Coronary Artery Inflammation AI: The Scan That Sees Risk Before It Strikes

Heart attacks kill without warning. That is the uncomfortable truth that cardiologists have wrestled with for decades: the majority of people who suffer a first myocardial infarction had cholesterol levels, blood pressures, and stress tests that gave no obvious signal of impending catastrophe. The problem is not that these standard tools are useless. The problem is that they measure the wrong thing. They capture the size of the fire, but they miss the spark. That spark, increasingly, is understood to be inflammation in and around the coronary arteries, and a new generation of artificial intelligence tools trained on cardiac CT data is now making it possible to detect that inflammation years, perhaps decades, before a plaque ruptures.

The technology in question is built around a signal called the Fat Attenuation Index (FAI), a subtle difference in the X-ray density of the fat tissue that wraps around coronary arteries. When a vessel wall is inflamed, the surrounding fat changes its composition in ways that are invisible to the human eye on a CT scan but detectable by machine learning algorithms trained to spot the difference. Researchers at the University of Oxford first described this principle in 2018, and subsequent clinical trials have validated it against hard cardiovascular outcomes. The result is a diagnostic capability that does not replace standard cardiac imaging but adds a dimension to it that standard imaging was never designed to provide: a readout of active vascular inflammation, not just anatomical plaque burden. This article explores the science behind coronary artery inflammation AI, the clinical evidence that supports it, and what it means for a preventive cardiology approach aimed at extending healthspan.

Why Inflammation, Not Just Plaque, Is the Target

Atherosclerosis has long been framed as a plumbing problem. Fat accumulates inside arterial walls, plaques build up, and eventually a pipe narrows enough to restrict flow. This model is intuitive and not entirely wrong, but it misses the most dangerous chapter of the story. Most heart attacks do not occur at the site of the most severely narrowed artery. They occur when a relatively modest plaque, one that might never have caused a symptom or even shown up as a significant finding on a stress test, becomes unstable and ruptures. The rupture exposes the contents of the plaque to the bloodstream, triggering a clot that can occlude the vessel in minutes.

What determines whether a plaque is stable or vulnerable? The answer is inflammation. An actively inflamed plaque has a thin fibrous cap stretched over a lipid-rich necrotic core, like a blister about to break. The inflammatory cells inside it, particularly macrophages that have gorged on oxidized LDL and transformed into foam cells, secrete enzymes that digest the collagen scaffold holding the cap together. This process, called plaque vulnerability, is driven by the same chronic low-grade inflammatory signaling that underlies so many other aspects of biological aging. [1]

The clinical evidence for inflammation as a causal driver of cardiovascular risk, independent of LDL cholesterol, has become overwhelming over the past decade. The JUPITER trial showed that statin therapy in people with normal LDL but elevated high-sensitivity C-reactive protein (hsCRP) dramatically reduced cardiovascular events. [2] The CANTOS trial then demonstrated that directly targeting the interleukin-1 beta inflammatory pathway with canakinumab, entirely apart from any lipid effect, reduced recurrent heart attacks. [3] The message is unambiguous: inflammation in the arterial wall is not a bystander. It is a perpetrator. The challenge has always been measuring it with enough precision and localisation to act on it clinically. That is exactly the gap that coronary artery inflammation AI is designed to close.

The Fat Attenuation Index: Reading Inflammation Through Fat

The coronary arteries do not sit in empty space. They are embedded in a sleeve of fat called pericoronary adipose tissue (PCAT), and this fat is not inert. It communicates with the underlying vessel wall through a continuous exchange of signaling molecules, cytokines and adipokines that travel outward from inflamed endothelium and inward from metabolically active fat cells. When the vessel wall is actively inflamed, this crosstalk drives changes in the lipid composition of the surrounding fat, specifically an increase in water content and a decrease in lipid droplet density, that shift the fat's attenuation value on CT imaging toward less negative Hounsfield units.

The Fat Attenuation Index quantifies exactly this shift. On a standard non-contrast or contrast-enhanced cardiac CT, the algorithm measures the mean CT attenuation of the pericoronary fat within a defined radial distance from the outer wall of each coronary artery. Healthier, more lipid-rich fat reads at around -90 Hounsfield units. Fat adjacent to an inflamed vessel shifts toward -70 Hounsfield units or above. The difference, roughly 20 units, sounds small in absolute terms, but it is reproducible, it tracks with histological markers of inflammation, and, critically, it predicts outcomes. [4]

The Fat Attenuation Index transforms a standard CT scan into a window onto vascular biology that was previously visible only under a microscope, after a patient had already died.

