AI-Assisted Chest CTs Spot Heart Risks Early 

An AI-based model can use routine chest CTs to accurately estimate the risk of major adverse cardiovascular events, according to research presented Thursday.

“Traditional risk models for major adverse cardiovascular events like heart attacks and cardioembolic strokes do not incorporate radiologic imaging directly,” said Amara Tariq, PhD, a data science analyst at Mayo Clinic in Scottsdale, AZ. “Instead, they rely on known clinical and demographic risk factors. These models do not achieve optimal performance when applied to external institutions outside of the clinical and demographic distributions that they were trained on.” 

Working with colleagues from Mayo Clinic and Emory University in Atlanta, Dr. Tariq helped develop a 3D convolutional neural network (CNN) model to estimate major adverse cardiovascular events risk from non-contrast chest CT. By using two populations with significant differences in demographic makeup, the researchers were able to better study the model’s generalizability.  

“We wanted to quantify the risk of cardiovascular disease in a way that is both widely applicable and has fair performance for all demographic subgroups,” Dr. Tariq said. 

The model incorporated causal intervention to enhance its ability to establish genuine cause-and-effect relationships and reduce spurious associations between socioeconomic factors and disease risk. 

Model Boosts Fairness in Cardiac Risk Prediction

Dr. Tariq and colleagues used 4,431 CTs from 4,231 patients at the Mayo Clinic for the internal cohort. A random set of 201 CTs from 191 patients at Emory without a history of major adverse cardiovascular events was used for external validation. 

The 3D CNN with and without causal intervention achieved comparable performance on the internal test set while outperforming a baseline machine learning model that used demographic and body composition statistics computed from CT. In external validation, 3D CNN with causal intervention outperformed 3D CNN without causal intervention, proving the benefits of causal intervention for model generalization.

“The initial results are very promising,” Dr. Tariq said. “Our model is not only more accurate than comparative models but is also more equitable and fairer.” 

Dr. Tariq emphasized that clinically, the model has the potential to provide earlier detection of cardiac risk and speed medical and lifestyle interventions across diverse patient populations beyond the capabilities of current clinical tools. 

For example, a patient getting routine non-contrast chest CT for any reason could also get an accurate risk assessment for future adverse cardiovascular events without any additional imaging. 

Using routine CT over more expensive imaging studies like CT with contrast to measure coronary calcium in the heart brings the predictive power of the model to more patients. 

“Our model uses routinely performed, non-contrast chest CT so that we can target a larger population,” Dr. Tariq said. “This not only provides accurate protection against future cardiovascular events, it also delivers fair and equitable performance across different subgroups.” 

The project is ongoing, with plans to build multimodal foundational models combining different imaging studies with clinical information for more accurate cardiovascular risk assessment.

Access the session, “Opportunistic Screening for Adverse Cardiovascular Events Using Routine Clinical Chest Computed Tomography – A Study on Performance and Generalization,” (R1-SSCA09-5) on demand at RSNA.org/MeetingCentral.