Cardiovascular risk from lung cancer screening images
Objective(s) of Data Request: We seek access to the full NLST dataset, including chest radiographs and CT scans, and corresponding X-rays—across all sites, with acquisition metadata. Our goal is to develop and validate deep learning-based pipelines that extract and integrate radiomics, calcium-omics, and fat-omics features. These models aim to enhance early diagnosis of cardiovascular disease and improve predictive accuracy for cardiac outcomes, while also allowing cross-domain analysis comparing cardiovascular and cancer outcomes.
Hypothesis: We hypothesize that an integrated AI framework combining lung cancer screening CT and CTCS-derived metrics, and clinical data can effectively detect and stratify risk for multiple cardiovascular conditions—such as MACE, heart failure, valve disease, and plaque progression—within a single model. This framework has the potential to enable earlier intervention and personalized surveillance. Furthermore, by jointly analyzing cardiovascular and cancer outcomes within the same cohort, we aim to uncover shared risk factors and imaging biomarkers that inform long-term multi-morbidity prediction.
Aim 1: Leverage available imaging data to develop automated deep learning segmentation models for key anatomical and pathological regions. These include coronary calcifications, epicardial and pericoronary adipose tissue, myocardial and chamber, and liver tissue—particularly from lung cancer screening datasets. Our segmentation pipelines will employ our most current, validated methods.
Aim 2: Engineer advanced imaging biomarkers across multiple domains, including: calcium-omics: (derived from coronary arteries, aortic valve, and mitral valve), fat-omics: Extracted from epicardial, paracardial, and pericoronary adipose tissues, shape-based features: Cardiac chamber geometry and morphometrics, radiomics texture features: Quantitative image biomarkers capturing tissue heterogeneity
We will investigate the association of these features with diverse cardiovascular and cancer-related outcomes.
Aim 3: Develop and validate predictive models including: multi-class classifiers for disease subtype stratification, multi-output regression models for concurrent outcome prediction, time-to-event models for survival and progression analysis. These models will be optimized to predict cardiovascular and oncologic outcomes, both independently and in combination.
Dr. Ammar Hoori, CWRU BME aoh11@case.edu
Dr. Juhwan, (CWRU BME, UH Cardiovascular Imaging Core Lab) jxl1982@case.edu
Dr. Sepideh Azarianpour-Esfahani, CWRU BME sxa786@case.edu
Ananya Subramaniam, CWRU BME axs1189@case.edu
Joshua Freeze, CWRU BME jcf100@case.edu
Tao Hu, CWRU BME txh272@case.edu
Dr. Hao Wu, CWRU BME hxw352@case.edu
Prerna Singh, CWRU BME, pxs524@case.edu