Theses and Dissertations

Advisor

Mittal, Sudip

Committee Member

Amirlatifi, Amin

Committee Member

Bozorgzad, Sean

Committee Member

Rahimi, Shahram

Date of Degree

8-7-2025

Original embargo terms

Immediate Worldwide Access

Document Type

Dissertation - Open Access

Major

Computer Science

Degree Name

Doctor of Philosophy (Ph.D.)

College

James Worth Bagley College of Engineering

Department

Department of Computer Science and Engineering

Abstract

Healthcare communication is plagued by fragmented medical knowledge, patient misunderstanding of care plans, and clinician burnout from documentation burdens. These inefficiencies cost the U.S. healthcare system over $300 billion annually due to preventable nonadherence and administrative waste [47, 19]. While Artificial Intelligence (AI) technology like Large Language Models (LLMs) offer potential solutions, existing systems fail to deliver personalized, context-aware guidance or automate documentation without sacrificing accuracy. This dissertation addresses these gaps through three novel context-aware AI frameworks such as MedInsight, ClinicSum, and ClinicDuo. MedInsight leverages a multi-source context augmentation approach to synthesize patient centric medical responses by integrating Electronic Health Records (EHRs), clinical guidelines, and medical literature. Evaluated on the MTSamples dataset, MedInsight achieves a answer similarity score of 0.93 and moderate agreement (60.07) from clinical experts, demonstrating its ability to ground answers in individualized patient contexts. On the other hand, ClinicSum automates Subjective-Objective-Assessment-Plan (SOAP) clinical summary generation from patient-doctor conversations using a hybrid architecture that combines retrieval-based filtering with fine-tuned language models. Trained on 1,473 expert-curated dialogue-summary pairs, CLINICSUM outperforms state-of-the-art baselines (F1 = 0.84) and is preferred by 61% of clinicians for its structured accuracy. Similarly, ClinicDuo bridges clinician and patient needs by optimizing EHR-based discharge summaries using In-Context Learning (ICL) and PEFT fine-tuning. Evaluations on MIMIC-IV show that PEFT outperforms ICL in relevance, factuality and readability. These findings underscore the value of domain-specific fine-tuning for reliable clinical documentation, particularly in resource-limited settings. Together, these frameworks establish a paradigm for context-aware AI that augments healthcare communication across three pillars: resolving fragmented knowledge via multi-source context augmentation, reducing documentation burdens through structured summarization, and balancing technical accuracy and readability in EHR summarization task. By grounding AI outputs in real world medical contexts, this work advances equitable care delivery while mitigating costs tied to miscommunication.

Share

COinS