Rao, Arvind2019-05-172019-11-012019-052019-04-17May 2019Barua, Souptik. "Leveraging structure in cancer imaging data using data-driven frameworks to predict clinical outcomes." (2019) Diss., Rice University. <a href="https://hdl.handle.net/1911/105939">https://hdl.handle.net/1911/105939</a>.https://hdl.handle.net/1911/105939Immunotherapy and radiation therapy are two of the most prominent strategies used to treat cancer. While both these strategies have succeeded in treating the disease in many patients and cancer types, they are known to not work well for all patients, sometimes even leading to adverse side effects. There is thus a critical need to be able to predict how patients might respond to these therapies and accordingly design optimal treatment plans. In this thesis, I have built data-driven frameworks that leverage different types of structure in cancer imaging data to predict clinical outcomes of interest. I demonstrate my findings using two kinds of cancer image data: multipexed Immuno-Fluorescence (mIF) images from the field of pathology, and Computed Tomography (CT) from radiology. In mIF images, I quantify spatial structure in terms of proximities of cancer cells and various immune cells by developing ideas from spatial statistics to estimate the nearest neighbor distribution function of a spatial point process. I demonstrate that this quantity, called the G-function, can be used as a visual signature of the infiltration patterns of different cell types of interest. I compute summary metrics from the G-function which are prognostic of clinical outcome in multiple cancer types such as pancreatic, breast, and skin cancer. I design a functional analysis pipeline to more efficiently summarize the G-function and predict the risk of progression in pancreatic cysts. Finally, I propose approaches to capture the heterogeneity in the tumor microenvironment and demonstrate the importance of quantifying spatial structure separately in different tumor regions through a case study in lung cancer. In CT images acquired at different time points, I capture the temporal evolution of a specific class of image features called radiomic features using ideas from functional analysis. I then show that the temporal dynamics of radiomic features can be used to predict clinical outcomes such as the likelihood of complete response to radiation therapy and the risk of developing long-term radiation injuries such as osteoradionecrosis. Overall, I envisage these data-driven frameworks can be incorporated into a clinical setting not only as a tool for prognosis and treatment design but also in basic science research to understand the biological underpinnings of cancer development.application/pdfengCopyright is held by the author, unless otherwise indicated. Permission to reuse, publish, or reproduce the work beyond the bounds of fair use or other exemptions to copyright law must be obtained from the copyright holder.cancer imagingmachine learningspatial statisticsfunctional data analysisimmunotherapyradiation therapyThesis2019-05-17