Rachel Dzieran, a doctoral candidate in systems engineering, will defend their dissertation titled “A Domain-Independent Framework for Human-Centered AI in High-Stakes Medical Decisions: Integrating Transplant Surgeon Expertise Through Fuzzy Logic in Kidney Allocation” Their advisor, Dr. Cihan Dagli is an professor in engineering management and systems engineering. The dissertation abstract is provided below.

This research introduces the Transplant Surgeon Fuzzy Associative Memory (TSFAM) framework, a domain-independent, human-centered Artificial Intelligence (AI) system that captures individual transplant surgeons' expertise through fuzzy logic to support decision-making in deceased-donor kidney allocation. TSFAM formalizes tacit clinical knowledge through a three-phase structured elicitation process: Define Ontology, Membership Function Calibration, and Decision-Making Rule Development. The framework develops an adaptable System-of-Systems architecture that integrates surgeon-specific ontologies and decision rules with the Final Acceptance deep learning model, generating context-sensitive recommendations while preserving clinical autonomy. Current AI systems in organ allocation treat human expertise as either irrelevant through full automation or merely advisory through standardized algorithms. These tools overlook surgeon expertise, institutional capacities, and local patient populations. TSFAM addresses this gap by creating a collaborative decision space where the human-AI balance emerges dynamically from surgeon-defined rule sets rather than from fixed algorithmic ratios. This research makes three primary contributions. First, a novel framework that integrates surgeon-specific ontologies and membership functions with deep learning models, demonstrating how fuzzy logic captures nuanced clinical judgment. Second, a systematic methodology for eliciting and formalizing tacit expertise through iterative fuzzy rule refinement, providing a reproducible template for human-AI teaming that adapts to individual practitioner preferences and local healthcare contexts. Third, empirical verification that the TSFAM model manages ambiguity in marginal organ evaluations, and confirmation that human-AI teams can integrate computational and human judgment when the AI system adapts to surgeon-specific decision rules. Fuzzy Associative Memory for Expertise (FAME) was operationalized in MATLAB Fuzzy Logic Toolbox (R2025b) and instantiated as TSFAM for deceased-donor kidney transplant acceptance decision-making support. The three-phase elicitation protocol was executed with two participating transplant surgeons, producing initial verified rule bases for each surgeon through the four-step rule development process and the verification protocol. The methodology and verification protocol are domain-independent and transferable to other high-stakes medical decision-making domains in which expert judgment augments a population-level AI model. Beyond kidney allocation, this work establishes foundational principles for developing adaptive AI systems and presents a unique framework for nuanced clinical judgment in complex healthcare decisions. The approach preserves clinician expertise while leveraging computational power. Its three-phase elicitation protocol, hybrid expert-data construction methodology, and initial verification methodology offer a blueprint for responsible AI adoption across medical specialties, where uncertainty, context, and human values remain paramount to decision quality.