Learning more with less data using domain-guided machine learning: the case for health data analytics

Varatharajah, Yogatheesan

Permalink

https://hdl.handle.net/2142/109413 Copy

Description

Title

Learning more with less data using domain-guided machine learning: the case for health data analytics

Author(s)

Varatharajah, Yogatheesan

Issue Date

2020-12-02

Director of Research (if dissertation) or Advisor (if thesis)

Iyer, Ravishankar

Doctoral Committee Chair(s)

Iyer, Ravishankar

Committee Member(s)

• Hasegawa-Johnson, Mark
• Koyejo, Sanmi
• Robinson, Gene
• Worrell, Gregory

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Machine Learning
• Neurology
• Artificial Intelligence
• Domain-guided
• Healthcare
• Signal Processing
• Interpretability
• Intelligence Augmentation

Abstract

The United States is facing a shortage of neurologists with severe consequences: a) average wait-times to see neurologists are increasing, b) patients with chronic neurological disorders are unable to receive diagnosis and care in a timely fashion, and c) there is an increase in neurologist burnout leading to physical and emotional exhaustion. Present-day neurological care relies heavily on time-consuming visual review of patient data (e.g., neuroimaging and electroencephalography (EEG)), by expert neurologists who are already in short supply. As such, the healthcare system needs creative solutions that can increase the availability of neurologists to patient care. To meet this need, this dissertation develops a machine-learning (ML)-based decision support framework for expert neurologists that focuses the experts’ attention to actionable information extracted from heterogeneous patient data and reduces the need for expert visual review. Specifically, this dissertation introduces a novel ML framework known as domain-guided machine learning (DGML) and demonstrates its usefulness by improving the clinical treatments of two major neurological diseases, epilepsy and Alzheimer’s disease. In this dissertation, the applications of this framework are illustrated through several studies conducted in collaboration with the Mayo Clinic, Rochester, Minnesota. Chapters 3, 4, and 5 describe the application of DGML to model the transient abnormal discharges in the brain activity of epilepsy patients. These studies utilized the intracranial EEG data collected from epilepsy patients to delineate seizure generating brain regions without observing actual seizures; whereas, Chapters 6, 7, 8, and 9 describe the application of DGML to model the subtle but permanent changes in brain function and anatomy, and thereby enable the early detection of chronic epilepsy and Alzheimer’s disease. These studies utilized the scalp EEG data of epilepsy patients and two population-level multimodal imaging datasets collected from elderly individuals.

Graduation Semester

2020-12

Type of Resource

Thesis

Permalink

http://hdl.handle.net/2142/109413

Copyright and License Information

Copyright 2020 Yogatheesan Varatharajah

Owning Collections