Advancing White Matter Tractometry of the Brain Using Diffusion MRI and Machine Learning PDF Download

Author: Bramsh Qamar Chandio
Publisher:
ISBN:
Category : Brain
Languages : en
Pages : 0

Get Book

Book Description
The human brain contains billions of axons that bundle together in tracts and fasciculi. These can be reconstructed in vivo by collecting diffusion MRI data and deploying tractography algorithms. The outputs of tractography algorithms are called tractograms. These tractograms are represented digitally using streamlines, which are representations of 3D curves traversing the brain. Diffusion MRI and tractography provide crucial information about brain connectivity and microstructural changes due to underlying conditions such as Alzheimer's, Parkinson's, and Schizophrenia disease. However, often generated whole-brain tractograms have millions of streamlines with many false positives and anatomically implausible streamlines. Therefore, tractograms require novel processing pipelines that can reduce such issues and provide anatomically relevant outcomes. For example, a) bundle segmentation methods extract anatomically relevant streamlines and white matter tracts/bundles from the whole-brain tractograms. b) bundle registration methods are used to create common spaces across subjects, and c) statistical methods can then be applied to study microstructural changes in groups and populations along the length of the bundles. This process of quantifying microstructural changes due to a disease or condition along the length of the digitally reconstructed white matter tracts is called tractometry.In this dissertation, we introduced new methods to advance tractometry using machine learning and functional data analysis approaches. For the problem of bundle segmentation and streamline filtering, we introduced the auto-calibrated RecoBundles method that precisely extracts bundles from tractograms with only one reference exemplar. We also developed an unsupervised method, FiberNeat, that filters out spurious streamlines from bundles in latent space. To solve the registration problem, a novel method, BundleWarp, was created for the nonlinear registration of white matter bundles where users can control the amount of deformations with a single free regularization parameter (Lambda). In the category of tractometry methods, we created a publicly available advanced tractometry pipeline called BUndle ANalytics (BUAN). BUAN provides a completely automatic, end-to-end streamline-based solution that connects bundle segmentation, registration, analysis of bundle anatomy, and bundle shape analysis. BUAN reports the exact locations of population differences along the length of the tracts. BUAN also includes metrics and methods for quality assurance of extracted white matter tracts in large populations. Furthermore, in BUAN 2.0, instead of treating points on the streamlines as independent observations in statistical analysis, we proposed using functional data analysis (FDA) methods where each streamline is considered a function. This dissertation moves beyond the standard processing of brain images to a tractography-based analysis of the brain tissue microstructure and connectivity by introducing robust, fast, and simple-to-use algorithms. Results are shown on Parkinson's disease data from Parkinson's Progression Markers Initiative (PPMI) and Alzheimer's disease from Alzheimer's Disease Neuroimaging Initiative phase 3 (ADNI3) datasets. The methods developed as part of this dissertation are made publicly available through DIPY.org.