Generative digital twins for longitudinal simulation and augmentation of multi-modal patient data

Theodorou, Brandon

Permalink

https://hdl.handle.net/2142/129388 Copy

Description

Title

Generative digital twins for longitudinal simulation and augmentation of multi-modal patient data

Author(s)

Theodorou, Brandon

Issue Date

2025-04-10

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

Sun, Jimeng

Doctoral Committee Chair(s)

Sun, Jimeng

Committee Member(s)

• Banerjee, Arindam
• Rehg, Jim
• Xiao, Cao

Department of Study

Siebel School Comp & Data Sci

Discipline

Computer Science

Degree Granting Institution

University of Illinois Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Machine learning in healthcare
• generative modeling
• synthetic data
• digital twins

Abstract

Generative modeling and digital twin technologies have emerged as transformative approaches to addressing critical challenges in healthcare, particularly regarding data availability, privacy, completeness, and quality. While machine learning has demonstrated significant potential in areas such as patient outcome prediction, drug discovery, and clinical trial optimization, its widespread application in real-world clinical settings remains constrained by the inherent limitations of medical data. These limitations include fragmented data repositories, modality-specific gaps, missing data, and stringent privacy requirements, all of which restrict data sharing and integration. My PhD research addresses these fundamental challenges by developing innovative generative digital twin frameworks designed to robustly simulate, repair, augment, and enhance multi-modal patient data. This research comprises several key methodological contributions: (i) Generation of privatized synthetic electronic health records (EHRs) to facilitate data sharing without compromising data quality or realism; (ii) Enhancement of existing EHR datasets to address inherent biases and quality concerns through advanced generative modeling; (iii) Generative augmentation techniques specifically tailored for medical imaging to repair and complete datasets via conditional simulation, improving downstream analytical performance; and (iv) Creation of universal, modality-agnostic representations of medical imaging data that enable robust model development despite heterogeneous and limited datasets. Collectively, these generative digital twin approaches significantly improve the usability and integrity of real-world multi-modal healthcare data. By overcoming critical barriers related to data incompleteness, fragmentation, and privacy, this work unlocks greater potential for advanced machine learning techniques to be practically applied, enhancing clinical decision-making, patient care, and overall healthcare outcomes.

Graduation Semester

2025-05

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/129388

Copyright and License Information

Copyright 2025 Brandon Theodorou

Owning Collections