Machine Learning Approaches that Extend Healthcare: Algorithms & Applications

Modern clinical systems frequently exhibit sporadic patient visits, delayed diagnoses, and unequal care distribution among diverse populations. Often, diseases aren’t identified until they reach advanced stages. The scarcity of specialists and disparities in healthcare access further complicate the long-term monitoring, timely intervention, and unbiased assessments. This thesis addresses the above challenges by developing artificial intelligence (AI) and machine learning (ML) algorithms and building practical systems that use these algorithms to solve key problems in healthcare and medicine. Specifically, on the algorithms front, the thesis introduces principled ML approaches to achieve fair, unbiased, and generalizable AI models, addressing core challenges in real-world medical data which encompass four main axes: • Label Scarcity: The thesis presents a novel self-supervised learning scheme that learns periodic and frequency information in data without labels, enabling representation learning for periodic tasks like vital signs estimation with minimal labeling efforts. • Data Imbalance: The thesis develops new ML algorithms to address data imbalance in regression, filling the gap in techniques for practical imbalanced regression problems. • Domain Generalization: The thesis presents theoretically grounded learning methods that ensure generalization across imbalanced domains and unseen environments. • Subpopulation Shifts: The thesis studies learning in the presence of underrepresented subgroups, providing actionable insights for model deployment in real-world settings. On the applications front, the thesis develops new AI-driven biomarkers and systems for human disease and medicine leveraging the proposed algorithms, enabling discovery and advancing delivery and equity in healthcare: • Early Diagnosis Biomarker for Parkinson’s: The thesis presents an AI-based biomarker for Parkinson’s disease that enables early detection years before standard clinical diagnosis, as well as longitudinal progression tracking using nocturnal breathing signals. • In-Home Touchless Monitoring of Sleep Posture: The thesis designs novel AI systems for continuous and contactless sleep posture monitoring overnight in the user’s own home using wireless signals. • Equitable Medical AI Deployments In The Wild: The thesis establishes best practices for medical imaging AI models that maintain their performance and fairness in deployments beyond their initial training contexts, across diverse populations and unseen sites.

Name thesis:degree_name

Doctoral

Department dc:contributor.department

Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science

Grantor dc:publisher

Massachusetts Institute of Technology

Year dc:date.issued

2024

Author dc:creator

• Yang, Yuzhe

Advisor dc:contributor.advisor

• Katabi, Dina

Statement dc:rights

• In Copyright - Educational Use Permitted
• Copyright retained by author(s)

Licence dc:rights.uri

• https://rightsstatements.org/page/InC-EDU/1.0/

Handle dc:identifier.uri

https://hdl.handle.net/1721.1/156620

OAI identifier oai:identifier

oai:dspace.mit.edu:1721.1/156620