The information is there. We just don't routinely measure it.
For generations, medicine has reduced the complexity of the human body to a single ratio of weight and height — a number that says nothing of what lies beneath the skin. Now, a team of researchers from Germany and the United Kingdom has used artificial intelligence to analyze over 66,000 MRI scans, producing the first detailed reference maps of how fat and muscle distribute across the body and what those patterns foretell about disease and death. The work, published in Radiology, does not merely refine an old tool — it quietly retires one, suggesting that the quality of what we carry inside us matters as much as how much we carry.
- BMI has long served as medicine's shorthand for bodily risk, but it cannot tell a muscular body from an overweight one, nor can it see the fat quietly infiltrating muscle tissue — a gap that has left countless patients misread.
- The new AI-driven analysis reveals that intramuscular fat raises the risk of major cardiovascular events by 54 percent and that low skeletal muscle mass predicts a 44 percent higher risk of death from any cause — findings that reorder the hierarchy of what clinicians should be watching.
- Researchers built and released an open-source deep learning tool that extracts five key body composition metrics from routine CT and MRI scans already being performed, meaning the data has always been present — medicine simply lacked the means to read it.
- The tool is already being aimed at cancer treatment monitoring and at patients on GLP-1 weight-loss drugs, where distinguishing healthy fat loss from dangerous muscle loss could meaningfully change clinical decisions.
- Validation studies in specific clinical populations are underway, with the field now oriented not toward building new imaging technology, but toward finally learning to interpret the images it has been collecting for years.
For decades, the body mass index — a simple ratio of height and weight — has been medicine's primary lens for assessing bodily risk. It is a blunt instrument. It cannot distinguish muscle from fat, nor can it reveal what is happening deep within tissue. A research team spanning Germany and the United Kingdom has now built something far more precise: using AI to analyze whole-body MRI scans from more than 66,000 people, they have produced the first detailed reference maps of how fat and muscle distribute across the human body, and what those distributions predict about diabetes, heart disease, and death.
The findings, published in Radiology, complicate a long-held assumption. Visceral fat — the kind that wraps around organs — does raise diabetes risk significantly, by more than twofold. But muscle quality emerged as an equally powerful signal. Low skeletal muscle mass predicted a 44 percent higher risk of death from any cause. Intramuscular fat, the kind that infiltrates muscle tissue itself, was associated with a 54 percent increased risk of major cardiovascular events. These are not subtle gradations. They are the kind of numbers that reframe how medicine understands the body.
Led by Jakob Weiss and Matthias Jung at the University Medical Center Freiburg, the team drew from two major biobanks — the UK Biobank and the German National Cohort — analyzing scans taken between 2014 and 2022. Their deep learning algorithm measured five metrics per scan: subcutaneous fat, visceral fat, skeletal muscle volume, muscle fat content, and intramuscular fat. Each was converted into a z-score, allowing fair comparison across age, sex, and height. A 70-year-old woman could finally be measured against her true peers.
What makes the work immediately translatable is its accessibility. The researchers released an open-source web calculator that extracts body composition data from routine CT or MRI scans already being performed for other clinical reasons. The information, as Weiss noted, has always been there — embedded in images radiologists review every day. AI now makes it possible to surface and standardize that hidden layer of data without any new equipment or dedicated scans.
The applications extend well beyond general risk assessment. In oncology, the tool could help predict how patients will tolerate treatment. For those taking GLP-1 weight-loss drugs, it could distinguish beneficial fat loss from the muscle loss that sometimes accompanies rapid weight reduction. Validation studies in specific clinical populations are already underway. The next step in medicine's understanding of the body may not require new technology at all — only a more careful reading of what has long been in plain sight.
For decades, doctors have relied on a single number to assess whether a patient's body poses a health risk: body mass index, a calculation based only on height and weight. It is crude, imprecise, and tells almost nothing about what is actually happening inside. A team of researchers working across Germany and the United Kingdom has now built something far more granular. Using artificial intelligence to analyze whole-body MRI scans from more than 66,000 people, they have created the first detailed reference map of how fat and muscle distribute themselves across the human body—and what those patterns predict about who will develop diabetes, suffer a heart attack, or die.
The work, published in Radiology, a journal of the Radiological Society of North America, upends a comfortable assumption: that visceral fat—the dangerous kind that wraps around organs—is the primary culprit in cardiometabolic disease. The data tells a more complicated story. Yes, high visceral fat increases the risk of diabetes by more than twofold. But the quality of muscle matters just as much, sometimes more. People with low skeletal muscle faced a 44 percent higher risk of death from any cause, even after accounting for other cardiovascular risk factors. Intramuscular fat—fat that infiltrates the muscle tissue itself—predicted major heart events with a 54 percent increased risk. These are not marginal differences. They are the kind of findings that reshape how medicine thinks about the body.
