The silent protein clogging hearts
ATTR cardiomyopathy (amyloid protein buildup in the heart) used to be considered rare.
Today, doctors realize it's much more common than once thought, especially in older adults. New treatments can slow it down — but only if changes in the heart are spotted early and accurately.
Tracking those changes is harder than it sounds.
Echocardiography (heart ultrasound) is the workhorse test for monitoring ATTR-CM.
But echoes have a well-known weakness: they depend heavily on the person doing them. Two cardiologists reading the same patient's echoes may get slightly different numbers — and those small differences can obscure real disease progression.
As new ATTR therapies become available, knowing whether the heart is getting worse — or holding steady — matters more than ever.
The old way vs. the new question
Traditionally, serial echoes have been measured by human readers who manually trace chambers, walls, and motion on the screen.
That works, but variation is inevitable — especially between a senior cardiologist and a less-experienced reader.
What's new: fully automated AI algorithms can now perform many of these measurements without human input. The question this study asked is whether AI is reliable enough to track patients over time, alongside — or instead of — human readers.
How it works, in simple terms
Picture an echocardiogram as a moving blueprint of the heart.
A cardiologist clicks through frames, marks the walls and chambers, and calculates measurements like ventricular wall thickness and chamber volume. It's careful work, but mouse clicks and judgment calls vary.
An AI algorithm, trained on thousands of echoes, does the same job almost instantly. It always "clicks" the same way given the same image — so in theory, it should be highly consistent.
Think of it like two bakers measuring flour by eye versus a kitchen scale — the scale may not be perfectly calibrated, but it's always the same scale every time.
The study at a glance
Researchers retrospectively analyzed 178 serial annual echocardiograms from 62 patients with ATTR-CM.
Each echo was evaluated four ways: by a reference (expert) cardiologist, by a second cardiologist, by a novice reader, and by an automated AI system (Us2.ai).
They used Bland-Altman analysis and intraclass correlation coefficients (ICCs) to judge agreement between readers and measured each reader's repeatability when re-reading images blindly.
The picture was mixed — but encouraging for AI in an important way.
AI showed moderate agreement with the reference cardiologist for two key measurements: interventricular septal thickness (IVSd) and left ventricular end-diastolic volume (LVEDV). The ICCs were 0.65 and 0.51.
Cardiologist-to-cardiologist agreement was better (ICC 0.79 and 0.84) with very small systematic differences.
But here's the more interesting result: AI's repeatability — how consistent it was when re-measuring the same images — was comparable to experienced cardiologists. The novice reader, by comparison, was noticeably less consistent.
This means AI isn't replacing the expert's eye, but it's already more reliable than a less-experienced one.
A subtle but important point
For tracking disease progression, repeatability may matter even more than absolute accuracy.
If an AI systematically undershoots a measurement by a small amount — but does so consistently every single time — it can still reliably flag real changes over time.
That's the key takeaway for longitudinal monitoring: detecting change is what matters, and AI did that well.
The authors suggest AI-based echo analysis could support long-term monitoring of ATTR-CM, especially in centers that lack a dedicated expert reader for every patient.
It won't replace experienced cardiologists anytime soon. But it could ease workload, standardize measurements across visits, and offer a reliable second opinion — particularly valuable when an echo is read by someone less experienced.
More broadly, this study fits the trend of AI tools earning their place as assistants rather than replacements in medical imaging.
If you have ATTR cardiomyopathy and your echoes are being reviewed by a dedicated cardiac team, you're likely in good hands already.
At centers where AI-assisted readings are available, those tools may add consistency from one annual echo to the next — helping your care team detect subtle progression sooner.
You don't need to ask for AI analysis specifically. But knowing it exists may help explain why your reports are sometimes accompanied by automated measurements alongside the cardiologist's.
Honest limitations
This was a single-center retrospective study of 62 patients with one AI system.
Results might not apply to other AI algorithms or other hospital settings. The "reference" in any agreement study is itself imperfect — so saying AI agrees "moderately" with an expert doesn't necessarily mean AI is wrong. Sample sizes for some sub-analyses were small.
And echocardiography is just one way to monitor ATTR-CM; imaging like cardiac MRI and blood biomarkers also play roles.
Expect AI-assisted echocardiography to expand rapidly over the next few years.
Larger multi-center studies will help pin down exactly how much variation AI tools add — and reduce — compared to human readers. Direct comparisons across AI platforms will also matter.
For ATTR-CM specifically, this is good news: as treatments evolve, reliable tracking is critical, and AI may help make that tracking more consistent across clinics of different sizes and expertise levels.
