From Data to Knowledge: Temporal Rule Learning from Electronic Health Records for Modelling Patient Histories

Schulz, Nicolas (2023) From Data to Knowledge: Temporal Rule Learning from Electronic Health Records for Modelling Patient Histories. Master's Thesis / Essay, Artificial Intelligence.

Preview	Text Final_Master_Thesis_N.A.Schulz.pdf Download (4MB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (183kB)

Abstract

Large patient data is gathered continuously at every hospital. If the digital representation thereof is in the form of a standardized Electronical Health Record (EHR), reusable and scalable analytics pipelines can be developed to uncover data-driven insights in patient populations. In a sensitive context such as medical decision-support, ideally, these insights are generated with interpretable models and result in a visual representation which can be assessed by medical professionals without technical expertise. In this thesis, the foundation was laid to robustly extract data from standardized EHR, transform it into a time series, learn temporal rules from these hospital encounter histories and visualize the results in a hierarchical and directed graph. Three models of computational intelligence, namely (1) a baseline transition matrix, (2) a Temporal Association Rule Mining and (3) a Dynamic Bayesian Network structure learning approach, are implemented in separate pipelines and the results are compared across models, datasets and hyperparameters using real-world lung cancer patient data from Germany. It was shown that all three approaches possess individual strengths and use cases. However, learning the structure of a Dynamic Bayesian Network from data was found to be the most robust approach and resulted in the most meaningful temporal rules.

Item Type:	Thesis (Master's Thesis / Essay)
Supervisor name:	Schippers, M.B. and Bunte, K.
Degree programme:	Artificial Intelligence
Thesis type:	Master's Thesis / Essay
Language:	English
Date Deposited:	09 Nov 2023 10:30
Last Modified:	09 Nov 2023 10:30
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/31619

Actions (login required)

View Item