The original Oxford group published the proof-of-concept data in Nature Medicine in 2018, demonstrating that FAI around the right coronary artery predicted all-cause and cardiac mortality in two independent cohorts totalling over 3,900 patients. [4] Patients in the top quartile of FAI had a more than three-fold higher risk of cardiac death compared to those in the bottom quartile, even after adjustment for conventional risk factors, plaque burden, and coronary artery calcium score. The finding was not a modest statistical association. It was a signal strong enough to suggest that FAI was capturing something fundamentally different, and more prognostically important, than anything the existing imaging toolkit could see.

From Signal to Score: The AI-CT-FFR and Cardiac Risk Index

Raw FAI values are useful, but clinical decision-making requires a more integrated picture. The next development was training machine learning models to combine the FAI signal with other CT-derived features, including plaque volume, plaque composition, coronary calcium score, and measures of myocardial perfusion, to generate a single composite cardiac risk index. The Oxford group's platform, now commercialized as the AI-Risk tool within the CaRi-Heart system, received CE marking in Europe in 2021 and has been validated in large prospective cohorts. [5]

The key validation study, published in the European Heart Journal in 2022, followed 3,912 patients who had undergone clinically indicated coronary CT angiography (CCTA). The AI-derived cardiac risk index, incorporating FAI alongside anatomical plaque features, predicted major adverse cardiovascular events (MACE) with a C-statistic of 0.79, substantially outperforming both the conventional clinical risk score (C-statistic 0.61) and the coronary artery calcium score alone. [5] Translated from statistical language: the AI tool correctly identified patients who would go on to suffer heart attacks or cardiac deaths in roughly four out of five cases, compared to roughly six in ten for standard risk assessment. That is not an incremental improvement. It is a reclassification of risk for a meaningful fraction of patients.

Particularly notable was the tool's performance in patients who were classified as intermediate or low risk by conventional metrics. Among those who standard assessment would have labelled as not requiring aggressive intervention, a subset had high AI-derived FAI scores, and these individuals went on to experience events at rates comparable to the high-risk group. This is the clinical scenario that current cardiology struggles with most acutely: the patient with a normal or near-normal lipid panel, no symptoms, and perhaps a modest calcium score, who nonetheless harbors active vascular inflammation that will eventually culminate in a sudden, apparently unexpected coronary event. [5]

The CRISP-CT Trial and Prospective Evidence

Observational data, however compelling, cannot establish clinical utility on its own. The CRISP-CT study addressed this by prospectively enrolling 1,872 patients from two independent international cohorts and evaluating whether FAI-based risk stratification provided incremental prognostic value beyond the established clinical standard, the Duke Clinical Risk Score combined with coronary CT findings. [6] The results were published in JAMA Cardiology in 2021 and confirmed the original Oxford findings in an entirely independent population.

FAI of the right coronary artery was associated with a hazard ratio of 2.15 for cardiac mortality across a median follow-up of 4.7 years, independent of all established risk factors. [6] Net reclassification improvement analysis showed that adding FAI to the standard assessment correctly reclassified 9.3% of patients who subsequently experienced events and 7.8% of those who did not. In absolute terms this means: for every 100 patients assessed, approximately 9 who would otherwise have been incorrectly labelled low-risk were correctly identified as high-risk, and 8 who would have been incorrectly labelled high-risk were correctly reassured. In a disease responsible for one in five deaths globally, those numbers represent enormous absolute benefit at the population level.

The CRISP-CT study also explored a biologically important question: is FAI measuring a fixed anatomical property of pericoronary fat, or is it a dynamic readout of ongoing inflammation that can change over time? Serial CT data in a subset of patients showed that FAI values do change, and that changes in FAI track with changes in clinical status, including the initiation of statin therapy, which is known to have anti-inflammatory effects beyond its lipid-lowering action. [6] This dynamism is crucial: it means FAI is not just a risk flag but potentially a treatment response biomarker, a way of confirming whether interventions aimed at reducing vascular inflammation are actually working at the target site.

