Show simple item record

dc.contributor.advisorHuang, Shuai
dc.contributor.authorSamareh Abolhasani, Banafsheh
dc.date.accessioned2019-10-15T22:58:52Z
dc.date.available2019-10-15T22:58:52Z
dc.date.submitted2019
dc.identifier.otherSamarehAbolhasani_washington_0250E_20602.pdf
dc.identifier.urihttp://hdl.handle.net/1773/44834
dc.descriptionThesis (Ph.D.)--University of Washington, 2019
dc.description.abstractTechnological innovations have given rise to data-rich environments that support the use of heterogeneous sensor measurements to monitor complex healthcare systems. Despite these advancements, however, there remains little understanding of how patient health evolves in real clinical settings and how changes in health condition generate manifestations that are captured by the data. To address this knowledge gap, my research aims to build disease trajectory modeling that can reconstruct the evolving patient’s health condition over time, termed the contemporaneous health index (CHI), by combining data and the natural history model of the disease. This global index may help increase the continuity of care, facilitate patient-provider communication, and assist with a range of clinical decision makings. However, lack of deep understanding of the disease and its progression, existence of patient heterogeneity, inherent uncertainty in predictive models and the emergence of large amounts of complex, and unstructured data, all could compromise prediction capabilities. In this dissertation, we developed innovative methodologies that reflect the progression of the underlying patient condition by learning personalized models, quantifying the uncertainty of those models, and creating effective biomarker engineering pipelines to analyze large amounts of complex data for effective incorporation into the calculation of CHI. We first, proposed a novel clinical data fusion framework, named DL-CHI: a dictionary learning-based CHI that quantifies the severity of the deterioration process over time and represents monotonic progression patterns with a systematic optimization formulation. DL-CHI mitigated the heterogeneity of the patients by incorporating dictionary learning to create personalized models for individual patients. We then developed the UQ-CHI framework: an uncertainty quantification-based model of CHI to further enhance the disease trajectory modeling with uncertainty quantification by considering imperfect and continuous delivery of knowledge via probabilistic nature of maximum entropy discrimination (MED) principle. Finally, we proposed effective biomarker engineering pipelines to enable possible extensions of the CHI for building trajectory models from complex data (video, audio, text, and mobile sensor reading data). We applied the proposed methodologies to real-world applications, including Alzheimer’s disease (AD), surgical site infection (SSI), depression and human activity recognition using wearable sensors
dc.format.mimetypeapplication/pdf
dc.language.isoen_US
dc.rightsCC BY
dc.subjectBayesian
dc.subjectConvex optimization
dc.subjectMachine learning
dc.subjectMulti-modality frameworks
dc.subjectUncertainty quantification
dc.subjectIndustrial engineering
dc.subject.otherIndustrial engineering
dc.titleContemporaneous Health Monitoring and Biomarker Discovery by Integration of Patient Data and Disease Knowledge
dc.typeThesis
dc.embargo.termsOpen Access


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record