The researchers, led by Jakob Weiss and Matthias Jung at the University Medical Center Freiburg, faced a practical problem. Clinicians still rely on BMI and waist circumference because they are easy to measure. No special equipment required. But BMI is a blunt instrument. It cannot distinguish between a person who is muscular and a person who is overweight. It ignores the fact that body composition changes dramatically with age and differs significantly between men and women. The medical field lacked reference standards—a baseline understanding of what normal looks like across different populations. Without that, it was impossible to know whether an individual's body composition put them at genuine risk.
The team drew their data from two large biobanks: the UK Biobank and the German National Cohort. The participants, averaging 57.7 years old, underwent whole-body MRI scans between 2014 and 2022. The researchers then deployed a deep learning algorithm—an open-source framework they built themselves—to automatically measure five key metrics from each scan: subcutaneous fat (the kind under the skin), visceral fat, skeletal muscle volume, the fat content within muscle, and intramuscular fat. Each measurement was then converted into a z-score, a statistical tool that shows how far an individual deviated from the norm for their age, sex, and height. This normalization was crucial. It meant that a 70-year-old woman could be compared fairly to her age-matched peers, not to a 30-year-old man.
The findings challenged conventional wisdom in ways both subtle and profound. A person with high visceral fat faced a 2.26-fold increased risk of developing diabetes. But someone with high intramuscular fat—something BMI cannot detect—faced a 1.54-fold increased risk of major cardiovascular events. And skeletal muscle emerged as a protective factor. Low muscle mass predicted mortality independent of other risk factors. As Jung explained, it is not simply the quantity of muscle that matters. It is the quality. Intramuscular fat acts as a window into muscle health that traditional tools like BMI, bioelectrical impedance analysis, or DEXA scans cannot easily provide.
What makes this work immediately practical is that the researchers did not require dedicated whole-body MRI scans. They released an open-source web calculator that allows clinicians to extract body composition data from routine CT or MRI scans already being performed for other reasons. A chest scan, an abdominal scan—the information is already there, embedded in images that radiologists review every day. AI now makes it possible to quantify that hidden layer of data in a reproducible, standardized way. Weiss put it plainly: "We're already imaging patients every day. The information is there. We just don't routinely measure or report it."
The implications ripple outward. In oncology, the tool could help predict treatment toxicity and survival. In patients taking weight-loss drugs like GLP-1 agonists, it could distinguish between healthy fat loss and the unwanted muscle loss that sometimes accompanies rapid weight reduction. The researchers are already planning validation studies in clinical populations, with an eye toward developing disease-specific reference values for different patient groups. The next frontier is not new imaging technology. It is learning to read what we are already looking at.
Citações Notáveis
BMI does not reliably reflect a person's actual body composition.— Dr. Jakob Weiss, University Medical Center Freiburg
It's not only how much muscle you have, but also the quality of that muscle. Intramuscular fat gives us a window into muscle quality that other methods can't easily provide.— Dr. Matthias Jung, University Medical Center Freiburg
A Conversa do Hearth Outra perspectiva sobre a história
Why does muscle quality matter more than we thought?
Because intramuscular fat—fat that infiltrates the muscle tissue itself—is a marker of metabolic dysfunction that BMI completely misses. A person can look fit by traditional measures and still have compromised muscle quality.
So BMI has been misleading us this whole time?
Not misleading exactly, but incomplete. BMI tells you weight relative to height. It says nothing about what that weight actually is—whether it's muscle, bone, or fat, or how that fat is distributed. Two people with identical BMI can have completely different health trajectories.
How does the AI actually measure this from an MRI scan?
It uses deep learning to automatically segment the scan into different tissue types—fat under the skin, fat around organs, muscle, fat within muscle. Then it normalizes those measurements against age, sex, and height to create a z-score. That score tells you whether you're above or below average for your demographic.
And clinicians can use this on scans they're already taking?
Exactly. If someone comes in for a routine chest or abdominal CT, the algorithm can extract body composition data without any additional imaging. The information was always there. Medicine just wasn't systematically measuring it.
What happens next? Does this change how doctors assess risk?
That's the validation phase. The researchers need to test these reference curves in actual clinical populations—cancer patients, people on weight-loss drugs, others with specific conditions. If it holds up, it could reshape risk stratification across multiple specialties.