< Terug naar vorige pagina

Publicatie

Tensor-Based Global-to-Local Morphometric Analyses in Neurodegenerative Diseases

Boek - Dissertatie

Tensor-based morphometry (TBM) is a technique to identify morphological differences among different groups of interest based on structural MRI images. Specifically in this work, morphological analyses can provide valuable information about the form altering effects of neurodegenerative diseases, such as Alzheimer and frontotemporal dementia, in the human brain. This can result in a better understanding of the disease impact and development of image-based diagnostic tools that are essential towards disease-modifying therapies. However, standard tensor-based methods have difficulties identifying the subtle volumetric changes occurring at early stages of disease development. Those methods analyse group differences within the sizes of pre-defined and non-overlapping regions and interpret the statistical results in an independent region-wise manner. In contrast, this work proposes a fundamental extension to TBM based on a global-to-local tensor-based image segmentation and a comprehensive inference testing framework that allows to expose the complexity of volumetric interactions affected by neurodegenerative diseases at different levels of detail. The first contribution of this work is providing a TBM-based data-driven image segmentation that partition the brain into regions with correlated volumetric variations, based on the volumetric variability observed within a given population. Additionally, the segmentation is defined hierarchically with various global-to-local levels of spatial coverage. In its result, the complete brain partitioning constitutes of different global-to-local levels with overlapping regions across different levels and with non-overlapping regions in each level that are well-integrated internally and less integrated with other regions. The second contribution is a complete assessment of morphological variations within each of the regions and integrated across different levels in the hierarchical chain. The completeness is obtained by (1) analysing inter-group inferences on different aspects of volumetric patterns in the regions such as integrated univariate volumetric changes (size) and multivariate patterns of volumetric covariation (shape), (2) analysing inferences at different levels of detail and (3) the propagation of results across those levels in the global-to-local hierarchy. Alzheimer's disease and frontotemporal dementia are common neurodegenerative diseases and are specifically investigated in this work. The proposed segmentation technique and hypothesis testing framework were applied to study early development of both diseases. We localised size and shape effects at different locations in the brain and with different global-to-local penetrations of the effects. The completeness of the proposed inference framework adds new information to the TBM analyses for both disease groups. Our method showed a strong advantage over conventional TBM analyses in the way that it adapted an exhaustive, but complete, set of inference analyses into a hierarchically form integrated structure that is easy to understand. The proposed extension to TBM did not compromise on statistical power and allowed us to derive a single compact test statistic, through a local-to-global propagation of the inference results.
Jaar van publicatie:2018
Toegankelijkheid:Open