DOI
https://doi.org/10.25772/SWE6-BY31
Defense Date
2020
Document Type
Dissertation
Degree Name
Doctor of Philosophy
Department
Computer Science
First Advisor
Preetam Ghosh
Second Advisor
Jatinder Palta
Third Advisor
Bridget McInnes
Fourth Advisor
Lulin Yuan
Fifth Advisor
Thang Dinh
Abstract
Radiation oncology is the field of medicine that deals with treating cancer patients through ionizing radiation. The clinical modality or technique used to treat the cancer patients in the radiation oncology domain is referred to as radiation therapy. Radiation therapy aims to deliver precisely measured dose irradiation to a defined tumor volume (target) with as minimal damage as possible to surrounding healthy tissue (organs-at-risk), resulting in eradication of the tumor, high quality of life, and prolongation of survival. A typical radiotherapy process requires the use of different clinical systems at various stages of the workflow. The data generated in these different stages of workflow is stored in an unstructured and non-standard format, which hinders interoperability and interconnectivity of data, thereby making it difficult to translate all of these datasets into knowledge that supports decision-making in routine clinical practice. In this dissertation, we present an enterprise-level informatics platform that can automatically extract and efficiently store clinical, treatment, imaging, and genomics data from radiation oncology patients. Additionally, we propose data science methods for data standardization, safety, and treatment quality analysis in radiation oncology. We demonstrate that our data standardization methods using word embeddings and machine learning are robust and highly generalizable on real-word clinical datasets collected from the nationwide radiation therapy centers administered by the US Veterans' Health Administration. We also present different heterogeneous data integration approaches to enhance the data standardization process. For patient safety, we analyze the radiation oncology incident reports and propose an integrated natural language processing and machine learning based pipeline to automate the incident triage and prioritization process. We demonstrate that a deep learning based transfer learning approach helps in the automated incident triage process. Finally, we address the issue of treatment quality in terms of automated treatment planning in clinical decision support systems. We show that supervised machine learning methods can efficiently generate clinical hypotheses from radiation oncology treatment plans and demonstrate our framework's data analytics capability.
Rights
© The Author
Is Part Of
VCU University Archives
Is Part Of
VCU Theses and Dissertations
Date of Submission
8-6-2020