Back to Projects

AI-enabled Imaging

Investigating the development of the brain and its microstructure using diffusion MRI

Project ID: 2019_017

1st supervisor: J.-Donald Tournier, King’s College London
2nd supervisor: Emma Robinson, King’s College London

Diffusion Magnetic Resonance Imaging (dMRI) is a non-invasive in vivo imaging method sensitive to the random, thermally-driven motion of water molecules over cellular length scales. Since the motion of water is affected by cellular barriers and membranes, this technique provides a unique means of investigating the microstructure of living tissue non-invasively. The KCL-led developing Human Connectome Project (dHCP; currently underway at St Thomas’ Hospital has already collected data from more than 700 term and pre-term neonates, and over 100 fetuses, using advanced acquisition methods tailored specifically for this project [1]. In addition to high-resolution anatomical and functional MRI data, high quality multi-shell high-angular resolution dMRI (HARDI) data are acquired for each subject, along with a large amount of clinical, genetic, and behavioural information, including 2 year outcomes and beyond. This provides one of the richest, highest-quality datasets in this difficult-to-image population, with huge potential for novel discoveries.

There are however many challenges in the analysis of advanced dMRI data, particularly in this cohort. To date, most approaches for the analysis of dMRI data rely on biolophysical models of tissue microstructure, primarily developed for the healthy adult population (e.g. [2], [3]). These are ill-suited to the study of tissue microstructure during the fetal and neonatal period, where the tissue is still very much developing and poorly represented using current models. A more promising approach, and the focus of this project, is to learn the relevant features of the signal using a fully data-driven approach, to characterise the signal in terms of its angular and radial features (in other words, how the signal varies with degree and orientation of the diffusion weighting), and the variation of these features both across the brain and during development, in a way that it unbiased by the imposition of any particular a priori model of microstructure. These features can then be assessed by comparison with the current literature on brain development, and also correlated with long-term outcomes.

This builds on our previous work on data-driven modelling of so-called multi-shell dMRI data [4], [5], coupled with our work on time-resolved group-wise analysis of diffusion MRI data [6]. This project will use advanced machine learning techniques to learn a compact representation of the signal and its evolution during development, coupled with biophysically-motivated prior knowledge and constraints, and in this way identify the most salient characteristics of the developing brain. This will then be compared to current anatomical knowledge about the various microstructural changes that are known to take place at different locations and at various times during this time period, to enable interpretation of the signal from a microstructural point of view. This in itself would provide crucial insights into how to interpret the vast amount of information that dMRI provides for each subject.

This framework will also be used to identify features any features in the signal that are predictive of later outcome [7]. This is particularly important in the management of preterm birth, since accurate prognosis is currently problematic.

[1] J. Hutter et al., “Time-efficient and flexible design of optimized multishell HARDI diffusion,” Magn. Reson. Med., May 2017.
[2] H. Zhang, T. Schneider, C. A. Wheeler-Kingshott, and D. C. Alexander, “NODDI: practical in vivo neurite orientation dispersion and density imaging of the human brain,” NeuroImage, vol. 61, no. 4, pp. 1000–1016, Jul. 2012.
[3] D. S. Novikov, J. Veraart, I. O. Jelescu, and E. Fieremans, “Rotationally-invariant mapping of scalar and orientational metrics of neuronal microstructure with diffusion MRI,” NeuroImage, vol. 174, pp. 518–538, Jul. 2018.
[4] B. Jeurissen, “Multi-shell multi-tissue constrained spherical deconvolution,” 2008, pp. 1–37.
[5] D. Christiaens et al., “Learning compact q-space representations for multi-shell diffusion-weighted MRI,” IEEE Trans. Med. Imaging, pp. 1–1, 2018.
[6] M. Pietsch et al., “A framework for multi-component analysis of diffusion MRI data over the neonatal period,” NeuroImage, Oct. 2018.
[7] E. C. Robinson, A. Hammers, A. Ericsson, A. D. Edwards, and D. Rueckert, “Identifying population differences in whole-brain structural networks: a machine learning approach,” NeuroImage, vol. 50, no. 3, pp. 910–919, Apr. 2010.
[8] T. T. Vu, B. Bigot, and E. S. Chng, “Combining non-negative matrix factorization and deep neural networks for speech enhancement and automatic speech recognition,” in 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2016, pp. 499–503.

Back to Projects