It detects diseases that we doctors cannot see
En los laboratorios de la Universidad de Columbia, investigadores han enseñado a una máquina a escuchar lo que el corazón susurra antes de gritar. EchoNext, un sistema de inteligencia artificial entrenado con más de un millón de registros cardíacos, detecta enfermedades estructurales del corazón a partir de un simple electrocardiograma con una precisión que supera a la de los cardiólogos experimentados. En una época en que muchas enfermedades cardíacas avanzan en silencio hasta desencadenar un infarto o un ictus, esta herramienta plantea una pregunta tan técnica como profundamente humana: ¿cuántos desastres pueden prevenirse si aprendemos a leer las señales antes de que sea demasiado tarde?
- La enfermedad cardíaca estructural puede progresar sin síntomas durante años, convirtiendo el primer aviso en una catástrofe: un infarto, un ictus, un corazón que ya no puede repararse fácilmente.
- EchoNext identificó correctamente al 77% de los pacientes con enfermedad cardíaca estructural, frente al 64% logrado por trece cardiólogos experimentados enfrentados a los mismos datos.
- El sistema fue entrenado con 1,2 millones de pares de electrocardiogramas y ecocardiogramas de 230.000 personas, aprendiendo a detectar patrones eléctricos invisibles al ojo clínico humano.
- En un estudio de 85.000 personas, EchoNext marcó a más de 7.500 individuos como de alto riesgo, pacientes que de otro modo habrían seguido su vida ignorando la amenaza silenciosa en su pecho.
- El verdadero desafío ahora no es tecnológico sino institucional: si hospitales, aseguradoras y sistemas sanitarios adoptarán esta herramienta antes de que más pacientes lleguen demasiado tarde.
Investigadores de la Universidad de Columbia han desarrollado EchoNext, un sistema de inteligencia artificial capaz de detectar enfermedades estructurales del corazón a partir de un electrocardiograma convencional, con una precisión superior a la de los cardiólogos. La enfermedad cardíaca estructural —aquella que compromete la capacidad del corazón para bombear sangre— avanza con frecuencia sin síntomas hasta que ocurre algo grave: un infarto, un ictus. Si se detecta a tiempo, muchas de estas condiciones pueden tratarse con procedimientos mínimamente invasivos. El problema siempre ha sido saber quién necesita una prueba más costosa, como el ecocardiograma, y quién no.
EchoNext responde precisamente a esa pregunta. Entrenado con más de 1,2 millones de pares de electrocardiogramas y ecocardiogramas de 230.000 personas, el sistema aprendió a reconocer en los datos eléctricos del corazón los patrones que anticipan problemas estructurales visibles solo en ultrasonido. Al ser evaluado con 3.200 electrocardiogramas nuevos, identificó correctamente al 77% de los pacientes afectados. Trece cardiólogos experimentados, enfrentados a la misma tarea, alcanzaron el 64%.
Pierre Elias, profesor de medicina e informática biomédica en Columbia, describe EchoNext como el primer modelo de IA capaz de detectar todas las formas de enfermedad cardíaca estructural a partir únicamente de un electrocardiograma. "Básicamente, EchoNext usa la prueba más barata para determinar quién necesita la más cara", explica. En un estudio de 85.000 personas, el sistema identificó a más de 7.500 individuos como de alto riesgo, pacientes que de otro modo habrían permanecido sin diagnóstico.
Lo que hace posible este avance es la escala: ningún cardiólogo puede retener en su mente un millón de casos, pero una máquina sí puede encontrar los susurros eléctricos que preceden al colapso. La pregunta que queda abierta es si el sistema sanitario —hospitales, aseguradoras, clínicas— estará dispuesto a adoptarlo antes de que más corazones fallen en silencio.
Researchers at Columbia University have developed an artificial intelligence system that can spot hidden heart disease from a simple electrical reading of the heart—and it does so more reliably than experienced cardiologists.
