Date of Graduation

5-2026

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Engineering (PhD)

Degree Level

Graduate

Department

Industrial Engineering

Advisor/Mentor

Zhang, Shengfan

Committee Member

Milburn, Ashlea

Second Committee Member

Cantanzaro, Donald

Third Committee Member

Liao, Haitao

Keywords

Healthcare Systems; Personalized Care; Data Digitization; Decision Analytics

Abstract

This dissertation traces a progression from foundational data infrastructure to individualized clinical decision-making, developing and evaluating three computational frameworks that advance healthcare efficiency and personalization through deep learning and decision analytics. The first study presents an end-to-end pipeline for automated recognition of handwritten medical forms, addressing persistent challenges in health data digitization. Integrating a YOLO-based field detection model, a Convolutional Recurrent Neural Network (CRNN) for text transcription, and a confidence-based human-in-the-loop quality assurance framework, the system achieved 99.75\% Exact Match Accuracy and a 0.13\% Character Error Rate on medical forms collected from a tuberculosis research project in Moldova. A GPU-accelerated batch processing architecture reduced the average processing time to 1.84 seconds per document, establishing a scalable solution for clinical research and public health data management. The second study develops a preference-adaptive deep reinforcement learning framework for individualized preventive decision-making in women at elevated hereditary breast cancer risk, encompassing confirmed pathogenic variant carriers (BRCA1, BRCA2, PALB2) and individuals with elevated familial risk without a confirmed variant. By encoding patient quality-of-life priorities as dynamic state variables rather than fixed population-level parameters, the Proximal Policy Optimization (PPO) agent achieved quality-adjusted life expectancy gains of 1.31–1.45 QALYs over prophylactic surgery-first strategies and 2.33–4.33 QALYs over guideline-concordant care across mutation-carrier cohorts. Pareto efficiency and SHAP-based interpretability analyses further confirmed the clinical coherence of learned policies. The third study develops a discrete-time Markov Decision Process framework for subtype-specific sequential treatment optimization in metastatic breast cancer, encompassing hormone receptor-positive/HER2-negative, HER2-positive, and triple-negative subtypes. Parameterized from clinical trial and real-world evidence, the PPO agent identified clinically interpretable tipping-point thresholds for transitions to best supportive care and demonstrated appropriate age-stratified policy differentiation. A key finding — that survival-dominated reward signals attenuate the influence of patient preference weights under the current architecture — motivates a constrained MDP reformulation as a priority for future work. Collectively, these studies demonstrate that the path from raw data digitization to personalized clinical care can be systematically bridged through deep learning and decision analytics, offering scalable frameworks with implications for health data infrastructure, cancer prevention, and oncological treatment sequencing.

Download

Share

COinS