Main content

Voxel-Wise Image Analysis for White Matter Hyperintensity Segmentation

Show full item record

Title: Voxel-Wise Image Analysis for White Matter Hyperintensity Segmentation
Author: Knight, Jesse
Department: School of Engineering
Program: Engineering
Advisor: Taylor, Graham
Abstract: White matter hyperintensities (WMH) are regions of increased pixel intensity in T2-weighted MRI which are correlated with several neurodegenerative diseases. Human segmentation of WMH is time consuming and inconsistent, motivating automation of WMH segmentation. While many algorithms for this task have previously been proposed, few have been validated on MRI from different sources, despite the sensitivity of most algorithms to source-specific image features. This thesis presents a segmentation algorithm called “Voxel-Wise Logistic Regression” (VLR), which provides both good interpretability and segmentation performance. VLR uses FLAIR MRI to estimate the WMH class probability image using spatially varying logistic parameters β(x). These “parameter images” also concisely summarize the model class discrimination. Additionally, a validation framework called “Leave-One-Source-Out Cross Validation” (LOSO-CV) is introduced, which provides more realistic estimation of model performance on “never-before-seen” MRI sources. Segmentation performance of the VLR model under LOSO-CV is presented using 96 open-source images from 7 MRI sources.
Date: 2017-12
Rights: Attribution 2.5 Canada
Terms of Use: All items in the Atrium are protected by copyright with all rights reserved unless otherwise indicated.

Files in this item

Files Size Format View Description
Knight_Jesse_201712_MASc.pdf 10.36Mb PDF View/Open thesis

This item appears in the following Collection(s)

Show full item record

Attribution 2.5 Canada Except where otherwise noted, this item's license is described as Attribution 2.5 Canada