1552. doi:10.1002/mrm.26235.
4. Palombo, M., Alexander, D.C. and Zhang, H. (2019) ‘A generative model of realistic brain cells with application to numerical simulation of the diffusion-weighted MR signal’, NeuroImage, 188, pp. 391–402. doi:10.1016/j.neuroimage.2018.12.025.
5. Callaghan, R. et al. (2020) ‘Config: Contextual fibre growth to generate realistic axonal packing for diffusion mri simulation’, NeuroImage, 220, p. 117107. doi:10.1016/j.neuroimage.2020.117107.
6. Ginsburger, K. et al. (2019) ‘MEDUSA: A GPU-based tool to create realistic phantoms of the brain microstructure using tiny spheres.’, Neuroimage, 193, pp. 10–24. doi:10.1016/j.neuroimage.2019.02.055.
7. Villarreal-Haro, J.L. et al. (2023) ‘CACTUS: a computational framework for generating realistic white matter microstructure substrates’, Frontiers in Neuroinformatics, 17. Available at: https://www.frontiersin.org/articles/10.3389/fninf.2023.1208073.