Doctoral Dissertations

Sima Azizi

Abstract

“Mathematical models can be combined with deep learning and machine learning methods to provide new insights in neuroscience. The field of neuroscience is characterized by rich datasets that include fluid biomarkers, EEG signals, and advanced neuroimages. Recent advances in natural language processing have led to the opportunity to gain additional insights from rapidly growing text databases as well as electronic health records. In this research, we focus on applying computational intelligence methods to the analysis of three different complex data sources: blood levels of disease biomarkers, EEG signals from schizophrenic patients, and disease phenotypes encoded in electronic health records. First, we have developed a kinetic model to describe the rise and fall of blood biomarkers after mild traumatic brain injury. The model is based on a pharamacokinetic model of oral drug absorption into the bloodstream. We used published papers to estimate the model parameters and performed a Monte Carlo simulation to characterize uncertainty about model predictions. In a companion paper, we have extended the kinetic model to steady state blood levels of biomarkers in patients with a variety of neurological conditions. Second, we utilized logistic regression to discriminate schizophrenia patients from healthy controls using EEG signals. We converted sensor level resting state EEG signals to source level EEG signals utilizing a partial differential equation. We found that classification results were superior in the source space. Third, we developed a pipeline based on a deep learning model followed by a lookup table to create a high throughput clinical phenotyping method. We trained a named entity recognition model using a BERT transformer and annotated text. We mapped the entities predicted by the model to machine readable codes in an ontology using a lookup table. Our intention was to extend NLP methods to recognize neurological concepts in text as either a named entity itself or the description of the named entity. NLP named entity recognition methods show promise for this task”--Abstract, page iv.

Advisor(s)

Wunsch, Donald C.

Committee Member(s)

Ferdowsi, Mehdi
Beetner, Daryl G.
Alsharoa, Ahmad
Hier, Daniel
Liu, Jinling

Department(s)

Electrical and Computer Engineering

Degree Name

Ph. D. in Electrical Engineering

Comments

The journal articles included in this dissertation are presented in a different order than outlined on the Publication Dissertation Option page.

Publisher

Missouri University of Science and Technology

Publication Date

Fall 2021

Journal article titles appearing in thesis/dissertation

  • A Kinetic Model for Blood Biomarker Levels after mild Traumatic Brain Injury
  • A Kinetic Model to Relate Steady State Blood and CSF Levels of Neurologic Biomarkers
  • Schizophrenia classification using resting state EEG functional connectivity: Source level outperforms sensor level
  • High Throughput Deep Clinical Phenotyping Using Transformers

Pagination

xiii, 107 pages

Note about bibliography

Includes bibliographic references.

Rights

© 2021 Sima Azizi, All rights reserved.

Document Type

Dissertation - Open Access

File Type

text

Language

English

Thesis Number

T 11937

Download

Share

 
COinS