The system, called EchoNext, was built to solve a practical problem in medicine. Structural heart disease—conditions that impair the heart's ability to pump blood through the body—often progresses silently, without obvious symptoms, until something catastrophic happens: a heart attack, a stroke. By then, the damage is done. Traditionally these conditions required open surgery, but many can now be treated with minimally invasive procedures if caught early enough. The challenge is catching them at all.
A standard electrocardiogram, or ECG, is cheap and painless. It records the electrical activity of the heart with each beat, producing a pattern of waves that show the impulses moving through the heart's chambers. But an ECG alone cannot definitively diagnose structural heart disease. For that, doctors need an echocardiogram—an ultrasound of the heart—which is more expensive and time-consuming. The question became: how do you know which patients actually need that more costly test?
EchoNext answers that question. Researchers at Columbia and New York Presbyterian Hospital trained the AI system on more than 1.2 million pairs of ECGs and echocardiograms from 230,000 individuals. The system learned to recognize patterns in the electrical data that correlate with structural problems visible only on ultrasound. When tested on 3,200 new ECGs, EchoNext identified 77 percent of patients with structural heart disease. Thirteen experienced cardiologists given the same task achieved 64 percent accuracy. The machine outperformed the doctors.
Pierre Elias, a professor of medicine and biomedical informatics at Columbia's medical school, explains that EchoNext represents something new: the first AI model to detect all forms of structural heart disease from electrocardiograms alone. Other AI systems have been developed to spot coronary disease in imaging or to help radiologists read mammograms more accurately, but this is different in scope. "Basically, EchoNext uses the cheapest test to figure out who needs the more expensive one," Elias says. "It detects diseases that we doctors cannot see in the electrocardiogram."
The practical implications are significant. In a study of 85,000 people, EchoNext flagged more than 7,500 individuals as high-risk for undiagnosed structural heart disease. Those are people who might otherwise have walked around unaware, waiting for a crisis. The combination of a routine ECG and artificial intelligence could reshape how hospitals and clinics screen for these conditions—a new paradigm, as Elias describes it, for identifying patients who need intervention before their hearts fail them.
What makes this work is scale and pattern recognition. No human cardiologist can hold 1.2 million cases in their mind. A machine can. It can find the subtle electrical signatures that predict structural problems, the whispers in the data that precede the shout of a heart attack. The question now is whether hospitals will adopt it, whether insurance will cover it, whether the system can be trusted at scale. For patients with undiagnosed heart disease, the answer matters.
Citações Notáveis
EchoNext uses the cheapest test to figure out who needs the more expensive one. It detects diseases that we doctors cannot see in the electrocardiogram.— Pierre Elias, professor of medicine and biomedical informatics, Columbia University
A Conversa do Hearth Outra perspectiva sobre a história
Why does structural heart disease stay hidden for so long?
Because it doesn't announce itself. Your heart can be structurally broken—a valve not closing right, a wall too thin—and you'll feel fine until suddenly you won't. By then you're in an ambulance.
And the ECG can't catch it?
Not reliably. An ECG shows electrical activity, not physical structure. It's like listening to a song and trying to guess if the instrument is broken. You might hear something wrong, but you might not.
So EchoNext is basically a translator?
Exactly. It translates electrical noise into structural diagnosis. It learned from millions of examples what the electrical patterns look like when the physical heart is damaged.
Why does it beat cardiologists?
Because cardiologists are pattern-matching machines too, but limited ones. They've seen thousands of cases. EchoNext has seen millions. And it doesn't get tired, doesn't miss the subtle ones.
What happens to the 7,500 people it flagged as high-risk?
That's the real test now. They should get echocardiograms. Some will have disease that needs treatment. Some might have nothing. But at least they'll know. And they won't be waiting for a heart attack to find out.
Is this the future of diagnosis?
It's part of it. Not replacing doctors—augmenting them. Doing the grunt work of screening so doctors can focus on the people who actually need their judgment.