AI Tool Unveils Hidden Heart Risks in Routine CT Scans, Predicting Future Cardiovascular Events Accurately

Mass General Brigham researchers, in collaboration with the United States Department of Veterans Affairs (VA), have developed a new AI tool called AI-CAC to identify individuals at high risk for cardiovascular events based on coronary artery calcium (CAC) levels in chest CT scans. This innovative approach is particularly significant because it uses nongated CT scans, which are routinely performed for other purposes like lung cancer screening, rather than the more specialized gated CT scans. The study, published in NEJM AI, demonstrates that AI-CAC can accurately detect and quantify CAC in these nongated scans. The researchers trained the deep learning algorithm on a vast dataset of chest CT scans from veterans treated at 98 VA medical centers. When tested on 8,052 CT scans, AI-CAC achieved 89.4% accuracy in identifying the presence of CAC and 87.3% accuracy in determining whether the CAC score was above or below 100, indicating a moderate cardiovascular risk. Importantly, the AI-CAC model was also effective in predicting 10-year all-cause mortality. Patients with a CAC score over 400 were found to have a 3.49 times higher risk of death over a 10-year period compared to those with a score of zero. Four cardiologists confirmed that 99.2% of the patients identified by AI-CAC with very high CAC scores (greater than 400) would benefit from lipid-lowering therapy, further validating the tool's clinical utility. Hugo Aerts, PhD, director of the Artificial Intelligence in Medicine (AIM) Program at Mass General Brigham and senior author of the study, emphasized the potential of AI-CAC to revolutionize cardiovascular risk assessment. "Millions of chest CT scans are performed each year, often in healthy individuals," Aerts stated. "Our research shows that valuable information about cardiovascular risk is currently overlooked in these scans. By leveraging AI, we can help clinicians identify and manage this risk proactively, potentially preventing heart disease before it leads to a cardiac event." Raffi Hagopian, MD, a cardiologist and researcher at the VA Long Beach Healthcare System and first author of the study, pointed out the practical implications of using AI-CAC with existing imaging data. "Approximately 50,000 gated chest CT scans are available within the VA system, but there are millions of nongated scans that could be utilized for cardiovascular risk evaluation," Hagopian noted. "This tool has the potential to shift medical practice from a reactive to a proactive paradigm, reducing long-term health complications and costs." The study highlights the growing role of AI in improving medical diagnostics and patient care. By integrating AI-CAC into routine clinical practices, healthcare providers can take a more preventive approach to managing cardiovascular health, potentially catching high-risk patients early and initiating appropriate interventions. However, the researchers acknowledge limitations, such as the algorithm being trained on a veteran population, which may not fully represent the general public. Future studies are planned to validate the tool’s effectiveness in broader patient demographics and to assess its impact on the management of lipid-lowering therapies. This development underscores the importance of combining AI with existing medical imaging technologies to uncover hidden health risks. Industry insiders praise the potential of AI-CAC, noting that it could significantly enhance patient outcomes by enabling early detection and intervention. The collaboration between Mass General Brigham and the VA showcases the power of interdisciplinary research in advancing healthcare innovation. Mass General Brigham, a leading healthcare organization, and the VA, a large and comprehensive healthcare system, are well-positioned to continue pushing the boundaries of AI in medicine. The initial success of AI-CAC opens the door for further exploration and implementation in various healthcare settings. With additional funding and broader validation, this tool could become a standard part of patient care, helping to reduce the burden of cardiovascular disease on individuals and healthcare systems alike.