I-Corps: Translation Potential of Artificial Intelligence (AI)-Enabled Cardiovascular Risk Prediction from Routine Imaging and Clinical Data

This I-Corps project is based on the development of two complementary artificial intelligence (AI) technologies that estimate a person's risk of heart attack, stroke, heart failure, or cardiovascular death using information collected during routine care. Current risk assessment tools depend on a limited set of values that may be missing or outdated and reduce data-rich scans to a single measurement. These two technologies address those limitations by evaluating risk directly from routine images and combining available clinical and electronic health record data. The technologies may help clinicians identify high-risk patients earlier, support hospitals and population-health programs in directing preventive care, and allow imaging or health-information vendors to add risk assessment to existing systems. The applications may include health-system software, payer analytics, and licensing or integration partnerships with imaging, electronic health record, and clinical artificial intelligence companies. This may support more individualized cardiovascular risk assessment and may remain useful even when some clinical data are missing. This I-Corps project utilizes experiential learning coupled with first-hand investigation of the industry ecosystem to assess the translation potential of an image-only computed tomography model, and a structured clinical and electronic health record model. Cardiovascular disease remains the leading cause of death globally, yet many of the clinical tools used to estimate cardiovascular risk were developed decades ago and rely on simplified statistical models. The first technology applies three-dimensional deep learning and clinically grounded nested representation learning directly to routine non-contrast computed tomography volumes, preserving prognostic patterns that conventional approaches discard when scans are reduced to coronary artery calcium scores, predefined biomarkers, or handcrafted radiomics. The second technology uses gradient-boosted classification and survival modeling to combine available demographics, laboratory values, medications, diagnoses, risk factors, clinical scores, calcium measurements, and derived imaging features from 1 to 10 years while accommodating real-world missingness. Together these technologies may provide more complete risk assessment without requiring additional clinical testing. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.