Coronary Artery Inflammation AI and the Broader Landscape of Cardiac Imaging

To appreciate what FAI-based AI adds, it helps to understand what existing cardiac imaging tools do and do not capture. The coronary artery calcium (CAC) score, derived from a non-contrast CT scan, quantifies calcified plaque in the coronary arteries. It is an excellent predictor of long-term cardiovascular risk and is now widely used in clinical guidelines to guide statin and aspirin decisions. But calcium in a plaque is actually a marker of relative stability: calcification is the body's way of walling off old, burnt-out plaque. High calcium scores identify patients with extensive prior disease but may miss the young, actively inflamed non-calcified plaque that is most likely to rupture. [7]

Coronary CT angiography goes further, providing detailed images of both calcified and non-calcified plaque and allowing assessment of stenosis severity. AI-assisted analysis of CCTA data for plaque characterization, including high-risk plaque features such as positive remodeling, low attenuation plaque, and spotty calcification, has itself become an active area of investigation. The HeartFlow FFR-CT platform, which uses computational fluid dynamics to estimate fractional flow reserve from CCTA data, is another validated AI application in cardiac imaging. [8] FAI sits alongside these tools as a complementary, not competing, measurement. Where HeartFlow asks "how much is this plaque restricting blood flow?" and CCTA plaque analysis asks "what does this plaque look like?", FAI asks "how inflamed is this vessel right now?" Each question addresses a different dimension of cardiovascular risk.

Calcium scores measure yesterday's fire. The Fat Attenuation Index measures the heat that is present today.

Positron emission tomography (PET) imaging with fluorodeoxyglucose (FDG) can also detect vascular inflammation by identifying metabolically active macrophages within plaque, but it is expensive, requires radiation exposure from two modalities, and is not widely available in routine clinical settings. FAI derived from CCTA that a patient may already be having for other indications represents a far more scalable solution: no additional radiation, no additional cost beyond the computational analysis, and information available from imaging that is already being performed. [4]

The Biology Connecting Systemic Inflammation and Pericoronary Fat

Understanding why pericoronary fat is such a sensitive reporter of vascular inflammation requires a brief excursion into the biology of adipose tissue as an active endocrine organ. Fat has long been thought of as passive storage. It is nothing of the sort. Adipose tissue secretes dozens of signaling molecules, collectively termed adipokines, that regulate insulin sensitivity, appetite, immunity, and vascular tone. The adipose tissue that surrounds coronary arteries is developmentally and functionally distinct from subcutaneous fat and is in particularly intimate communication with the underlying vessel wall, separated from the adventitia by only a thin connective tissue layer. [9]

When endothelial cells in the coronary artery wall activate an inflammatory program, as they do in response to oxidative stress, dyslipidemia, hypertension, hyperglycemia, or the direct mechanical effects of turbulent blood flow at arterial branch points, they release cytokines including TNF-alpha, IL-6, and reactive oxygen species into the periadvential space. These signals reach the surrounding fat almost immediately. The fat cells respond by altering their lipid metabolism: they begin releasing stored lipid and accumulating water, and in doing so, they shift the attenuation of the tissue on CT. Think of it as the fat tissue acting like a mood ring for the vessel underneath it, changing its signal in response to the biochemical environment the artery is broadcasting. [9]

This biology has a direct implication for the relationship between metabolic health and coronary inflammation. Obesity, particularly visceral adiposity, chronic hyperglycemia, and insulin resistance are all associated with a pro-inflammatory adipokine profile that amplifies coronary artery inflammation from both the inside and outside of the vessel wall simultaneously. This is one reason why cardiometabolic risk is so much more than just a cholesterol problem. The same metabolic milieu that drives visceral fat accumulation is also stoking the pericoronary inflammatory fire that FAI can now detect. [7]

AI Architecture: How the Algorithm Sees What the Eye Cannot

The technical underpinnings of the FAI measurement deserve attention because they explain why this is genuinely a machine learning problem rather than simply a new measurement that radiologists could perform manually. Pericoronary fat is an irregular, three-dimensional structure that must be segmented from surrounding tissues including the myocardium, the lung, and other mediastinal fat depots, each of which has overlapping attenuation ranges. Doing this accurately and reproducibly by hand for three coronary arteries across dozens of CT slices would take a trained analyst approximately two hours per scan. The AI pipeline reduces this to minutes and does so with less operator-dependent variability. [5]

