EchoNext is an AI model capable of detecting structural heart disease from standard 12-lead ECGs. Trained on over 1.2 million ECG-echocardiogram pairs, it achieved an AUROC of 85.2%. The model was validated on external datasets and demonstrated a diagnostic accuracy of 77.3% in a blinded survey, surpassing cardiologists' average accuracy. EchoNext identified 3,444 high-risk patients in a silent deployment, achieving a positive predictive value of 74%. The system offers significant potential for improving the diagnosis of underdiagnosed conditions such as heart failure and valvular heart disease.
In a blinded survey of 150 ECGs, EchoNext achieved a diagnostic accuracy of 77.3%, surpassing the average cardiologist's 64% accuracy.
The model integrated both ECG waveform data and standard tabular ECG features, such as age, sex, and heart rate.
EchoNext identified 3,444 high-risk patients who had not yet received echocardiograms, achieving a 74% positive predictive value in the follow-up subset.
Trained on more than 1.2 million ECG-echocardiogram pairs, the deep learning system achieved an area under the ROC curve of 85.2%.
#ai-in-healthcare #structural-heart-disease #electrocardiograms #deep-learning #cardiology-innovations
Collection
[
|
...
]