Robust quantification of diffusion kurtosis parameters.

Diffusion-weighted magnetic resonance imaging is a non-invasive technique to reveal the brain's microstructural properties by probing the local diffusion of water molecules. Fitting mathematical models to diffusion MRI data allows to extract quantitative information and, among these models, the diffusion tensor imaging (DTI) model is the most commonly applied. However, recent literature has shown that the diffusion kurtosis imaging (DKI) model can provide more accurate estimates of diffusion tensor properties as well as additional information in clinical applications. Unfortunately, the quality of the diffusion metrics extracted from these models is degraded by several acquisition artefacts, such as Gibbs ringing, eddy current distortions and susceptibility-induced artefacts. Besides these well-known artefacts, voxels in DW images may suffer from additional problems: ● signal intensity outliers resulting from motion, cardiac pulsation or system instabilities can compromise the parameter estimates to an extent that they are no longer useful; ● image voxels are relatively large (2 to 3 mm isotropic) and thus susceptible to partial volume effects, which is particularly a problem in brain images when cerebrospinal fluid contamination occurs making the interpretation of diffusion markers ambiguous and no longer tissue-specific. A first aim of this project is to improve and validate an outlier-robust framework for diffusion and kurtosis parameter estimation. During the initial phase of the PhD project, the performance of such a framework was assessed in simulation experiments, thereby ignoring spatial correlations of outliers. As a logical step for improving the method, prior information on how outliers correlate within a slice will be included. Subsequently, a validation study will be performed to assess the reproducibility of DKI metrics in real test-retest datasets. In the second and third year of the PhD project, an advanced bi-compartment model based on the combination of diffusion and relaxometry data has been proposed for correcting free water contamination in multi-shell multi-echo diffusion data. This work has resulted in a journal paper that will be submitted in the second quarter of 2021. This promising approach exploits the combination of diffusion and relaxometry data as a rich source of information, but is not applicable to datasets acquired with a single echo time, which are typically acquired in clinical practice. For this reason, our next research goal is to implement and validate approaches for partial volume correction in single/multi-shell single-echo acquisitions. For this purpose, the potential of artificial intelligence solutions will be explored to deal with the ill-conditioned parameter estimation problem. Finally, as part of the Horizon 2020 initial training network (ITN) B-Q MINDED, the ultimate aim of the project will be to integrate the developed techniques in a regulatory approved quantitative MR product that can be used in clinical trials and, in a later stage, in daily clinical practice for improved assessment of drug efficacy and patient follow-up.

Keywords:MAGNETIC RESONANCE IMAGING (MRI)

Disciplines:Health information systems of medical informatics