Log in to MyACC
ECG-Based Deep Learning Improves Outcome Prediction After CRT
Quick Takes
- Developed with the use of artificial intelligence, FactorECG may predict the combined clinical endpoint of death, heart transplant, or left ventricular assist device implantation, significantly outperforming QRSAREA and guideline ECG criteria.
- FactorECG identified inferolateral T-wave inversion, smaller right precordial S- and T-wave amplitude, ventricular rate, and increased PR interval and P-wave duration to be important predictors for poor outcome.
Study Questions:
Can an explainable deep learning-based model (FactorECG) predict hard outcomes and echocardiographic response after cardiac resynchronization therapy (CRT) using only a standard electrocardiogram (ECG)?
Methods:
A deep learning algorithm, trained on 1.1 million ECGs, was used to compress the median beat ECG, thereby summarizing most ECG features into 21 explainable factors (FactorECG). Pre-implantation ECGs of 1,306 CRT patients were converted into their respective FactorECG. These were assessed for their ability to predict the combined clinical endpoint of death, left ventricular assist device, or heart transplantation, and compared with QRSAREA and guideline ECG criteria.
Results:
FactorECG predicted the combined clinical endpoint (c-statistic 0.69), significantly outperforming QRSAREA and guideline ECG criteria (c-statistic 0.61). The addition of 13 clinical variables was of limited added value for the FactorECG model when compared with QRSAREA. FactorECG identified inferolateral T-wave inversion, smaller right precordial S- and T-wave amplitude, ventricular rate, and increased PR interval and P-wave duration to be important predictors for poor outcome.
Conclusions:
FactorECG is superior for the prediction of clinical outcome when compared with guideline criteria and QRSAREA.
Perspective:
Despite CRT having been available for over two decades, there remain challenges about identifying patients most likely to benefit. The current standard of care relies on QRS duration and some morphology characteristics. A variety of left ventricular electrical activation patterns are concealed in the ECG. Some clinical variables and QRSAREA have shown additional predictive ability. A prior artificial intelligence ECG model predicted future atrial fibrillation from a sinus rhythm ECG to identify patients at elevated stroke risk. The authors of the present study used deep learning algorithms to identify “explainable” (visualizable) ECG abnormalities that are associated with poor response after CRT. The analysis suggests that using FactorECG may identify a large group of patients with guideline Class I and IIa indications for CRT, who did not improve with CRT. Of note, the authors made the algorithm available at https://crt.ecgx.ai/ where users may for research purposes upload files from GE MUSE ECG system. It is highly likely that this and other deep learning-based models will continue to evolve and provide clinicians with additional insights, allowing for more targeted therapies for the right patients.
Clinical Topics: Arrhythmias and Clinical EP, Cardiac Surgery, Heart Failure and Cardiomyopathies, Invasive Cardiovascular Angiography and Intervention, Noninvasive Imaging, Implantable Devices, SCD/Ventricular Arrhythmias, Atrial Fibrillation/Supraventricular Arrhythmias, Cardiac Surgery and Arrhythmias, Cardiac Surgery and Heart Failure, Acute Heart Failure, Heart Transplant, Mechanical Circulatory Support, Interventions and Imaging, Echocardiography/Ultrasound
Keywords: Arrhythmias, Cardiac, Artificial Intelligence, Cardiac Resynchronization Therapy, Echocardiography, Electrocardiography, Heart-Assist Devices, Heart Failure, Heart Transplantation, Intelligence, Machine Learning, Outcome Assessment, Health Care
< Back to Listings
