Date of Award

January 2023

Document Type

Open Access Thesis

Degree Name

Medical Doctor (MD)



First Advisor

Sanjay Aneja


Cancer patients are a unique population with increased mortality from cardiovascular disease, however only half of high-risk patients are medically optimized. Physicians ascertain cardiovascular risk from several risk predictors using demographic information, family history, and imaging data. The Agatston score, a measure of total calcium burden in coronary arteries on CT scans, is the current best predictor for major adverse cardiac events (MACE). Yet, the score is limited as it does not provide information on atherosclerotic plaque characteristics or distribution. In this study, we use deep learning techniques to develop an imaging-based biomarker that can robustly predict MACE in lung cancer patients. We selected participants with screen-detected lung cancer from the National Lung Screening Trial (NLST) and used cardiovascular mortality as our primary outcome. We applied automated segmentation algorithms to low-dose chest CT scans from NLST participants to segment cardiac substructures. Following segmentation, we extracted radiomic features from selected cardiac structures. We then used this dataset to train a regression model to predict cardiovascular death. We used a pre-trained nnU-Net model to successfully segment large cardiac structures on CT scans. These automated large cardiac structures had features that were predictive of MACE. We then successfully extract radiomic features from our areas of interest and use this high-dimensional dataset to train a regression model to predict MACE. We demonstrated that automated segmentation algorithms can result in low-cost non-invasive predictive biomarkers for MACE. We were able to demonstrate that radiomic feature extraction from segmented substructures can be used to develop a high-dimensional biomarker. We hope that such a scoring system can help physicians adequately determine cardiovascular risk and intervene, resulting in better patient outcomes.


This is an Open Access Thesis.

Open Access

This Article is Open Access