James, 72, felt tired for months. His doctor said it was just aging. Then came the shortness of breath. By the time he was diagnosed with a rare heart condition, his heart was already stiff and weak. His story is far too common.
This condition, called transthyretin amyloid cardiomyopathy (ATTR-CM), happens when a protein called transthyretin builds up in the heart. It makes the heart muscle stiff, so it can’t fill with blood properly. Over time, it leads to heart failure.
It’s not rare. It affects up to 1 in 10 older adults with heart failure. But most cases go undiagnosed for years.
Why? Because the early signs are vague—fatigue, swelling, trouble breathing. They’re easy to mistake for normal aging or other heart problems.
And the tests that can confirm it—special scans or biopsies—are hard to get, especially in rural or underserved areas.
But here’s the twist: most people with ATTR-CM already had an ECG, sometimes years before diagnosis.
An ECG is a simple, painless test that records the heart’s electrical activity. It’s done in clinics, hospitals, even during routine checkups. Millions are performed every year.
Yet doctors couldn’t use them to spot ATTR-CM—until now.
The AI that sees what doctors miss
Researchers trained an artificial intelligence model to study old ECG images. Not raw data—actual pictures of 12-lead ECGs, the kind printed on paper or viewed on a screen.
The AI learned to spot tiny, invisible patterns linked to ATTR-CM. Think of it like recognizing a face in a crowd by the shape of the ears or the curve of the jaw—details most of us would overlook.
It’s not looking for classic heart attack signs. It’s spotting a silent buildup of protein that changes how the heart’s electrical signals travel.
Imagine a highway where traffic flows smoothly. Now picture ice forming under the pavement. At first, cars still move. But over time, cracks appear, lanes close, and traffic slows. The AI sees those early cracks—before the road shuts down.
This model scored an AUROC of 0.87, a strong result meaning it can tell who has the disease with high accuracy.
And it worked just as well in Black and Hispanic patients—groups at higher risk but often left behind in heart research.
Works across real-world settings
The team tested the AI on data from eight countries across the US and Europe. It performed consistently, even in places with very few cases.
That’s rare. Many AI tools fail when moved from one hospital to another. This one held up.
In three real-world screening groups, the AI flagged people at high risk. These included older adults with heart failure and those who’d had carpal tunnel surgery—a known early clue of ATTR-CM.
The AI didn’t replace diagnosis. It pointed doctors to who should get a confirmatory scan.
This doesn't mean this treatment is available yet.
But there’s a catch.
The AI is not in hospitals today. It hasn’t been approved by regulators. And while it finds risk, it can’t confirm disease.
Patients still need a special scan or blood test to get a final diagnosis.
Also, the study used past data. The real test comes when the AI runs live in clinics, reading ECGs as they come in.
Experts say this is a major step—but not the final one.
“The beauty is using a test we already do,” said one cardiologist not involved in the study. “If this works in real time, it could close a huge gap in care.”
Who could benefit most
Older Black men have the highest risk of ATTR-CM due to a common genetic variant. Yet they’re less likely to be referred for advanced testing.
This AI could help level the field. Since ECGs are common and low-cost, the tool could be used in community clinics, not just big medical centers.
Early treatment can slow the disease. New drugs help clear the protein buildup. But they work best when started early—before the heart is too damaged.
That’s why timing matters.
The road ahead includes clinical trials where the AI guides real-time referrals. If those go well, it could be part of routine care within a few years.
For now, the message is hope—not action.
Talk to your doctor if you have heart failure, carpal tunnel, or a family history. Ask if ATTR-CM could be a cause.
The AI isn’t ready. But the conversation can start today.
The next step is testing the tool in live clinics. Researchers plan trials in diverse health systems to see how well it works in daily practice. Results could come in the next two to three years.