Loading...
DEVELOPING NOVEL MACHINE LEARNING APPROACHES FOR CLASSIFICATION OF MILD TRAUMATIC BRAIN INJURY
Vedaei, Faezeh
Vedaei, Faezeh
Citations
Altmetric:
Genre
Thesis/Dissertation
Date
2023-09
Advisor
Group
Department
Bioengineering
Subject
Permanent link to this record
Collections
Research Projects
Organizational Units
Journal Issue
DOI
http://dx.doi.org/10.34944/dspace/9506
Abstract
Traumatic brain injury (TBI) is one of the common brain diseases and a public health concern. It causes significant social burden affecting people’s quality of life depending on
the severity of the disease. Mild TBI (mTBI) is a type of TBI characterized with cognitive and emotional deficits within few weeks after injury that may persist for several years. The
chronic symptoms of Mubi called post-concussion syndrome may develop at some points in patients including cognitive impairment, memory loss, depression, irritability, lack of
concentration, and anxiety. Brain function alteration caused by mTBI can be detected using brain imaging including resting-state functional magnetic resonance imaging (rs-fMRI)
and positron emission tomography (PET). Machine learning (ML) algorithms have been used in neuroimaging to provide brain signatures of neurological and psychiatric disorders.
These approaches have been able to overcome the limitation of the conventional massunivariate group analysis methods and develop individually based imaging biomarkers of
brain conditions through multi-variate mathematics.
In the need of robust biomarkers in prediction of mTBI patients, this thesis provides several approaches using functional brain imaging in conjunction with ML methods developing brain patterns of mTBI. First, we present application of conventional ML method using support vector machine (SVM) in classification of mTBI patients from healthy controls (HCs) group using several rs-fMRI metrics. The SVM has been used in neuroimaging in identification of brain disorders promising robust classification accuracies. However, this method needs prior engineering of raw data to reduce the dimensionality of features and optimize the learning process. Lastly, to address the need to obtain more efficient automatic approach to classify patients from HCs, we present deep learning (DL) based strategy developing artificial neural
network (ANN) architecture to extract the optimal feature representations that contribute the most to the model training. Single, and multimodal data could be used as input features without need for prior feature engineering resulting in robust and less bias-prone patterns of brain function. Together, this approach can potentially be extended into clinical settings aiding in prediction of brain disorders at the level of individuals.
Description
Citation
Citation to related work
Has part
ADA compliance
For Americans with Disabilities Act (ADA) accommodation, including help with reading this content, please contact scholarshare@temple.edu