The segmentation component uses convolutional neural networks trained on thousands of manually annotated CCTA datasets to identify coronary artery centrelines and outer vessel walls, then extracts all voxels within a defined radial band around each artery that fall within the attenuation range of fat (-190 to -30 Hounsfield units). The mean attenuation of this region is the raw FAI. This raw value is then fed into the composite risk model alongside other CT-derived features to generate the final cardiac risk index. The composite model itself uses gradient boosting algorithms trained and validated on outcome data, with MACE and cardiac mortality as the endpoints. [5]

Crucially, the model was not developed on a single homogeneous dataset and then published. It was trained and validated across multiple international cohorts with different scanner manufacturers, imaging protocols, and patient demographics. Robustness across these variables is a prerequisite for clinical deployment, and the published data support it. The one caveat that investigators have consistently acknowledged is that the tool was validated primarily in patients who had already been referred for CCTA based on some level of clinical suspicion. Whether its performance characteristics hold equally well in truly asymptomatic screening populations remains an active area of investigation. [6]

Clinical Implementation: Who Should Be Scanned, and What Happens Next

The question of which patients should receive FAI analysis sits at the intersection of clinical cardiology, health economics, and preventive medicine philosophy. Current evidence is strongest for patients who are already undergoing CCTA for clinical indications, including evaluation of chest pain, assessment of intermediate coronary stenosis, and pre-surgical cardiac evaluation. For these patients, adding FAI analysis to the existing scan is a computational step with no additional patient burden and potentially significant prognostic benefit. Several European centers have begun integrating the CaRi-Heart platform into their standard CCTA reporting workflow on this basis. [5]

The more contested and more interesting question is whether FAI should be used in asymptomatic individuals presenting for cardiovascular risk assessment in a preventive or longevity medicine context. The argument for this is compelling. A significant proportion of people who appear low risk by conventional metrics, normal blood pressure, LDL below treatment threshold, no diabetes, non-smoker, harbor the kind of active pericoronary inflammation that FAI can detect and that predicts events. If they can be identified and treated more aggressively before a plaque ruptures, the benefit is enormous. The argument against is that the evidence base in purely asymptomatic populations is still being built, and that identifying high FAI without clear treatment pathways risks generating anxiety without clinical benefit. [6]

What is clear is that a finding of elevated FAI changes the clinical calculus. Patients with high pericoronary inflammation scores are likely to benefit from more intensive lipid-lowering therapy, from anti-inflammatory interventions such as colchicine (now supported by the LoDoCo2 and COLCOT trials for secondary prevention), and from aggressive optimization of the metabolic factors that drive vascular inflammation. [10] For patients managing cardiometabolic risk through programs targeting insulin resistance and visceral adiposity, such as GLP-1 Longevity Care or the SGLT2 Protocol, FAI may eventually serve as an objective biomarker of whether metabolic interventions are translating into reduced coronary inflammation at the vessel wall level. Both GLP-1 receptor agonists and SGLT2 inhibitors have demonstrated cardiovascular outcome benefits that appear to exceed their metabolic effects alone, and pericoronary fat inflammation is a plausible biological mediator of those benefits. [11]

Integrating FAI into a Longevity-Oriented Cardiovascular Workup

Longevity medicine differs from conventional cardiology in its orientation: rather than waiting for symptoms or established disease to trigger investigation, it seeks to identify the earliest biological signals of dysfunction and intervene while the system is still adaptive. Coronary artery inflammation AI fits this framework exceptionally well, because it measures a biological process, active vascular inflammation, rather than its structural consequence, calcified plaque.

In practical terms, a comprehensive cardiovascular longevity assessment now has the potential to include several complementary layers of information. Advanced lipid testing, including LDL particle number, Lp(a), and oxidized LDL, characterizes the atherogenic substrate. Inflammatory biomarkers including hsCRP, interleukin-6, and fibrinogen capture systemic inflammatory tone. Coronary artery calcium scoring provides a structural index of cumulative plaque burden. And FAI-based AI analysis of CCTA adds the dynamic inflammatory readout at the vessel wall. No single test is sufficient alone. Together, they create a picture of cardiovascular biological age that is far richer than conventional risk calculators can generate from age, cholesterol, blood pressure, and smoking status.

The Longevity Optimization program includes advanced biomarker assessment as part of its core protocol, recognizing that meaningful healthspan extension requires understanding a person's biological risk profile at a level of granularity that standard annual physicals cannot provide. Cardiovascular disease remains the leading cause of years of healthy life lost globally, and the ability to detect and act on coronary inflammation years before a plaque ruptures represents exactly the kind of upstream intervention that longevity medicine is designed to enable.

