Cardiovascular risk prediction from small, low-densitycalcifications detected on CT calciumscore exams

The National Heart, Lung, and Blood Institute awarded Case Western Reserve University $109,076 on September 30, 2025, under the Cardiovascular Diseases Research program (CFDA 93.837) to develop improved cardiovascular risk prediction models using advanced computational methods applied to cardiac CT imaging. The research applies deep learning to detect small, low-density calcifications on coronary artery calcium score exams that are predictive of early-stage atherosclerosis and near-term cardiovascular events. The work comprises detection optimization of sub-Agatston calcifications, derivation of image features predictive of cardiovascular risk, causal association of calcification features with socioeconomic and clinical risk factors including pericoronary and epicardial adipose tissue, and validation of sub-Agatston calcification predictive value for major adverse cardiac events in over 100,000 patients through Cox proportional hazards modeling. Research is performed at Case Western Reserve University, Office of Research Administration in affiliation with University Hospitals of Cleveland in Cleveland, Ohio, with performance through September 29, 2029.

Abstract

Atherosclerotic cardiovascular disease continues to be a significant health concern worldwide. Improving risk stratification is particularly crucial to identify individuals at risk for near term cardiovascular incidents and detect disease earlier. Using cardiac CT scans for coronary artery calcium (CAC) scoring is a well-established risk assessment method, but the traditional agatston calcium scoring method lacks sensitivity for near-term and early-stage disease detection. On cardiac CT images, small, low-density, "spotty" calcifications have been linked to high-risk plaque characteristics, like inflammation and stress, and signal an early stage of atherosclerosis before progressing into larger calcifications. Yet, these small calcifications are usually overlooked in standard agatston scoring due to their low density and/or small size. The objective of the proposed research is to apply advanced computational methods like deep learning to optimize the detection of these "sub-agatston" calcifications (sub-aim 1A) and derive reliable image features that characterize these sub-agatston cacs and are predictive of cardiovascular risk (sub-aim 1B). To gain insights into the mechanisms of atherosclerosis initiation and progression, sub-agatston calcification features will be causally associated with potential drivers of atherosclerosis, including socioeconomic, clinical, and image-derived-metabolic risk factors (pericoronary and epicardial adipose tissue) (sub-aim 2A). Lastly, the predictive value of sub-agatston calcifications for major adverse cardiac events will be assessed in over 100,000 patients through COX proportional hazards modeling (sub-aim 2B). This proposed research will yield: (1) improved risk prediction models accounting for new (sub- agatston) and understudied features (socioeconomic status). (2) new insights into atherosclerosis onset and progression (3) technical innovations in machine learning and causal inference for medical imaging. (4) translational software tools suitable for clinical risk assessment. This proposal, levering big CT data analytics, will provide crucial knowledge on the prognostic value of small, low-density calcifications detected on CT images, advance preventative cardiovascular medicine through the early detection of heart disease, and enable a new understanding of atherosclerosis initiation and progression.