AI Technology Uncovers Undetected Heart Disease in Existing Patient Scans

Recent advancements in artificial intelligence (AI) have led to the development of a new tool designed to analyze previously collected chest CT scans for indications of hidden heart disease. This innovative AI system, known as AI-CAC, was created through a collaboration between researchers at Mass General Brigham and the United States Department of Veterans Affairs (VA). The findings of this research were published in the journal NEJM AI.

Every year, millions of chest CT scans are performed, often on patients without any apparent health issues, typically for lung cancer screenings. However, these scans may contain critical data regarding cardiovascular risks that often go unnoticed. The senior researcher on the project, Dr. Hugo Aerts, emphasizes the potential for AI to transform clinical practices by enabling healthcare providers to engage with patients sooner, potentially preventing the progression of heart disease.

The AI-CAC tool effectively identifies coronary artery calcium (CAC) deposits, which are associated with an increased risk of heart attacks. While the traditional method for measuring CAC involves gated CT scans synchronized with the heartbeat to minimize motion during imaging, the majority of routine chest CT scans are non-gated. Recognizing this gap, the research team developed AI-CAC to extract and quantify CAC data from these non-gated scans.

In developing the AI algorithm, the researchers utilized chest CT scans from a broad sample of veterans across 98 VA medical centers. They subsequently tested the AI-CAC's efficacy on a dataset of 8,052 CT scans, simulating a routine screening for CAC. The results indicated that the AI tool achieved an impressive accuracy rate of 89.4% in detecting the presence of CAC. Additionally, for scans exhibiting CAC, the model demonstrated an accuracy of 87.3% in assessing whether the calcium score was moderate (over 100), which signifies a significant cardiovascular risk.

Moreover, the AI-CAC model proved to be predictive of long-term outcomes, revealing that individuals with a CAC score exceeding 400 had a 3.49 times greater risk of mortality over a decade compared to those with no CAC. Among the patients categorized as having high CAC levels by the AI, nearly all (99.2%) were deemed likely to benefit from lipid-lowering treatments, as confirmed by a panel of cardiologists.

The implications of this research are profound, as the VA currently possesses millions of non-gated chest CT scans that could be repurposed for cardiovascular risk assessments. Dr. Raffi Hagopian, the study's lead author and a cardiologist, noted that employing AI in such a manner could transition healthcare from a reactive model to a proactive one, potentially reducing the incidence of severe health issues and associated healthcare costs.

Despite these promising results, the study does have limitations, notably that the algorithm was trained exclusively on a veteran population. The research team plans to conduct future studies that will include a broader demographic to evaluate the AI's effectiveness across various populations and its ability to monitor changes in CAC scores in response to lipid-lowering therapies.