Alexandria Digital Research Library

Multi-Output Multi-Modal Parts-Based Regression for High Dimensional Data with Low Sample Size

Author:
Joshi, Swapna
Degree Grantor:
University of California, Santa Barbara.Electrical & Computer Engineering
Degree Supervisor:
ManjunathGrafton B.SScott
Place of Publication:
[Santa Barbara, Calif.]
Publisher:
University of California, Santa Barbara
Creation Date:
2012
Issued Date:
2012
Topics:
Electrical engineering and Medical imaging and radiology
Keywords:
subspace analysis
pattern recognition
regression
computer vision
medical image analysis
machine learning
Description:

It is generally hypothesized that there are significant differences in the structural anatomy of the brains of normal people when compared to those who are diagnosed as psychopaths, yet very little quantitative data exists. This research addresses various computer vision and medical imaging applications, mainly focusing on the problem of correlating clinically assigned psychopathic scores (called PCL-R scores) with Magnetic Resonance Imaging (MRI) brain scans of the subjects.

We propose a novel data-driven parts-based regression algorithm for the analysis of such cross-sectional anatomical data. The method helps capture and localize biologically significant parts in the brain exhibiting regression with respect to the associated clinical variable. This analysis is further extended to the case of functional MRI brain scans wherein we present a new multi-modal regression method to capture the correlating

anatomical parts between the modalities that are undergoing changes due to the clinical variable. Finally, a formulation of the regression method is provided to learn the complex relationship between each data modality and multiple clinical labels associated with them.

Format:
Text
Collection(s):
UCSB electronic theses and dissertations
ARK:
ark:/13030/m5q81drv
Merritt ARK:
ark:/13030/m5q81drv
Rights:
Inc.icon only.dark In Copyright
Access: At the request of the author, this item is currently restricted via delayed access (embargo), and may become available at a later date. Please check our FAQs for more information.