Abstract
Pancreatic cancer is a particularly lethal and aggressive malignancy characterized by its late detection and 5-year survival of less than 10% for both men and women as of 2022. While many cancers come with symptoms of varying degrees, pancreatic cancer is largely asymptomatic until the disease reaches its later stages. A variety of factors, both clinical and image-based, may be used to improve patient prognosis after diagnosis of this disease.Quantitative imaging analysis, namely radiomics, has been gaining popularity in medicine over recent years in identifying cancer variants that are particularly difficult to diagnose when symptoms are absent. In addition, radiomics is becoming a valuable tool for patient prognosis and to assess therapy response that may lead to treatment re-planning. This thesis will discuss radiomics-based survival modeling of patients with pancreatic cancer with the goal of early identification and supplementary intervention for patients based on this imaging-based technique.
This study retrospectively considers 75 abdominal MR images of borderline resectable pancreatic cancer patients treated at the University of Nebraska Medical Center from 2006 through 2017. Medical images were used to extract numerical features that may be used to supplement patient-specific treatment along with the standard clinical factors. Survival analysis is performed to evaluate the overall success rate of these predictive models and the impact each subset of features has on ultimate patient outcomes. It is shown that radiomics features may provide invaluable data regarding individual risk associated with pancreatic cancer that may lead to improved prognosis and individual treatment plans. It is also discussed that this methodology is not specific to pancreatic cancer and may improve prognosis across a breadth of diseases.