STRUCTURAL CHARACTERIZATION OF HUMAN BRAINSTEM NETWORKS CRITICAL TO CONSCIOUSNESS WITH HIGH- AND LOW-FIELD DIFFUSION TENSOR IMAGING
Name
olchanyi-olchanyi-phd-hst-2026-thesis.pdf
Size
59.06 MB
Format
Adobe PDF
Checksum (MD5)
db5e07df044c296965044043aef227d3
Author(s)
Olchanyi, Mark David
Advisor(s)
Brown, Emery N.
Edlow, Brian L.
Date Issued
May 2026
Publisher
Massachusetts Institute of Technology
Abstract
The brainstem is a compact brain structure that houses circuitry for the initiation and modulation of many functions vital to life. However, its size, location deep within the brain, and dense architecture of interwoven gray and white matter structures make it difficult to assess with noninvasive imaging methods. A central goal of this thesis is to provide segmentation methods for structures within the brainstem for use across various magnetic resonance imaging (MRI) modalities. First, I introduce a Bayesian segmentation strategy for gray matter regions within the brainstem that serve as nodes of the ascending arousal network (AAN) that is based on a probabilistic atlas generated from \textit{ex vivo} imaging of human brain specimens. I also introduce a companion algorithm that uses a convolutional neural network model to segment brainstem white matter bundles, a subset of which contain axonal connections to, from and between AAN nuclei. This algorithm relies on white matter contrast and brainstem-specific tractography reconstructions in diffusion tensor imaging (DTI)/diffusion MRI sequences. Second, I propose a DTI sequence and correction/enhancement framework for portable, ultra-low-field (ULF) MRI hardware. This includes a Bayesian model for the correction of direction-specific bias fields unique to ULF DTI coupled with a spatio-angular superresolution algorithm to recover white matter contrast degraded by poor spatial and angular resolutions, low signal-to-noise ratios, and presence of extensive artifacts. The latter superresolution method is termed "DiffSR", and is by extension meant to be applicable to DTI/dMRI scans of any quality. Collectively, the methods described in this thesis aim to provide more accurate segmentation and evaluation of deep brain structures, particularly the brainstem, and expand the utility of DTI/dMRI for studying white matter across a range of modalities and various levels of scan quality.
MIT Department
Harvard-MIT Program in Health Sciences and Technology
Terms of Use
Attribution 4.0 International (CC BY 4.0)
Copyright retained by author(s)
Persistent DSpace Link
