Empowering vision machine perception for robust telehealth applications

Hoang, Trung Hieu

Permalink

https://hdl.handle.net/2142/132490 Copy

Description

Title

Empowering vision machine perception for robust telehealth applications

Author(s)

Hoang, Trung Hieu

Issue Date

2025-11-25

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

Do, Minh N

Doctoral Committee Chair(s)

Do, Minh N

Committee Member(s)

• Cunningham, Brian T
• Hsiao-Wecksler, Elizabeth T
• Shomorony, Ilan
• Wang, Yuxiong

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

AI for healthcare, computer vision, machine learning, human-motion analysis, biosensors, test-time adaptation

Abstract

Today, the healthcare system faces significant challenges driven by a shortage of health workers, the growing demand for personalized care, and an aging population. This creates a crisis of bottlenecks in symptom triage and longitudinal monitoring. Digital telehealth toolbox and AI-assisted diagnosis systems offer promising solutions as force multipliers to alleviate these challenges. Among them, smartphone vision-based telehealth exams have gained significant interest due to their powerful computing power, serving as valuable point-of-care sensors. Yet, developing vision-based modules to extract digital biomarkers from visual data and reliably deploying them in real-world scenarios remains a key challenge. This dissertation concentrates on two primary research paths: empowering computer vision for telehealth and deploying robust machine learning (ML) models for telehealth applications. Under the first focus, the Digitized Neurological Examination (DNE) system is introduced for comprehensive vision-based neurological examination using smartphones, validated for clinical relevance and abnormality detection and documentation. Additionally, the smartphone-based viral pathogen detection system, PathTracker, is presented for rapid point-of-care diagnosis through innovative image processing techniques. In the second focus, although test-time adaptation (TTA) techniques offer promise in handling domain-shift challenges during ML model deployment, they are susceptible to error accumulation and even adversarial attack. We extensively investigate this issue, resulting in the introduction of “persistent TTA” and “reusing of incorrect prediction attack (RIP)” to ensure stability in dynamic testing environments. These contributions drive forward robust telehealth solutions for neurological care and viral pathogen detection, providing effective responses to future healthcare challenges.

Graduation Semester

2025-12

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/132490

Copyright and License Information

Copyright 2025 Trung Hieu Hoang

Owning Collections