This is a paper about body fat having a group chat with your heart. Not metaphorically in the fluffy wellness-blog sense. Biologically. Chemically. Possibly with read receipts.

The new review by Khanna, Mann, Bhat, and colleagues in Nature Reviews Cardiology argues that cardiometabolic risk is not just about how much fat someone carries, but where that fat lives, what mood it is in, and which organs it has been texting at 2 a.m. The authors call this the “unified adipose tissue” model: instead of treating fat depots like isolated storage lockers, they frame them as a connected system shaping inflammation, metabolism, blood vessels, and heart remodeling Khanna et al., 2026.

Follow the adipose trail, friends. The evidence is wearing a trench coat.

The Scale Was Always a Little Too Confident

For decades, medicine has leaned heavily on body weight and BMI, which is understandable because they are easy to measure and fit neatly in spreadsheets. But BMI is basically the “vibes-based hiring manager” of health metrics. It sees height and weight, makes a snap judgment, and refuses to ask follow-up questions.

This review pushes a sharper idea: two people can have similar body weight but very different fat biology. Subcutaneous fat under the skin may behave differently from visceral fat around organs. Epicardial adipose tissue, or EAT, sits directly around the heart, tucked inside the pericardial sac like an overly involved neighbor. It can release inflammatory and metabolic signals close enough to influence the coronary arteries and myocardium through paracrine, vasocrine, neural, and endocrine routes.

Coincidence that the heart’s local fat depot gets its own secret communication channels? I think not.

Wikipedia-level background actually helps here: adipose tissue is not just passive padding. It is an endocrine organ, producing hormones and cytokines, and visceral fat has long been linked to insulin resistance, type 2 diabetes, and cardiovascular disease. Epicardial fat is a particularly spicy subplot because it is metabolically active and physically close to cardiac tissue.

The Plot Twist: Fat Has a Personality

The review’s core claim is that fat “quality” may matter as much as fat quantity. Imaging can now measure not only volume, but density, attenuation, texture, and radiomic signatures. Translation: CT and MRI are no longer just taking anatomical mugshots. They are becoming forensic accountants.

Recent AI work makes this feel less like sci-fi and more like a suspiciously well-funded Tuesday. Miller and colleagues showed that deep learning could measure epicardial fat volume and attenuation from low-dose ungated CT scans in under 2 seconds, compared with about 15 minutes for expert manual annotation. In 8,781 patients, elevated EAT volume and attenuation predicted death or myocardial infarction even after adjusting for factors like BMI, coronary artery calcium, and perfusion findings Miller et al., 2024.

Foldyna and colleagues took the same “don’t waste the CT scan” energy into lung cancer screening. In 24,090 heavy smokers followed for over 12 years, AI-derived EAT volume and density were independently associated with all-cause and cardiovascular mortality Foldyna et al., 2024. The lung scan came for cancer screening and accidentally brought cardiovascular receipts. Very normal. Nothing to see here.

Then there is “fat-omics,” which sounds like a startup that would serve cold brew at a cardiology conference. Hoori and colleagues extracted 148 EAT features from CT calcium-score scans and found that a 15-feature model predicted major adverse cardiovascular events better than simpler EAT volume, mean attenuation, or BMI measures Hoori et al., 2024. The hidden variables included local thickness patterns and high-Hounsfield-unit regions, which may reflect inflammatory changes. The fat was not just there. It had texture. Motive. Possibly a burner phone.

Why AI Shows Up at the Crime Scene

The machine learning angle matters because manual fat segmentation is slow, tedious, and not exactly what clinicians dream about while drinking bad hospital coffee. AI can turn existing CT or MRI scans into quantitative maps of fat depots, potentially adding risk information without new scans.

A 2025 review by Kandi and colleagues describes how AI is moving adipose imaging from manual measurement toward automated segmentation, radiomics, and multimodal prediction models that combine imaging with clinical and molecular data Kandi et al., 2025. That lines up with Khanna et al.’s big-picture thesis: cardiometabolic disease is not one organ misbehaving. It is a network. And networks are where machine learning likes to lurk, wearing sunglasses indoors.

The Therapeutic Angle, AKA Follow the GLP-1s

The review also points toward treatment. If adipose depots have distinct biology, therapies might change not only weight, but fat inflammation and depot behavior. In the STOP randomized trial, semaglutide was associated with reduced epicardial adipose tissue volume in people with type 2 diabetes Manubolu et al., 2024. That does not prove epicardial fat is the magic lever behind cardiovascular benefit, but it does make the conspiracy board more crowded.

The honest version: this field still needs prospective trials showing that measuring and modifying specific fat depots improves outcomes. AI models also need external validation across scanners, populations, ethnic groups, sex-specific biology, and clinical workflows. Otherwise we risk building a very elegant risk calculator that performs beautifully in one hospital and then panics when shown a different CT protocol.

The Takeaway Hidden in Plain Sight

Khanna and colleagues are asking cardiology to stop treating adipose tissue like a single blob with a bad reputation. Fat is regional, biological, inflammatory, measurable, and possibly more organized than several group projects I have survived.

If the unified adipose tissue model holds up, cardiometabolic risk assessment could become more personal: not just “how much fat,” but “which depot, what phenotype, what signal, and what downstream damage?” That is a better question. Also a more suspicious one. Naturally, I approve.

References

Khanna, S., Mann, G., Bhat, A., et al. “Role of systemic and epicardial adipose tissue in cardiometabolic disease.” Nature Reviews Cardiology (2026). DOI: 10.1038/s41569-026-01304-9

Miller, R. J. H., Shanbhag, A., Killekar, A., et al. “AI-derived epicardial fat measurements improve cardiovascular risk prediction from myocardial perfusion imaging.” npj Digital Medicine 7, 24 (2024). DOI: 10.1038/s41746-024-01020-z

Foldyna, B., Hadzic, I., Zeleznik, R., et al. “Deep learning analysis of epicardial adipose tissue to predict cardiovascular risk in heavy smokers.” Communications Medicine 4, 44 (2024). DOI: 10.1038/s43856-024-00475-1

Hoori, A., et al. “Artificial Intelligence Prediction of Cardiovascular Events Using Opportunistic Epicardial Adipose Tissue Assessments From Computed Tomography Calcium Score.” JACC: Advances (2024). PMID: 39372475, PMCID: PMC11450955, DOI: 10.1016/j.jacadv.2024.101188

Kandi, S. R., Khera, R., Rajagopalan, S., et al. “AI in Adipose Imaging: Revolutionizing Visceral Adipose Tissue, Ectopic Fat, and Cardiovascular Risk Assessment.” Current Atherosclerosis Reports 27, 101 (2025). DOI: 10.1007/s11883-025-01356-1

Manubolu, V. S., Lakshmanan, S., Kinninger, A., et al. “Effect of Semaglutide on Epicardial Adipose Tissue in Type 2 Diabetes.” Journal of the American College of Cardiology 84(9), 865-867 (2024). PMID: 39168573, DOI: 10.1016/j.jacc.2024.05.065

Disclaimer: This blog post is a simplified summary of published research for educational purposes. The accompanying illustration is artistic and does not depict actual model architectures, data, or experimental results. Always refer to the original paper for technical details.