In a study recently published in the NPJ | Cardiovascular Health, researchers developed a deep learning framework to predict cardiovascular mortality risks from aortic disease in heavy smokers.
Study: Deep learning to predict cardiovascular mortality from aortic disease in heavy smokers. Image Credit: Random.graphy/Shutterstock.com
The approach aimed to improve how we assess cardiovascular risk by using imaging data from non-contrast chest CT scans, specifically analyzing features of the thoracic aorta.
The study showed that this deep learning technique could provide more accurate predictions than traditional methods, opening up new possibilities for personalized prevention in high-risk groups.
Cardiovascular Disease and Existing Risk Assessment Methods
New advances in artificial intelligence (AI) and deep learning are reshaping how we approach medical imaging and diagnostics. By analyzing complex data patterns, deep learning can automatically identify and measure anatomical and pathological features in imaging studies, helping to improve the accuracy of cardiovascular risk assessments.
Coronary artery disease (CAD) is a leading cause of illness and death worldwide, which highlights the need for reliable ways to predict risk. Traditional tools, like the Framingham risk score and the ASCVD risk score, mainly look at factors like age, gender, and medical history.
However, these tools often fall short for people with early signs of atherosclerosis, as they do not account for certain hidden risk factors.
Using imaging techniques like non-contrast chest CT scans can help identify additional indicators, such as characteristics of the thoracic aorta, that may better predict cardiovascular events and provide a more comprehensive picture of risk.
Deep Learning for Thoracic Aortic Disease Quantification
In this study, researchers developed and validated a fully automatic deep learning framework that quantified features of thoracic aortic disease directly from non-contrast chest CT scans.
The algorithm was designed to identify individuals at higher risk of cardiovascular mortality, surpassing traditional markers like maximum aortic diameter, standard cardiovascular risk factors, and coronary artery calcium (CAC). By leveraging previously unused imaging data, the study proposed an innovative way to improve existing cardiovascular risk assessment methods.
The researchers used data from the National Lung Screening Trial (NLST), which included 24,770 participants, mostly heavy smokers aged 55 to 74. They trained the deep learning model on a subset of these CT scans, employing a hierarchical approach to segment the thoracic aorta and quantify key features, including maximum diameter, volume, and calcifications.
These features were then analyzed for their association with cardiovascular mortality. The team applied Cox proportional hazards regression models to adjust for traditional cardiovascular risk factors, resulting in a more precise assessment of mortality risk.
Key Findings and Insights
The study revealed that both thoracic aortic volume and calcifications were stronger predictors of cardiovascular mortality than the maximum aortic diameter, which is commonly used in clinical settings.
Specifically, the c-index values—an indicator of predictive accuracy—were 0.61 for maximum diameter, 0.63 for volume, and 0.70 for calcifications, highlighting the added value of volume and calcifications in predicting mortality risk.
Over a median follow-up of 6.5 years, 440 participants (about 1.8 %) died from cardiovascular causes. Those with higher-risk aortic features, such as a maximum diameter of 4.5 cm or more and increased calcifications, showed poorer survival rates. Notably, the hazard ratios were 3.68 for a diameter of ≥4.5 cm and 2.19 for a volume of ≥210 ml, indicating a strong link between these features and cardiovascular mortality.
The study also found that adding thoracic aortic volume and calcifications to existing risk models improved predictive accuracy. This suggests that measuring these features from routine chest CT scans could help refine risk assessments and identify individuals who might benefit from targeted prevention.
Potential Applications in Clinical Practice
This research could make a real difference in how we assess cardiovascular risk. By using existing imaging data from non-contrast chest CT scans, healthcare providers can more accurately identify people at high risk, allowing for earlier interventions and customized preventive care—especially beneficial for heavy smokers and other high-risk groups.
Additionally, integrating these deep learning tools into electronic medical record systems could simplify the analysis of imaging data. This would allow for real-time risk assessments without the need for extra imaging procedures. This approach matches modern trends toward personalized medicine, where clinical decisions are based on individual patient characteristics and risk profiles.
Enhancing Cardiovascular Risk Assessment Through AI
In short, this new deep learning algorithm showed real promise for measuring thoracic aortic features from non-contrast chest CT scans, offering valuable insights into cardiovascular mortality risk.
The research suggests that these advanced imaging techniques could improve traditional risk assessment methods, helping to better identify people at high risk for cardiovascular events.
Looking ahead, more studies across diverse populations and in different clinical settings are needed to confirm these findings. Clinical trials could also help determine how adding AI-driven aortic measurements into regular medical care might affect patient outcomes. As medical imaging technology advances, using AI in this way could transform cardiovascular risk assessment and lead to better, more personalized patient care.
Journal Reference
Rau, A., Michel, L., Wilhelm, B. et al. Deep learning to predict cardiovascular mortality from aortic disease in heavy smokers. npj Cardiovasc Health 1, 28 (2024). DOI: 10.1038/s44325-024-00029-3, https://www.nature.com/articles/s44325-024-00029-3
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.