Files and links (1)
PDFOpen Access (License Unspecified), Open Access
Thesis
Cross-Sectional Cohort Separation of AV1451 Uptake Patterns using Voxel-wise Barycentric Discriminant Analysis (VoBADA)
Master of Science (M.S.), Drexel University
Sep 2015
DOI:
https://doi.org/10.17918/etd-6695
Abstract
Alzheimer's disease (AD) is expected to affect 1 out of 85 people by 2050, and with no currently-available disease-modifying treatment, understanding the underlying pathology which is hypothesized to bring about AD is critical for research and development of novel drugs to prevent the onset and progression of the disease. AD is believed to be caused by an accumulation of beta amyloid plaques in the grey cortex which brings about the destabilization of tau protein in neuronal microtubules, leading to apoptosis and regional atrophy. As definitive diagnosis of AD can only be made at autopsy, understanding disease pathology and developing tools to aid in more accurate, earlier diagnosis becomes imperative to developing disease-modifying treatments. The tau-specific PET tracer AV1451 is utilized to understand in vivo information regarding tau accumulation and propagation. To quantify the differential diagnostic capabilities of AV1451 in differentiating between AD and other clinical diagnoses, voxel-wise barycentric discriminant analysis (VoBADA), an extension of multi-block barycentric discriminant analysis (MUBADA) is utilized on a battery of 202 subjects. Groups were defined by paired amyloid status as assessed by Amyvid PET scans and by clinical diagnosis which categorized the patients as either cognitively normal, mild cognitive impairment (MCI), or Alzheimer's disease (AD). Old cognitively normal (OCN) and young cognitively normal (YCN) subjects were defined based on age 50. Final groups selected were amyloid negative OCN, MCI, & AD and amyloid positive MCI & AD. Images were normalized by grey cerebellum counts and vectorised using a whole brain mask. The factor space was generated by running a generalized principal component analysis (GPCA) on the mean centered data. A combination of the Scree test, percentage of total variance, and bootstrapping the decomposition revealed that the first dimension is the sole dimension worth investigating to determine variance between clinical groups. Projecting the individual subjects onto the factor space through fixed effects modeling and random effects modeling (through jackknifing) revealed the distribution of factor scores across the first dimension. A general trend of separation was seen between the amyloid negative groups, the amyloid positive MCIs, and the amyloid positive ADs, but two much variance within the amyloid positive groups prevented a significant separation from being realized. This was also reflected when calculating accuracy of group assignment: while fixed effects modeling showed a better-than chance classification through a Chi-square test (p = 0.0139), random effects modeling did not yield significant classification (p = 0.0528). To better focus on voxels which contribute to group separation, a partial inertia map was generated for the first dimension based on the "eigenbrain" from the decomposition with weights for each voxel corresponding to the degree of separation between groups for that voxel. To ensure that the weights were only corresponding to moving in a direction towards AD, the "eigenbrain" was bootstrapped and pseudo-standard scores were generated. Significant weights are found in the temporal, parietal, occipital, and frontal lobes. While the accuracy of distinguishing between groups is moderate in the random effects model, it only reflects the separation quality of the voxels within the whole brain. The significant voxels extracted from the bootstrapped "eigenbrain" can be used in further analyses to better classify subjects which fit an AD-like pattern of uptake. Focusing on these voxels in quantitative analyses and qualitative reads can better assist in the diagnosis of AD.
Metrics
34 File views/ downloads
40 Record Views
Details
- Title
- Cross-Sectional Cohort Separation of AV1451 Uptake Patterns using Voxel-wise Barycentric Discriminant Analysis (VoBADA)
- Creators
- Ian Andrew Kennedy - DU
- Contributors
- Michael D. Devous (Advisor) - Drexel University (1970-)
- Awarding Institution
- Drexel University
- Degree Awarded
- Master of Science (M.S.)
- Publisher
- Drexel University; Philadelphia, Pennsylvania
- Number of pages
- 48 pages
- Resource Type
- Thesis
- Language
- English
- Academic Unit
- School of Biomedical Engineering, Science, and Health Systems (1997-2026); Drexel University
- Other Identifier
- 6695; 991014632232904721