.png){kind=icon}
.png){kind=icon}
Return
Efficient ECG classification method for arrhythmia using MODWPT and adaptive incremental broad learning
.png){kind=icon}
PRE
AI
DOI：10.1016/j.bspc.2025.107516.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Arrhythmia, which is defined as a problem with the rhythm or rate at which the heart operates, is highly imperceptible yet insidious and can result in severe complications. In this study, in order to develop and evaluate an efficient arrhythmia detection method, we propose a lightweight network for identifying the electrocardiograph (ECG) signal which combines maximum overlap discrete wavelet packet transform (MODWPT) and adaptive incremental broad learning (Ada-IBL). The adaptive technology of Ada-IBL can well avoid the limitations of empirical selection parameters, and has the ability to balance time cost and accuracy. The classification process is as follows: firstly, the baseline drift noise of ECG signal is removed by using a low-computation median filter, then the multi-angle features of the de-noised signal are extracted, including the energy spectrum features based on MODWPT and the conventional morphological and rhythmic features. Finally, the above features are fused as the input data of Ada-IBL. We evaluate the proposed network on MIT-BIH arrhythmia database based on AAMI ECAR-1987. Experimental results show that, compared with convolutional neural networks, Ada-IBL has better computational efficiency and classification accuracy, which only takes 0.31 s to achieve 99.15% classification accuracy on the test set, and the training time is less than 2.4 s. Moreover, our overall performance is better than the most advanced networks, especially the accuracy of ventricular and supraventricular ectopic beats has reached 97.2% and 99.9%.
Keywords:
Arrhythmia
ECG
MODWPT
Multi-angle feature
Adaptive incremental broad learning
Journal
IF:
4.9
Papers:
9.8K
Citations:
2.4W