Back to search

University of Washington

Improving MRI Operations through Data-Driven Methods and Predictive Modeling

Abstract

dc:description.abstract

Magnetic resonance imaging (MRI) is one of the most clinically valuable and operationally complex modalities in modern radiology departments. Its operations are characterized by significant exam duration variability, complex scheduling demands, and a heavy reliance on technologist expertise that is difficult to develop and assess systematically. Despite theavailability of rich operational data from sources like electronic health records (EHRs) and modality log files (MLFs), MRI operations management has remained largely experience-based and reactive. This dissertation addresses this disconnect through five interconnected studies that apply data-driven methods to operational improvement in MRI services. The first study established the feasibility of a virtual hub-and-spoke model for deploying MRI expertise across geographically distributed sites, demonstrating that key radiology functions can be performed remotely while improving patient turnaround time and wait time. The second analyzed modality log file data to identify actionable sources of workflow inefficiency that are invisible to EHR-based analysis alone. The third characterized the structure and distributional properties of EHR data underlying subsequent predictive work and demonstrated that clinical text embeddings derived from result narratives using BioClinicalBERT carry predictive signal for exam duration. The fourth introduced a concordance-based validation framework for integrating EHR and MLF data, raising the concordance correlation coefficient between systems from 0.33 to 0.87, and showed that a Random Forest model trained on the validated dataset outperformed template-based scheduling for 11 of 12 procedure codes with mean absolute error reductions ranging from 2% to 57%. The fifth compared Gated Recurrent Unit (GRU) neural networks against XGBoost on three established benchmark datasets, finding that GRU models consistentlyoutperformed XGBoost with average advantages ranging from 8.5% to 18.8%, particularly at longer temporal windows and with minimal feature engineering. Collectively, these studies demonstrate that meaningful gains in MRI operational efficiency and patient access are achievable by employing data-driven methods, and that the choice of analytical approach can matter as much as the availability of data. The findings provide both practical tools for MRI operational improvement and a methodological foundation for future temporal modeling of dynamic technologist competency.

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Li, Lun
Advisor dc:contributor.advisor
  • Mastrangelo, Christina

Subjects

dc:subject × 3

Rights

dc:rights
Statement dc:rights
  • CC BY-NC-SA
Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/1773/57421

Chain of custody

source
Harvested from
University of Washington
Base URL
digital.lib.washington.edu/server/oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Li, Lun. Improving MRI Operations through Data-Driven Methods and Predictive Modeling. 2026. https://hdl.handle.net/1773/57421