Most cardiovascular disease does not announce itself. It develops quietly over years, and by the time symptoms appear, the window for early intervention has often narrowed considerably. Researchers are now reporting that artificial intelligence may help close that gap, by extracting a previously overlooked data point from a scan that millions of people already receive.
The study, which followed nearly 12,000 adults over approximately 16 years, found that AI can measure pericardial fat, the layer of fat surrounding the heart, directly from coronary artery calcium scans. Those scans, commonly known as CAC scans, are non-invasive imaging tests that measure calcium deposits in the coronary arteries. They are already used as an early indicator of heart disease. The AI component adds a second layer of information from the same image, without requiring any additional imaging or procedures.
What pericardial fat reveals about cardiovascular risk
Pericardial fat volume, as measured by the AI model in this study, was independently associated with a higher likelihood of developing cardiovascular disease. When researchers incorporated those measurements into existing risk models, predictive accuracy improved. The gains were most pronounced for patients in the low or intermediate risk categories, a group that current tools tend to handle with less precision.
That matters clinically because the patients who are hardest to assess accurately are often the ones whose treatment decisions carry the most uncertainty. A patient who falls just below the threshold for intervention may benefit from closer monitoring or earlier lifestyle changes, but without a clear signal from standard risk tools, clinicians have limited grounds to act. AI-derived pericardial fat measurements could provide that signal.
The current standard for estimating cardiovascular risk involves models like the American Heart Association’s PREVENT equation, which factors in age, blood pressure, cholesterol levels, and diabetes status, often combined with a CAC score. These tools are well-validated and widely used. The argument this study makes is not that they should be replaced, but that pericardial fat volume can complement them, particularly for patients whose scores land in the gray zone between low and high risk.
How the AI model was trained and what it can do
The model was trained on a set of manually annotated images, which allowed it to learn to identify and measure pericardial fat volume with consistency across a large dataset. The process does not require a radiologist to manually measure the fat each time. Once trained, the model can extract the measurement from a standard CAC scan automatically, making it practical to incorporate into existing workflows without adding significant time or cost.
Dr. Zahra Esmaeili, one of the researchers involved in the study, noted that elevated pericardial fat may carry meaningful risk information for patients with low coronary calcium scores. That is a significant finding because a low CAC score is often interpreted as reassuring. If a patient has low calcium deposits but a high volume of fat surrounding the heart, their actual cardiometabolic risk may be higher than the score alone suggests. This is particularly relevant for patients with a normal body mass index, who might not otherwise be flagged for additional cardiovascular monitoring.
What needs to happen before this reaches clinical practice
The researchers are careful to note that further work is needed before AI-derived pericardial fat measurements become a standard part of cardiovascular assessment. The current findings are promising, but translating a research tool into routine clinical use requires validation across broader and more diverse populations, as well as evidence that acting on the additional information actually improves patient outcomes.
The direction is clear, though. CAC scans are already part of the preventive cardiology toolkit. If AI can reliably extract additional predictive information from an image that patients are already receiving, it represents a low-friction path toward more precise risk stratification. For the patients currently sitting in the uncertain middle of cardiovascular risk categories, that precision is exactly what current tools have been missing.
