Date of Award

12-30-2025

Document Type

Open Access Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computational Sciences

First Advisor

Xiaohui Liang

Abstract

Early detection of cognitive decline and efficient medical image analysis remain critical challenges in healthcare. Traditional clinical assessments are infrequent and resource-intensive, while everyday speech data and unlabeled medical images remain largely unexploited. This dissertation develops computational methods integrating machine learning and artificial intelligence across speech, text, and imaging modalities to address challenges in medical data processing. For cognitive monitoring, this work first introduces methods using voice assistant systems to collect longitudinal speech data in home environments, demonstrating that incorporating historical session patterns significantly enhances detection of mild cognitive impairment. Building on this foundation, a framework combining large language model-driven prompt refinement with multimodal fusion extends the approach to unstructured daily voice commands, where automatically extracted linguistic features align with established cognitive markers such as word-finding difficulties and reduced coherence. To further improve generalizability across diverse populations, supervised contrastive learning combined with Product of Experts fusion addresses picture variability and achieves robust performance across languages and demographic subgroups. Beyond speech analysis, this dissertation also addresses medical imaging challenges through an open-source application using self-supervised learning with modified lesion detection filtering, enabling accurate content-based image retrieval without extensive manual annotations. Together, these contributions demonstrate that passive speech monitoring can detect cognitive changes with clinical relevance and that self-supervised approaches effectively address label scarcity in medical imaging, advancing computational methods for handling diverse biomedical data in clinical settings.

Comments

Free and open access to this Campus Access Thesis is made available to the UMass Boston community by ScholarWorks at UMass Boston. Those not on campus and those without a UMass Boston campus username and password may gain access to this thesis through Interlibrary Loan. If you have a UMass Boston campus username and password and would like to download this work from off-campus, click on the “Off-Campus Users” button.

Download

Share

COinS