From the moon to the grocery store.
I may be biased but it is amazing how much information you can glean from an Electrocardiogram (ECG) which looks like squiggles on a piece of paper to many physicians.
An ECG is the electrical signature of the heart. Below is an EKG strip from Neil Armstrong’s moon landing taken in 1969. Until now, we’ve looked at ECG tracings almost the same as we did then with our eyes but that is starting to change.
A recent paper from the SHOPS-AF study demonstrated a surprisingly accessible method for diagnosing Atrial Fibrillation (AF). Instead of on the Moon, ECGs were taken on a shopping cart. A single-lead sensor was built into the handle (like at the Gym) and as people pushed their buggies through a supermarket, the device quietly recorded tracings and flagged atrial fibrillation.
The research team used unsupervised AI, specifically called Generative Topographic Mapping which means it wasn’t told what AF looks like. It learned patterns on its own. It clustered beats by hidden properties we can’t see. (SEE BELOW IF INTERESTED)
It saw things NO Cardiologist reads when we glance at a strip.
This is where the machines RISE above us. AI isn’t replacing our judgment, it’s revealing signals we never learned to read. The same old strip from 1969 but with new AI eyes.
It can call Sleep apnea without a Sleep Study.
It can pick up signs of weak heart or Early LV dysfunction long before an echo would ever be ordered.
It can flag patterns linked to Pulmonary hypertension, Elevated Potassium, Valve problems like Moderate aortic stenosis, and even early Cardiac Amyloidosis.
Some studies show it can spot risk for future Coronary artery disease, Diabetes or even Anemia.
The machines are rising above us and fundamentally expanding the visual field of medicine, uncovering new truths we have been looking at all along.
The future is bright with the convergence of wearables and Machine Learning (ML) technologies the delivery of healthcare will move from in the office EKGs to in the future ECG-AI on your wearable (AliveCor EKG, Apple watches, WHOOP straps, and Everbeat rings). So the possibility of diagnosing sleep apnea, heart failure or elevated blood pressure in your lungs sooner will drastically improve .
Cardiologists must not Rage Against the Machine; rather, we should embrace the new capabilities of the machines. The machines will soon reveal insights that we never before imagined were present.
Generative Topographic Mapping (GTM), an unsupervised model that pulls dozens of ECG and HRV features into a two-dimensional map so similar rhythm patterns cluster together in an interpretable way. Think of a playground at recess. All the kids run out, nobody tells them where to go, but kids with similar interests drift toward each other. Soccer kids on one side, cheerleaders in another spot, basketball players in their own corner. GTM is like watching that playground from above. You don’t tell anyone where to stand. You just see where they gather and draw circles around the groups. In this study, the ‘kids’ are ECG snippets. Their ‘interests’ are the tiny heartbeat patterns we can’t see. And the ‘playground’ is a two dimensional map. The AF snippets ended up clustering in their own area without the model ever being told what AF is.
References
AI-driven clustering and visualization of electrocardiogram signals to enhance screening for atrial fibrillation: The supermarket/hypermarket opportunistic screening for atrial fibrillation study Bellfield, Ryan A.A. et al. Heart Rhythm O2, Volume 6, Issue 10, 1601 - 1612