Less Confusion in Diffusion MRI

With its unique ability to investigate tissue architecture and microstructure in vivo, diffusion MRI (dMRI) has gained tremendous interest and the society has been continuously triggered to develop novel dMRI image analysis approaches. With the overwhelming amount of strategies currently available it is unfortunately not always evident to the end-users how dMRI can be optimally used to address their application. In addition, differences in processing strategies lead to ambiguities as to which conclusions can reliably be drawn from dMRI data, resulting in controversies in the field. Such issues hamper a smooth transition of dMRI processing strategies into useful tools for applications. This thesis contributes to reducing the confusion in diffusion MRI by scrutinizing different steps of the processing pipeline. It focuses on making the topics accessible for a broad audience, and new methodology is proposed to make more intuitive and data-driven choices in dMRI data processing, to facilitate interpretation and visualization of dMRI data, and to investigate fundamental topics such as variability in characteristics of the dMRI signal and the geometrical organization of the brain pathways. Chapter 2 introduces the different steps of the dMRI processing pipeline and reviews the most commonly used and state-of-the-art dMRI processing techniques with a focus on the brain. Chapter 3 describes the MASSIVE (Multiple Acquisitions for Standardization of Structural Imaging Validation and Evaluation) brain dataset containing multi-modal MR data and 8000 dMRI volumes of a single healthy subject. Subsets of the MASSIVE dataset can serve as representative test beds for the development of new dMRI processing techniques. In Chapter 4 a robust parameter estimation procedure is proposed coined REKINDLE (Robust Extraction of Kurtosis INDices with Linear Estimation). By means of fast reweighted linear estimation of the diffusion kurtosis model, REKINDLE aims to identify and exclude outliers. Chapter 5 describes a data-driven framework that recursively finds single fiber population (SFP) voxels to calibrate the response function for spherical deconvolution, aiming at improved estimation of the fiber orientation distribution function. In Chapter 6, the recursive framework proposed in Chapter 5 is used to localize and characterize SFPs in multiple subjects and tracts. Chapters 7 and 8 focus on a recent debate on the existence of ‘sheet structures’ in the brain. It was proposed that pathways consistently cross each other orthogonally on surfaces somewhere along their trajectory. Others stated that these sheet structures are likely artifacts mainly based on qualitative findings. In Chapter 7, condition for sheet structure is recapitulated and a method to quantify a sheet probability index (SPI) from the data is proposed. Whereas the method in Chapter 7 requires the reconstruction of many pathways with tractography, Chapter 8 proposes a different method to calculate the SPI that does not rely on tractography and is less computationally intensive. Chapter 9 proposes a novel fiber tractography visualization approach that interactively and selectively adapts the transparency rendering of fiber trajectories based on their local or global orientation. This allows for improved 3D visualization and exploration of the fiber network. Chapter 10 discusses the findings in this thesis