An AI Rereads Your Chest-Pain CT Scan
The Hidden Second Reading
Here is the setup. If you show up at a hospital with chest pain, doctors already order a coronary CT angiogram, a scan that images the arteries feeding your heart. A University of Edinburgh study published in Radiology on June 30, 2026 took 1,722 of those scans, from patients mostly in their 50s, and asked a simple question: what else is sitting in that image that nobody has ever bothered to look at? Senior author Professor Michelle Williams and her team, part-funded by the British Heart Foundation, pointed an AI at the muscle around the ribs, the chest, back, pectoral and intercostal muscles, tissue that was captured in every one of those scans but that no radiologist had ever scored. The AI was not measuring how big that muscle is. It was measuring muscle attenuation, a reading of how dense and how fatty the muscle is. And it did it in under a minute per scan, a job that takes a radiologist several hours by hand. That speed is the whole point: it is the difference between a measurement nobody could afford to take and one every scan could produce automatically, for free, as a byproduct. Here is what that measurement bought. A 10-point rise in the muscle score tracked with a 31 percent lower chance of a heart attack and a 39 percent lower risk of death within ten years. And the link held up even after the researchers adjusted for age, sex and coronary calcium, the usual suspects that might otherwise explain away a result like this. Muscle size told the model nothing. Composition did. Nobody had scored that tissue before Williams pointed software at it.
What The Money Is Buying
The Edinburgh result landed the same month as two other signals worth putting side by side with it. The FDA cleared an 'AI-CVD' platform for opportunistic cardiothoracic screening. 'Opportunistic' is the key word: the software scans a CT image taken for some completely unrelated reason, say a lung check, and flags heart risk as a bonus finding nobody asked for. And a2z Radiology, backed by Khosla Ventures and SeaX, raised a $5 million seed round on the strength of its first FDA clearance, for opportunistic abdomen-pelvis CT screening. Notice what none of these are: a new diagnostic test. They are all software rereading old film for a new signal. No new scanner, no new radiation dose, no new appointment. Just a second pass on the imaging exhaust that hospitals already produce by the million. That is also why the money is moving faster than the evidence: a company can sell a risk score without ever touching a patient or a machine. The FDA cleared the AI-CVD platform this month. a2z Radiology closed its $5 million seed round on the same opportunistic-screening premise. And Edinburgh's own authors are the first to say the muscle-heart link still needs testing in people who were never scanned at all.
So what happens next? The real test is a follow-up study in an unscreened population, people who were never worried enough to get a scan in the first place, to see if the muscle signal holds outside patients who already had symptoms. Until that trial runs, the score stays a research finding, not something your doctor can act on. The frustrating part is that every hospital with a CT archive already has the raw data sitting there. What is missing is not more scans. It is that one study.