Accounting for Complex Structure in Diffusion Weighted Imaging Data using Volume Fraction Representations
The domain of diffusion weighted imaging (DWI) has come a long way since its initial development in the mid-1980s. Over the years, we have gained a better understanding of the accompanying techniques and necessary processing steps involved, furthermore leading to a wealth of new insights in the complex workings of the (human) brain. The introduction of diffusion tensor imaging (DTI) has played a crucial role in thisprocess, as it provided the first model intended to deal with anisotropic diffusion; a particular feature observed in the white matter (WM), asopposed to the other most common 'tissue' types found in the human brain, i.e. the gray matter (GM) and cerebrospinal fluid (CSF).
During the last decade, however, we have come to realize that the DTI model is severely lacking in its possibilities to represent voxels that contain so called 'crossing fibers', a general name that is often used to refer to a range of complex geometric fiber configurations caused by the partial volume effect.
In this PhD thesis, we intended to design certain representations of (information extracted from) DWI data that take into account the aforementioned variety of complex geometrical configurations. Our proposed novel representations aim to offer a greater flexibility that should inherently render many existing difficult problems (e.g. segmentation and registration) trivial; yet make as little assumptions aspossible on the nature of the data or the properties of the underlying structures.
A first major contribution is a generic framework for multi-shell multi-tissue (MSMT) representations, and a specific implementation tailored to represent WM, GM and CSF in the human brain. This representation was specifically designed to render the retransformation problem trivial. The latter was easily solved by a newly introduced preservation of principal volume fractions (PPVF) retransformation strategy.
A second major contribution is a more tangible track orientation distribution (TOD) representation for complex fiber track distributions. Our newly developed method to obtain such a TOD, is termed track orientation density imaging (TODI). This technique allowed us to gain further understanding in the amplitude of a short-tracks TOD, which can now be interpreted as a measure of track-like local support (TLS). Furthermore, we showed that employing the latter for TOD-based tractography results in guiding the tracks along directions that are morelikely to correspond to continuous structure over a longer distance; i.e. track-like structure!
Both representations (MSMT and the TOD) forthcoming from these major contributions, also allow for increased insightin many other aspects of the data they describe, and provide a large range of opportunities for future research.