Computational imaging and multiomic biomarkers for precision medicine: characterizing heterogeneity in lung cancer.
Degree type
Graduate group
Discipline
Subject
non-small cell lung cancer
precision medicine
radiogenomics
radiomics
tumor heterogeneity
Funder
Grant number
License
Copyright date
Distributor
Related resources
Author
Contributor
Abstract
Lung cancer is the leading cause of cancer deaths and is the third most diagnosed cancer in both men and women in the United States. Non-small cell lung cancer (NSCLC) accounts for 84% of all lung cancer cases. The inherent intra-tumor and inter-tumor heterogeneity in lung tumors has been linked with adverse clinical outcomes. A well-rounded characterization of tumor heterogeneity by personalized biomarkers is needed to develop precision medicine treatment strategies for cancer. Large-scale genome-based characterization poses the disadvantages of high cost and technical complexity. Further, a histopathological sample from a tumor biopsy may not be able to fully represent the structural and functional properties of the entire tumor. Medical imaging is now emerging as a key player in the field of personalized medicine, due to its ability to non-invasively characterize the anatomical and physiological properties of the tumor regions. The studies included in this thesis introduce analytical tools developed thorough machine learning and bioinformatics and use information from diagnostic images and other “omic” sources, to develop computational imaging and multiomic biomarkers to characterize intratumor heterogeneity. A novel radiomic biomarker, that integrates with PDL1 expression, ECOG status, BMI, and smoking status, to enhance the ability to predict progression-free survival in a preliminary cohort of patients with stage 4 NSCLC, treated with first-line anti-PD1/PDL1 checkpoint inhibitor therapy PEMBROLIZUMAB. This study also showed that mitigation of the heterogeneity introduced by voxel spacing and image acquisition parameters improves the prognostic performance of the radiomic phenotypes. We further performed a detailed investigation of the effects of heterogeneity in image parameters on the reproducibility of prognostic performance of models built using radiomic biomarkers. The results of this second study indicated that accounting for heterogeneity in image parameters is important to obtain more reproducible prognostic scores, irrespective of image site or modality. In the third study, we developed novel multiomic phenotypes in a larger cohort of patients with stage 4 NSCLC treated with PEMBROLIZUMAB. These multiomic phenotypes, formed by integration of radiomics, radiological and pathological information of the patients, enhanced precision in progression-free survival prediction upon combination with prognostic clinical variables. To our knowledge, our study is the first to construct a “multiomic signature for prognosis of NSCLC patient response to immunotherapy, in contrast to prior radiogenomic approaches leveraging a radiomics signature to identify patient categories based on a genomic biomarker-based classification. In the exploratory fourth study, we evaluated the performance of radiomics analyses of part-solid lung nodules to detect nodule invasiveness using several approaches: radiomics analysis in the presurgical CT scan, delta radiomics over three time-points leading up to surgical resection and nodule volumetry. The best performing model for the prediction of nodule invasiveness was the model built using a combination of immediate pre-surgical, delta radiomics, delta volumes and clinical assessment. The study showed that the combined utilization of clinical, volumetric and radiomic features may facilitate complex decision making in the management of subsolid lung nodules. To summarize, the studies included in this thesis demonstrate the value of computational radiomic and multiomic biomarkers in the characterization of lung tumor heterogeneity and have the potential to be utilized in the advancement of precision medicine in oncology.