Integrating Early Detection of Heart Disease with Lung Cancer Powered by AI-CXR
Principal Investigator
Name
Morteza Naghavi
Degrees
M.D.
Institution
HeartLung Technologies
Position Title
President
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-1501
Initial CDAS Request Approval
Mar 31, 2026
Title
Integrating Early Detection of Heart Disease with Lung Cancer Powered by AI-CXR
Summary
Cardiovascular disease (CVD) and lung cancer (LC) are the two leading causes of mortality, yet current screening paradigms address them separately and incompletely. Low-dose CT (LDCT) screening, as implemented in the National Lung Screening Trial (NLST), has demonstrated mortality reduction for lung cancer while simultaneously capturing rich cardiopulmonary information relevant to cardiovascular risk. We previously received approval for integrated cardiopulmonary risk assessment in NLST based on CT scans in 2024. Our research studies with multiple publications have shown that AI-based analysis of CT scans enables comprehensive cardiovascular phenotyping, including cardiac chamber volumetry and cardiometabolic biomarkers that predict heart failure, atrial fibrillation, stroke, and mortality.
Chest radiography (CXR), which was also performed in NLST, is far more scalable and widely available but remains underutilized for quantitative risk assessment. Recent studies suggest that AI can extract prognostic cardiovascular and oncologic signals from CXR. However, it is unknown whether CXR can approximate the predictive performance of CT for integrated cardiopulmonary risk assessment.
Central Hypothesis:
AI-based analysis of chest radiographs can enable simultaneous prediction of cardiovascular and lung cancer outcomes and, when benchmarked against CT-based models in the same NLST population, will provide a scalable alternative for opportunistic multi-disease screening.
Aims
Aim 1. Develop AI models for integrated prediction of cardiovascular and lung cancer outcomes from chest radiographs in NLST.
We will train deep learning models using NLST CXR images to predict cardiovascular mortality and lung cancer incidence.
Hypothesis: CXR-based AI models can extract latent cardiopulmonary features predictive of both CVD and lung cancer outcomes.
Aim 2. Benchmark CXR-based AI models against CT-based AI models for cardiopulmonary risk prediction.
Using the same NLST participants with both CXR and LDCT imaging, we will directly compare performance of CXR-based models with CT-based models previously developed and validated by our group.
Hypothesis: CT-based models will provide superior predictive performance, but CXR-based models will retain substantial prognostic value and may offer a practical alternative for large-scale screening.
Aim 3. Evaluate the ability of CXR to approximate CT-derived cardiopulmonary phenotypes.
We will assess whether CXR-based AI models can infer structural and functional features previously quantified from CT, including cardiac chamber enlargement and cardiometabolic phenotypes.
Hypothesis: Key CT-derived imaging biomarkers have correlates detectable on CXR that contribute to risk prediction.
Aim 4. Assess the clinical utility of CXR-based opportunistic screening for combined CVD and lung cancer risk.
We will evaluate risk stratification performance and identify individuals at high risk who would not be captured by current screening criteria.
Hypothesis: AI-enabled CXR analysis can identify high-risk individuals missed by traditional approaches, supporting a scalable, low-cost strategy for integrated cardiopulmonary screening.
Collaborators
Morteza Naghavi HeartLung Corporation
Anthony Reeves HeartLung Technologies
Hamed Zarei HeartLung Technologies
Kyle Atlas HeartLung Technologies
Mohammad Mozafary HeartLung Technologies
Reza Mirjalili HeartLung Technologies