.png){kind=icon}
返回
Diffusion based comprehensive approach for highly contaminated electrocardiogram segmentation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.bspc.2024.106693.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Electrocardiogram (ECG) segmentation is the foundation for cardiac disease diagnosis. The precise localization of the onsets, peaks and offsets of P waves, T waves and QRS complexes would contribute to the precise diagnosis of cardiac disease, such as cardiac arrhythmia, myocardial ischemia and myocardial infarction, etc. Existing methods can achieve reasonably satisfactory segmentation results using a public annotated dataset. However, we found that with strong noises or partial damages in electrocardiogram data will lead to a significant impact on the accuracy of electrocardiogram segmentation, and incorrect segmentation could greatly reduce the accuracy of algorithms for the subsequent clinical cardiac disease diagnosis. In this study, we proposed a comprehensive approach which is based on diffusion model and applied it to the segmentation task to solve the problem of inaccurate delineation for highly contaminated ECG. We introduced a twophase approach. In phase one, we designed a multi-scale feature extraction structure combined with a cross transformer mechanism as backbone, and integrated it into the diffusion model DDPM to acquire the denoised reconstructed signal. In phase two, for the sake of fairness in comparison, we use the traditional UNet to further achieve the goal of solving the problem of highly contaminated ECG segmentation to compare with other more advanced segmentation networks. To demonstrate the effectiveness of the proposed method, Lobachevsky University Electrocardiography Database (LUDB) was used for the benchmarking. We have designed five types of noise adding strategies and tested the corresponding segmentation effects separately. Compared with current main segmentation networks on the same tasks, it validates that the proposed approach has outperformed the state-of-the-art segmentation method in terms of the quality and capability of ECG segmentation. In particular, the Dice Similarly Coefficient for detecting P waves, T waves and QRS complexes in the most contaminated ECG signals are at least 88.33%, 88.94% and 86.90%, respectively.
Keyword:
ECG segmentation
Deep learning
Contaminated ECG
Transformer
Diffusion model
CNN
期刊
IF:
4.9
论文数:
1.0W
被引数:
2.4W
机构
引用论文
LUDB: A New Open-Access Validation Tool for Electrocardiogram Delineation AlgorithmsLUDB: 一种新的心电描记算法开放验证工具.png)
IEEE ACCESS
OAAI
IEEE ACCESSIF3.6
OAAI分享收藏Light‐Addressable Capsules as Caged Compound Matrix for Controlled Triggering of Cytosolic Reactions
PREAI
分享收藏
Arrhythmia detection using deep convolutional neural network with long duration ECG signals基于深度卷积神经网络的长持续时间心电信号心律失常检测
PREAI
分享收藏
查看更多内容