An artificial intelligence model known as EchoNext became the first AI tool to receive Food and Drug Administration clearance to detect six indications of structural heart disease (SHD) using only data from standard 12-lead electrocardiograms (ECGs).
Developed by physicians and scientists in the Center for Cardiovascular and Radiologic Deep Learning (CRADLE) at NewYork-Presbyterian and Columbia, the tool leverages ECGs to identify patients who should go on to receive an echocardiogram to confirm the diagnosis. EchoNext is now available free to clinicians through OpenEvidence, the AI-based clinical decision platform.
“We have colonoscopies, we have mammograms, but we have not had equivalents for most forms of heart disease,” says Pierre Elias, M.D., medical director for artificial intelligence at NewYork-Presbyterian, who leads the CRADLE lab. “Now we are able to diagnose critical, high-risk conditions that the human eye can’t, and potentially deliver lifesaving treatment earlier.”
Dr. Pierre Elias led the team of researchers and clinicians at NewYork-Presbyterian and Columbia who developed EchoNext.
Demonstrating the Accuracy of EchoNext
EchoNext is a convolutional neural network model that was trained on over 700,000 ECG-echocardiogram pairs from 230,000 patients collected over 14 years. In a validation study published in Nature in 2025, Dr. Elias and his colleagues demonstrated EchoNext’s high accuracy in identifying a range of structural heart problems, including heart failure due to cardiomyopathy, valve disease, pulmonary hypertension, and hypertrophy. In a head-to-head comparison of 3,200 ECG reviews — half conducted by a cardiologist without AI assistance, and half with AI assistance — EchoNext accurately identified 77% of SHD, compared with 69% for cardiologists who used AI-assisted reviews, and 64% who did not use AI.
To evaluate real-world clinical utility, investigators prospectively deployed EchoNext among nearly 85,000 patients undergoing ECGs who had not previously received echocardiography. The algorithm identified approximately 9% of these patients as high risk for previously undiagnosed structural heart disease. Among the roughly half who went on to get their first echocardiogram, nearly three-quarters were ultimately diagnosed with SHD, roughly doubling the diagnostic yield compared with standard referral patterns.
Rather than serve as an end diagnosis, EchoNext augments clinical decision-making by functioning as an early detection and triage tool. Its integration into OpenEvidence allows clinicians to submit an ECG reading on the platform and receive a prediction for structural heart disease. By incorporating the model into routine workflows, Dr. Elias is hoping the tool will enable earlier diagnosis, more timely intervention, and improved allocation of imaging resources.
Case Study: AI Structural Heart Disease Detection Leads to Heart Transplantation
In June, Dr. Elias and his colleagues published a case study in Nature Medicine describing how EchoNext was able to detect SHD in a patient who ultimately went on to receive a heart transplant, the first peer-reviewed account of its kind.
A 45-year-old man presented in the emergency department at NewYork-Presbyterian Queens after four days of progressively worsening shortness of breath and a productive cough following recent exposure to wildfire smoke. His heart rate was elevated, his oxygen level normal, and he had reduced breath sounds in both lungs. A blood test showed a mildly increased marker of cardiac injury, but other laboratory tests were normal, and a chest X-ray showed no abnormalities. He was ultimately treated with a bronchodilator and steroids for asthma and discharged when symptoms improved.
Now we are able to diagnose critical, high-risk conditions that the human eye can’t, and potentially deliver lifesaving treatment earlier.
— Dr. Pierre Elias
A 12-lead ECG performed on the patient in the ED was run through EchoNext as part of the prospective SAGE (Structural Heart Disease Diagnosis using Artificial Intelligence in the Emergency Department) trial, which evaluates whether AI-guided ECG screening in the ED can improve SHD detection and management. The algorithm identified him as high risk, prompting a follow-up echocardiogram that revealed advanced cardiomyopathy.
Subsequent evaluation identified that the patient had a rare variant in the LMNA gene known to be associated with dilated cardiomyopathy and risk of sudden cardiac death. He initially underwent medical therapy for heart failure, but his condition continued to worsen, requiring placement of a temporary mechanical heart pump and, ultimately, a successful heart transplant at NewYork-Presbyterian and Weill Cornell Medicine — only six months after his initial SHD diagnosis.
“It was a chain of people, decisions, and timing that ultimately helped this patient survive,” says David Majure, M.D., medical director of the Heart Transplant Service at NewYork-Presbyterian and Weill Cornell Medicine. “EchoNext was one important link. It helped us recognize the problem earlier and act faster. When you equip teams with better tools, patients have better chances to have the best outcomes.”
Dr. Elias says this is a prime example of how EchoNext can aid physicians in clinical decision-making. “We sought to develop, validate, and deploy AI technologies that would meaningfully change the way we take care of patients,” says Dr. Elias. “We’re now at a point where we can see that impact and meet patients who are benefiting from these technologies every day.”
Dr. Elias is a founder of Pathway Labs, a company that builds AI to help clinicians catch serious heart disease earlier, starting with EchoNext. Pathway Labs recently announced an $8.5 million seed financing round to support expansion across health systems, further clinical validation, and continued research and development. NewYork-Presbyterian provided funding as part of this round.
