Generation of 12-Lead Electrocardiogram with Subject-Specific, Image-Derived Characteristics Using a Conditional Variational Autoencoder

Deep learning models have proven their value in the analysis of electrocardiogram (ECG). Among these, deep generative models have shown their ability in ECG generation. In this paper, we propose a conditional variational autoencoder (cVAE) to automatically generate realistic 12-lead ECG signals. Our...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Proceedings (International Symposium on Biomedical Imaging) S. 1 - 5
Hauptverfasser: Sang, Yuling, Beetz, Marcel, Grau, Vicente
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 28.03.2022
Schlagworte:
Analytical models
Biological system modeling
Deep learning
ECG Generation
Electrocardiography
Heart
Heart Position and Orientation
Morphology
Sensitivity
Variational Auto-encoder
ISSN:1945-8452
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Deep learning models have proven their value in the analysis of electrocardiogram (ECG). Among these, deep generative models have shown their ability in ECG generation. In this paper, we propose a conditional variational autoencoder (cVAE) to automatically generate realistic 12-lead ECG signals. Our method differs from previous papers in that (i) it generates complete 12-lead studies and (ii) generated ECGs can be adjusted to correspond to specific subject characteristics, particularly those from images. We demonstrate the ability of the model to adjust to age, sex and Body Mass Index (BMI) values. Our model is the first to incorporate imaging information by including heart position and orientation as input conditions, to analyse anatomical influences on generated ECG morphology. The network shows high accuracy and sensitivity to different conditions. In addition, our method can extract a ten-dimensional latent space containing interpreted features of the 12 ECG leads, which correspond to interpretable ECG features.
ISSN:1945-8452
DOI:10.1109/ISBI52829.2022.9761431