The metabolic dimension of this picture also points toward the SGLT2 Protocol and GLP-1 Longevity Care as therapies with direct mechanistic relevance to pericoronary inflammation. GLP-1 receptor agonists reduce visceral adiposity, lower systemic inflammatory markers, and have demonstrated reductions in MACE in large cardiovascular outcomes trials. SGLT2 inhibitors reduce cardiac preload and afterload, lower uric acid and inflammatory cytokines, and have shown particularly striking reductions in heart failure hospitalizations and cardiovascular mortality. Both classes may be reducing FAI in ways that are now, for the first time, directly measurable. [11]

Limitations and the Road Ahead

Scientific rigor demands acknowledging what this technology cannot yet do. First, FAI analysis requires a CCTA scan, which involves radiation exposure and iodinated contrast administration. It is not a blood test or a wristband. For truly low-risk individuals with no clinical indication for CCTA, the benefit-risk calculation of obtaining a scan solely for FAI assessment is not yet established. Ongoing trials, including the ORFAN (Oxford Risk Factors and Non-Invasive Imaging) study, are prospectively enrolling tens of thousands of patients to address population-level questions about FAI-guided screening. [12]

Second, FAI is currently validated primarily as a prognostic biomarker rather than a therapeutic target in a formal randomized trial sense. There is no trial yet that has randomized patients to FAI-guided treatment versus standard care and demonstrated improved hard outcomes. This trial is logistically challenging and will take years to complete. The available evidence, while compelling, is observational and prospective cohort data, not the randomized controlled trial gold standard. Clinicians and patients should interpret FAI results in this context: as a powerful addition to risk stratification, not as a standalone mandate for any specific intervention. [6]

Third, the technology is currently available at a limited number of centers. The CaRi-Heart platform requires integration with CCTA acquisition and reporting workflows, and access remains concentrated in academic medical centers and specialized cardiovascular imaging programs. This will change as the evidence base expands and health systems recognize the cost-effectiveness of preventing events versus treating them, but the transition will take time. [5]

Looking forward, several developments are likely to extend FAI's clinical reach. First, machine learning models trained on larger and more diverse datasets may be able to extract FAI-equivalent information from non-contrast CT scans, eliminating the need for contrast administration and making the analysis applicable to calcium scoring scans that are already performed at high volume for cardiovascular risk screening. Second, integration of FAI with other AI-derived CT biomarkers, including aortic stiffness estimates, epicardial fat volume, and liver fat fraction, will enable truly comprehensive cardiometabolic phenotyping from a single imaging study. Third, longitudinal FAI tracking is likely to emerge as a treatment response biomarker, closing the loop between intervention and biological effect at the target organ. [12]

A New Vocabulary for Cardiovascular Risk

Every significant advance in medicine eventually changes the vocabulary clinicians and patients use to talk about disease. The introduction of the coronary artery calcium score gave people a concrete, quantitative way to think about their accumulated plaque burden. The concept of biological age, distinct from chronological age, gave longevity medicine a framework for communicating the gap between where someone is and where they could be. FAI-based coronary artery inflammation AI is poised to introduce a third vocabulary: vascular inflammatory age, a readout of how aggressively the immune system is currently attacking the coronary arterial wall.

This framing matters because it changes the emotional and motivational landscape of cardiovascular prevention. A coronary calcium score of zero is reassuring but can breed complacency: it says "no damage so far" but says nothing about the pace of the inflammatory process currently underway. A high FAI score says something different and more actionable: "this vessel wall is under active attack right now, and the attack is measurable, and interventions exist that can reduce it." That is the kind of information that motivates change, justifies clinical investment, and, when acted upon, can genuinely alter the trajectory of a person's cardiovascular life.

The science of coronary artery inflammation has been building for three decades, from the recognition of macrophage foam cells in vulnerable plaques, through the CANTOS trial proof-of-concept for anti-inflammatory therapy, to the current generation of AI imaging tools that can read the inflammatory state of a living coronary artery from the outside. Each step has moved medicine closer to the goal of preventing the first event rather than managing the consequences of it. FAI-based coronary artery inflammation AI is not the last word in this story. But it is, without question, a transformative next chapter.