For decades, cardiovascular imaging has concentrated heavily on a visually intuitive problem: how blocked is the artery? A severe coronary narrowing can restrict blood flow and clearly identifies disease, yet the relationship between visible blockage and future heart attacks is far from perfect. Many major cardiovascular events occur in people whose coronary CT scans show non-obstructive disease rather than the dramatic narrowing patients normally associate with an impending heart attack.
Artificial intelligence is beginning to interrogate something different inside those same images: inflammation.
Caristo Diagnostics’ CaRi-Heart software has received U.S. Food and Drug Administration De Novo authorization as a Class II device capable of assessing coronary vascular inflammation from coronary computed tomography angiography and estimating cardiovascular mortality risk using that information alongside clinical risk factors. The authorization covers adults aged 30 to 80 who have already been referred for coronary CT angiography.
How can a CT scan detect inflammation around a coronary artery?
The key is not simply examining the inside of the blood vessel.
Inflammation within the coronary artery wall alters biological signaling to the fat surrounding that artery. Those inflammatory signals can change the size, composition and lipid content of nearby fat cells.
Those changes influence the density of the surrounding tissue on CT imaging.
Researchers developed a measurement known as the perivascular fat attenuation index, or FAI, to quantify these changes. Instead of treating the fat around the coronary artery as irrelevant background tissue, the method effectively uses it as a sensor of inflammatory activity occurring within the adjacent vessel wall.
CaRi-Heart standardizes that information and incorporates it with other factors to create an individualized cardiovascular-risk assessment.
The scientific trick is elegant: the software does not need to directly biopsy an inflamed coronary artery because it measures how that inflammation changes the tissue around it.
Why is inflammation different from measuring plaque and artery blockage?
Traditional coronary CT angiography is excellent at showing plaque and determining whether a coronary artery has become narrowed.
But a heart attack frequently occurs when an atherosclerotic plaque becomes unstable and ruptures, triggering rapid clot formation.
The plaque responsible does not necessarily need to have produced severe obstruction beforehand.
Caristo notes that a large proportion of major cardiovascular events arise in patients who do not have obstructive coronary disease on their initial CT imaging. Research evaluating FAI and CaRi-Heart has therefore focused on identifying inflammatory activity and risk that anatomical narrowing alone may underestimate.
This changes the clinical question from “How narrow is this artery?” toward “How biologically active and dangerous is the disease inside this artery?”
What exactly did the FDA authorize CaRi-Heart to report?
The FDA defines CaRi-Heart as prescription software intended to assist healthcare professionals in patient management.
It analyzes coronary CT angiography images to assess vascular inflammation and provides an estimate of cardiovascular mortality risk associated with coronary inflammation and other clinical risk factors. The output must be interpreted alongside the original CT images, symptoms, history, diagnostic tests and physician judgment rather than being used as a standalone diagnosis.
This boundary matters.
CaRi-Heart does not tell a patient that a heart attack will occur on a particular date. It estimates longer-term cardiovascular risk by extracting biological information that conventional image interpretation may not routinely quantify.
That makes it a predictive medical device rather than a crystal ball.
Does coronary-inflammation AI actually change treatment decisions?
Early real-world research suggests that it can.
A 2025 study involving 164 patients undergoing coronary CT angiography found that adding FAI and an AI-derived risk estimate changed cardiovascular-risk classification in a substantial proportion of patients. Clinical management changed in approximately one-third after clinicians received the additional CaRi-Heart information.
The study is small and does not prove that every treatment change improves outcomes, but it demonstrates why an apparently abstract AI score can become clinically relevant.
A patient whose CT shows no major obstruction might ordinarily receive reassurance and routine preventive advice. If inflammation analysis instead identifies unexpectedly high risk, the physician might intensify cholesterol lowering, address blood pressure more aggressively or investigate other preventive strategies.
Conversely, additional risk information might prevent unnecessary escalation in genuinely lower-risk patients.
Could every coronary CT eventually become a heart-attack prediction test?
Technically, that is one of the most attractive possibilities.
Coronary CT angiography is already performed in large numbers of patients with suspected coronary artery disease. If an AI system can extract additional prognostic information from images already collected, healthcare systems do not need another scan to obtain that information.
This is sometimes described as extracting latent biomarkers from existing medical data.
The same principle is appearing elsewhere in medicine. Artificial intelligence can extract cardiac-risk signals from electrocardiograms, detect additional abnormalities in retinal photographs and derive prognostic features from radiology scans acquired for unrelated clinical purposes.
Medicine may therefore generate considerably more information than clinicians currently use.
AI provides a way to go back into those existing datasets and ask questions the original test was never designed to answer.
Could coronary inflammation be treated before an artery becomes severely blocked?
This is where the technology becomes potentially transformative.
If inflammation identifies high-risk disease earlier than conventional anatomical measures, physicians could intervene during a stage when the artery still appears relatively open.
Preventive therapy might involve intensive lipid reduction, blood-pressure treatment, smoking cessation, weight management or other interventions appropriate to the individual patient.
The underlying concept is not that AI itself prevents heart attacks. AI could identify which patients deserve more aggressive prevention.
That distinction is essential because predictive algorithms create value only when their prediction changes an actionable clinical decision.
What are the risks of relying on an AI heart-risk score?
Overtreatment is one.
A false high-risk prediction could cause anxiety, additional testing and more aggressive medication than the patient actually requires.
Underestimation creates the opposite danger if a patient or physician places too much confidence in a low-risk result.
Algorithm performance can also vary when technology moves between hospitals, scanners and patient populations.
That is why the FDA indication specifically requires CaRi-Heart outputs to be reviewed with other clinical evidence and professional judgment.
Another challenge is explaining probabilistic results to patients. “You have coronary inflammation associated with higher long-term cardiovascular risk” is much less intuitive than “this artery is 80% blocked.”
The future of predictive cardiology will therefore require better communication alongside better algorithms.
Could AI eventually show whether cardiovascular treatment is working?
This may become the next major step.
If inflammation can be measured quantitatively, researchers can investigate whether successful therapy produces measurable changes in that inflammation over time.
That could eventually create an imaging biomarker for treatment response.
Rather than waiting years to discover whether a patient suffers a cardiovascular event, clinicians might compare biological risk signals before and after aggressive preventive treatment.
Such use would require separate evidence and regulatory validation; the existing authorization should not be interpreted as automatically approving serial treatment monitoring.
Nevertheless, it illustrates where quantitative imaging is heading.
Will heart disease eventually be diagnosed before symptoms begin?
Cardiovascular medicine is moving steadily in that direction.
Traditional cardiology often encounters disease after chest pain begins, a stress test becomes abnormal or a heart attack has already occurred.
Future systems could combine genetics, cholesterol, blood biomarkers, wearable information, coronary plaque, vascular inflammation and artificial intelligence to identify dangerous disease years earlier.
The greatest obstacle may not be detecting risk but deciding which signals are strong enough to justify treatment.
An algorithm capable of finding abnormalities in millions of people is easy to imagine. A healthcare system capable of acting appropriately on those findings without creating unnecessary testing and anxiety is considerably harder to build.
CaRi-Heart provides an early glimpse of that future because it changes what physicians can ask from an existing coronary CT scan.
The scan no longer needs to answer only whether an artery is blocked. It can begin answering a more consequential question: does the biology surrounding that artery suggest that this patient is heading toward trouble even though the blockage does not yet look dramatic?
If predictive imaging proves that acting on those signals prevents heart attacks, cardiovascular medicine could move from detecting dangerous anatomy toward detecting dangerous biology.