Home
Resource
Advanced computational analysis of neuroimaging data for brain injury identification and decoding behavior
PDF
PDF format is widely accepted and good for printing.
Plug-in required
Access to this PDF has been restricted at the author's request. It will be publicly available after February 2, 2025.
Citation & Export
View Usage Statistics
Staff View

Citation & Export
Hide

Simple citation

Salsabilian, Shiva. Advanced computational analysis of neuroimaging data for brain injury identification and decoding behavior. Retrieved from https://doi.org/doi:10.7282/t3-2awv-k773

Export

Click here for information about Citation Management Tools at Rutgers.

Statistics
Hide

Description

TitleAdvanced computational analysis of neuroimaging data for brain injury identification and decoding behavior
NameSalsabilian, Shiva (author); Najafizadeh, Laleh (chair); Chen, Yingying YC (member); Zhang, Yuqian YZ (member); Haghani, Sasan SH (member); Margolis, David DM (member); Rutgers University; School of Graduate Studies
Date Created2023
Other Date2023-01 (degree)
SubjectElectrical engineering, Biomedical engineering, Computer science, Adversarial variational autoencoder, Applied machine learning, Cross-subject transfer learning, Mild traumatic brain injury identification, Neural decoding of behavior, Widefield calcium imaging
Extent148 pages : illustrations
DescriptionUnderstanding how the brain functions have been one of the major goals of neuroscience. To approach this challenging topic, artificial intelligence (AI) and various computational techniques have become valuable tools in analyzing and processing the data obtained from the brain. In this work, we present new computational methods applied to data obtained via widefield calcium imaging from transgenic mice, to study several problems in two domains: 1) how does the brain alter, in terms of functionality, after the occurrence of a mild traumatic brain injury (mTBI) and utilize it for early mTBI identification, and 2) how can behavior be inferred from brain recordings and identifying behavior from neural recordings.
The first part of this work focuses on problems related to mTBI diagnosis. Early diagnosis of mTBI is challenging, yet significantly important in order to grant the patients with timely treatment and mitigate the risks of possible long-term psychiatric and neurological disorders. Via a longitudinal study, we investigate how the brain’s functional networks change following an injury and identify the links that significantly contribute to the alternation in brain networks. Next, we employ graph theoretical analysis along with statistical analysis and machine learning, to identify the network measures that can be used as biomarkers for differentiating injured and healthy brains. We then propose to automate the process of feature extraction and develop an mTBI-detection framework based on graph embedding features, extracted via the Node2Vec algorithm, combined with convolutional neural networks. Finally, we address the problem of subject variability by developing an autoencoder feature learning model and propose a data-driven framework to extract subject-invariant representations from brain recordings to perform cross-subject transfer learning for mTBI identification.
The second part of this work focuses on problems related to inferring behavior from neuroimaging data. We first investigate changes in the brain's functional networks under various behavioral conditions and identify the links that contribute to the differences in the networks corresponding to different behavioral conditions. We then propose to utilize an adversarial variational encoder model in a two-step data-driven approach to extract cross-subject feature representations from neural activity in order to decode subjects’ behavior choices. Finally, to automate the analysis of behavior for unsupervised neural decoding, we propose a data-driven self-supervised LSTM-based adversarial variational autoencoder framework addressing two main challenges that exist in relating neural data and behavior: 1) automatic creation of labels from behavioral observations, and 2) dealing with subject variability and finding a model that works across individuals.
NotePh.D.
NoteIncludes bibliographical references
Genretheses
Persistent URLhttps://doi.org/doi:10.7282/t3-2awv-k773
LanguageEnglish
CollectionSchool of Graduate Studies Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.
Version 8.5.5
Rutgers University Libraries - Copyright ©2024