Date of Graduation
12-2016
Document Type
Dissertation (PhD)
Program Affiliation
Medical Physics
Degree Name
Doctor of Philosophy (PhD)
Advisor/Committee Chair
Laurence Court
Committee Member
Jinzhong Yang
Committee Member
Mary Martel
Committee Member
Daniel Gomez
Committee Member
Francesco Stingo
Committee Member
Tina Briere
Abstract
Radiation injury in the esophagus occurs with high frequency from the treatment of non-small cell lung cancer (NSCLC). Radiation esophagitis is an acute normal tissue toxicity that negatively affects treatment efficacy by limiting dose and potentially interrupting radiation therapy. Clinical quantification of this toxicity is typically achieved by utilizing physician grading scales, assigning complication severity on an ordinal scale of symptom presentation and/or physician chosen interventions. These criteria are subjective in nature, both from the physician assigning the grade and the patient reporting the symptom. Furthermore, radiation therapy planning guidelines for the esophagus are derived from toxicity prediction models utilizing these subjective grading scores as complication endpoints. Not only does this schema of toxicity analysis leads to a lack of consistency between models from different study populations, and thereby radiation therapy planning recommendations for the esophagus, but inherent patient radiosensitivity is not considered, leading to suboptimal treatment regimens.
The purpose of this work was to investigate radiation injury in the esophagus by first developing in-vivo imaging biomarkers of radiation-response in the esophagus using 4-dimensional computed tomography (4DCT) and 18fluorodeoxyglucose positron emission tomography (FDG-PET), separately. These imaging biomarkers were then compare with radiation esophagitis grade, using traditional and machine learning techniques, and shown to objectively quantify esophageal radiation toxicity. Metrics describing the esophageal radiation response from either imaging modality were strong classifiers of radiation esophagitis grade. Multivariate models to predict maximum esophagitis treatment grade (4DCT), and esophagitis symptom progression (FDG-PET) were developed and had strong performance for both scenarios.
These imaging biomarkers were then used to comprehensively investigate the influence of dose-geometry and radiation type (photon or proton) on esophageal response. Using these radiation-response biomarkers in esophageal dose-response analysis, dose metrics with spatial information of esophageal dose coverage, (e.g. dose to a subregion of the esophagus with specific percent cross-sectional area coverage), as well as without spatial information, (traditional dose-volume histogram), was analyzed separately using machine learning methods. No detectable difference in response was observed when comparing dose metrics with and without spatial information. Statistical analysis showed no significant difference (p
Inherent patient radiation sensitivity was investigated using esophageal expansion and delivered dose to the corresponding esophageal subregion. Cluster analysis was used to group patient patients based on their maximum expansion and delivered dose to the analyzed subregion of the esophagus. Patients clustered with proportionally higher expansion per delivered dose were considered radiosensitive. These results were then applied to NTCP toxicity modelling by using patient radiosensitivity cluster membership as a predictor variable. Models with the radiosensitive predictor outperformed models not including the cluster membership variable for prediction of grade 3 esophagitis.
Keywords
Radiation therapy, normal tissue toxicity, esophagitis, non small cell lung cancer, NTCP modelling, imaging biomarker, machine learning, functional imaging, 4DCT, FDG-PET