Electrocardiogram-based deep learning improves outcome prediction following cardiac resynchronization therapy

Abstract

Aims This study aims to identify and visualize electrocardiogram (ECG) features using an explainable deep learning–based algorithm to predict cardiac resynchronization therapy (CRT) outcome. Its performance is compared with current guideline ECG criteria and QRSAREA. Methods and results A deep learning algorithm, trained on 1.1 million ECGs from 251 473 patients, was used to compress the median beat ECG, thereby summarizing most ECG features into only 21 explainable factors (FactorECG). Pre-implantation ECGs of 1306 CRT patients from three academic centres were converted into their respective FactorECG. FactorECG predicted the combined clinical endpoint of death, left ventricular assist device, or heart transplantation [c-statistic 0.69, 95% confidence interval (CI) 0.66–0.72], significantly outperforming QRSAREA and guideline ECG criteria [c-statistic 0.61 (95% CI 0.58–0.64) and 0.57 (95% CI 0.54–0.60), P < 0.001 for both]. The addition of 13 clinical variables was of limited added value for the FactorECG model when compared with QRSAREA (Δ c-statistic 0.03 vs. 0.10). 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. An online visualization tool was created to provide interactive visualizations (https://crt.ecgx.ai). Conclusion Requiring only a standard 12-lead ECG, FactorECG held superior discriminative ability for the prediction of clinical outcome when compared with guideline criteria and QRSAREA, without requiring additional clinical variables. End-to-end automated visualization of ECG features allows for an explainable algorithm, which may facilitate rapid uptake of this personalized decision-making tool in CRT.

Dynamic prediction of malignant ventricular arrhythmias using neural networks in patients with an implantable cardioverter-defibrillator

“…The predictive capacity of latent space variables for arrhythmia prediction reportedly outperforms standard ECG criteria and clinical interpretations, which indicates the ability of autoencoders to map unique prognostic ECG features to a latent representation not reflected by traditional ECG parameters. 17 , 18 , 20 , 21 The framework used in this study, which integrates an unsupervised learning branch (the autoencoder) and a supervised learning branch (the dynamic ML model), is related to the concept of digital twins , which create virtual replicas of individual patients by continuously integrating and analysing their real-time physiological data. 50 , 51 Despite the potential of deep learning networks to extract features from high dimensional data and aid clinical decision-making, the interpretability and explainability of neural networks remains a significant challenge and an active field of research.…”

“…Prior studies have demonstrated encoder-decoder architecture neural networks to be able to extract physiologically-relevant ECG features in a latent feature space. 17 , 18 , 19 , 20 , 21 Considering that this latent space provides a comprehensive representation of the underlying ECG brought back to a pre-defined number of variables, it can be subsequently used for classification and regressions tasks. 22 In this study, we examine the potential of accommodating a supervised dynamic ML model with the learned low-dimensional representations from longitudinal ECGs spanning a duration of 44 months.…”

Longer and better lives for patients with atrial fibrillation: the 9th AFNET/EHRA consensus conference

“…Highlighting significant segments in ECGs or visualizing decision-making in structured data sets can provide insights that can be interpreted based on mechanistic understanding. 170 , 177–180 Additionally, the availability of generic frameworks enables the visualization of decisions from various deep neural networks, making them applicable to multiple data sources, including structured data sets. 181 …”

Revolutionizing Cardiology through Artificial Intelligence—Big Data from Proactive Prevention to Precise Diagnostics and Cutting-Edge Treatment—A Comprehensive Review of the Past 5 Years

“…Artificial intelligence has found its usefulness in ECG for the following purposes: diagnosis of pulmonary thromboembolism [ 95 ], prediction of sudden death and cardiovascular events [ 97 , 98 ], prediction of fatal events after cardiac resynchronization [ 99 ], prediction of paroxysmal atrial fibrillation [ 37 ], detection of ventricular hypertrophy [ 100 ], risk prediction in liver transplantation [ 101 ], detection of ventricular dysfunction [ 103 ], and prediction of recurrence after paroxysmal atrial fibrillation ablation [ 38 ].…”