Master of Engineering Science
de Ribaupierre, Sandrine
Surgical management for hydrocephalus is among the most common procedures performed by pediatric neurosurgeons. However, how to best predict postoperative outcomes is unknown. Neuroimaging studies could provide insight, though working with these images is non-trivial. This thesis aims to 1) evaluate registration and preprocessing methodologies to best prepare data for comparisons, and 2) assess the impact of postoperative lateral ventricle volume (LVV) as a predictor of white matter health in networks that are dysregulated in hydrocephalus patients. We found that skull-stripped, bias corrected images with the SyN algorithm produced most accurate registration. We also found large dysregulated white matter networks in patients, and postoperative LVV did not have a large impact in predicting these networks. Overall, these studies suggest an image processing pipeline for pathological pediatric images and adds to the knowledge surrounding both the impact of pediatric hydrocephalus on white matter networks and the association with postoperative LVV.
Summary for Lay Audience
Hydrocephalus is a neurological disease that occurs in approximately 0.1% of births world-wide and is characterized by increased cerebrospinal fluid in the ventricles of the brain. Treatment for hydrocephalus involves redirecting the excess fluid from the ventricles to another part of the body, for example the peritoneum which is seen in the ventriculoperitoneal (VP) shunt. Despite being one of most common surgeries performed in children, there are many questions surrounding how to best predict postoperative outcomes. Utilizing neuroimaging techniques can allow us to better understand the disease. One challenge when working with neuroimages of children with hydrocephalus is that there can be difficulty when trying to compare these neuroimages with those of healthy children (e.g., normalization). This difficulty arises from morphological differences such as large ventricles, other pathologies in the brain, and treatment related non-correspondence (i.e., the VP shunt). In the first study, various different image preprocessing steps and normalization algorithms were assessed. It was found that images that were bias corrected as well as skull-stripped had better normalization accuracy relative to those that were not, and the best performing algorithm was SyN by Advanced Normalization Tools. In the case wherein these preprocessing steps are not possible, the DARTEL toolbox also performed with relatively high accuracy. In the second study, taking a whole-brain approach, white matter integrity and connectivity were compared between pediatric patients with hydrocephalus and healthy controls. Furthermore, postoperative ventricle volume was explored as a predictor for the white matter metrics in hydrocephalus patients. We found a series of large, dysregulated networks in patients with hydrocephalus relative to controls, many suggested decreased white matter integrity, and decreased white matter connections. Most networks involved subcortical structures and those outside the frontal lobes. After correction for multiple comparisons only white matter metrics in two streamlines were predicted by ventricle volume. In summary, this thesis adds to the current knowledge of image processing pipelines for pathological pediatric images, and both the impact of pediatric hydrocephalus on white matter networks and the association with postoperative lateral ventricle volume.
Ragguett, Renee-Marie, "Quantitative Image Analysis of White Matter Dysregulation Using Brain Normalization for Diagnostic Analysis of Pediatric Hydrocephalus" (2022). Electronic Thesis and Dissertation Repository. 8575.
Available for download on Thursday, June 01, 2023