Back to Projects

AI-enabled Imaging, Emerging Imaging

High-resolution in vivo microstructure imaging using targeted 3D diffusion MRI

Project ID: 2021_021

1st Supervisor: J-Donald Tournier, King’s College London
2nd Supervisor: David Carmichael, King’s College London
Clinical Champion: Alexander Hammers, King’s College London

Aim of the PhD Porject:

  • The aim is to develop methods to acquire high-quality, high-resolution, eloquent diffusion MRI data at high field to investigate tissue microstructure.
  • Particularly useful for studies into disorders such as epilepsy, where resection of subtle cortical malformations is an effective therapy, but these malformations can be difficult to identify.

Project Description / Background:

Diffusion MRI (dMRI) is unique in its ability to provide information related to the microstructure of the tissue in vivo non-invasively, and has consequently been the focus of intense research efforts over the last few decades. dMRI is currently primarily used to study white matter due to its sensitivity to the orientation of the coherently oriented bundles of axonal fibres that project through it. This unique property means dMRI can be used to infer the orientations of the white matter fibres, providing a means to delineate the path of long-range white matter tracts in the brain in vivo, via so-called tractography techniques.

More recently, there has been renewed interest in using dMRI to probe and assess the microstructural properties of the tissue. A particularly interesting avenue of investigation is for imaging of the human cortex, which has to date remained relatively unexplored with dMRI. This is due primarily to the difficulty of obtaining images of sufficient resolution to observe the patterns that might be expected due to the different layers, which would require sub-millimetre resolution. Studying the cortex would be particularly interesting in the context of epilepsy, as the seizure focus is typically associated with subtle malformations of cortical development that can be difficult to identify with conventional imaging. It would also be of interest in studies of cortical development during the perinatal period, when the cortex undergoes rapid and profound changes in layering and dendritic arborisation.

Acquiring high quality dMRI data is difficult due to its exquisite sensitivity to microscopic motion, which affects in vivo data due to cardiac pulsation and involuntary subject motion. Motion during the diffusion-weighting preparation induces random phase variations throughout the sample, making it difficult to recombine shots acquired from multiple excitations without introducing ghosting artefacts.

In this project, we propose to address this issue by acquiring high-resolution 3D data from a small region of interest. This would consist of a 3D pulsed gradient spin-echo (PGSE) sequence with inner volume selection, with each excitation providing full k-space sampling along the read-out (x) axis, while sub-sampling both phase-encode (y) and partition (z) axes. Additional methods will be required to stabilise the echo train in the presence of the expected phase instabilities [1,2]. The reconstruction of the data would invert the forward model shown in Figure 1, and exploit the redundancy provided by parallel imaging methods to allow estimation of the per-shot phase variations, thereby allowing artefact-free reconstruction of the 3D volume. A simple iterative proof of principle reconstruction algorithm (not shown here), based on estimating the image from current estimates of the phase maps, then estimating the phase maps based on the current estimate of the image, shows a 10-fold reduction in the original artefact level. Additional redundancy can be exploited based on the expected correlations between volumes to further regularise the problem.


  1. Le Roux P, McKinnon G, Yen Y-F, Fernandez B. Realignment capability of the nCPMG sequence. J. Magn. Reson. 2011;211:121–133.
  2. Pipe JG, Huo D, Li Z, Aboussouan EA. Consistent Signal for non-CPMG echo trains. In: Proceedings of the International Society for Magnetic Resonance in Medicine. Hawai’i, USA: 2009. p. 164.
  3. Miller KL, Pauly JM. Nonlinear phase correction for navigated diffusion imaging. Magn. Reson. Med. 2003;50:343–353.
  4. Andersson JLR, Sotiropoulos SN. Non-parametric representation and prediction of single- and multi-shell diffusion-weighted MRI data using Gaussian processes. NeuroImage 2015;122:166–176.
Figure 1. The description is in the caption.

Figure 1: The model for diffusion-weighted image data generation in parallel acquisition mode.

Back to Projects