Date of Award

2024

Document Type

Open Access Master's Thesis

Degree Name

Master of Science in Health Informatics (MS)

Administrative Home Department

Department of Applied Computing

Advisor 1

Guy Hembroff

Committee Member 1

Ronghua Xu

Committee Member 2

Shane Meuller

Abstract

With advancements in technology and the growth of artificial intelligence, the gap between healthcare challenges and solutions is narrowing. This study explores the use of large language models (LLMs) to screen for Alzheimer's disease through patient-physician conversations. Fine-tuning and instruction-tuning techniques were applied to improve predictions. The performance of fine-tuned open-source models like LLaMA and GPT-3 was compared to smaller models such as TinyLLaMA and retrieval-augmented generation (RAG) models. The models were tested to determine their ability to classify individuals as cognitively normal (CN) or at any stage of Alzheimer's disease (AD). Reinforcement learning through human feedback was used to evaluate the models, while the DementiaBank dataset validated the evaluation metrics. This research demonstrates the potential of LLMs in advancing mental health diagnostics.

Share

COinS