Arrhythmias Classification by Integrating Stacked Bidirectional LSTM and Two-Dimensional CNN
Classifying different types of arrhythmias based on ECG signal is an important research topic in healthcare. Traditional methods focus on extracting varieties of features from ECG and using them to build a classifier. However, ECG usually presents high inter- and intra-subjects variability both in morphology and timing, hence, it’s difficult for predesigned features to accurately depict the fluctuation patterns of each heartbeat. To this end, we propose a novel arrhythmias classification model by integrating stacked bidirectional long short-term memory network (SB-LSTM) and two-dimensional convolutional neural network (TD-CNN). Particularly, SB-LSTM mines the long-term dependencies contained in ECG from both directions to depict the overall variation trend of ECG, while TD-CNN exploits local characteristics of ECG to characterize the short-term fluctuation patterns of ECG. Moreover, we design a discrete wavelet transform (DWT) based ECG decomposition layer and a Sum Rule based intermediate classification result fusion layer, by which ECG can be analyzed from multiple time-frequency resolutions, and the classification results of our model can be more accurate. Experimental results based on MIT-BIH arrhythmia database shows that our model outperforms 3 baseline methods, achieving 99.5% of accuracy, 99.9% of sensitivity and 98.2% specificity, respectively.
KeywordsArrhythmias classification Stacked bidirectional LSTM Convolutional neural network Wavelet decomposition Classification result fusion
This work was partially supported by the National Natural Science Foundation of China (No. 61332013, No. 61672161), the National Key Research and Development Program of China (No. 2016YFB1001400), and the China Scholarship Council (No. 201706290110).
- 5.ANSI/AAMI EC57: Testing and Reporting Performance Results of Cardiac Rhythm and ST Segment Measure Algorithms (2012)Google Scholar
- 8.Mant, J., Fitzmaurice, D.A., et al.: Accuracy of diagnosing atrial fibrillation on electrocardiogram by primary care practitioners and interpretative diagnostic software: analysis of data from screening for atrial fibrillation in the elderly (SAFE) trial. BMJ 7616, 335–380 (2007)Google Scholar
- 13.Jiang, H., Zhou, R., Zhang, L., Wang, H., Zhang Y.: Sentence level topic models for associated topics extraction. World Wide Web. https://doi.org/10.1007/s11280-018-0639-1
- 16.Liu, F., Zhou, X., Wang, Z., Wang, T., Ni, H., Yang, J.: Identifying obstructive sleep apnea by exploiting fine-grained BCG features based on event phase segmentation. In: IEEE BIBE, pp. 293–300 (2016)Google Scholar
- 18.Liu, F., Zhou, X., Wang, Z., Ni, H., Wang, T.: OSA-weigher: an automated computational framework for identifying obstructive sleep apnea based on event phase segmentation. J. Ambient Intell. Hum. Comput. (2018). https://doi.org/10.1007/s12652-018-0787-2
- 19.Liu, F., Zhou, X., Wang, Z., Wang, T., Zhang, Y.: Identification of hypertension by mining class association rules from multi-dimensional features. In: ICPR 2018, pp. 3114–3119 (2018)Google Scholar
- 22.Liu, F., Zhou, X., Wang, Z., et al.: A light-weight data preprocessing and integrative scheduling framework for health monitoring. In: IEEE-EMBS BHI, pp. 192–195 (2016)Google Scholar
- 23.Andreotti, F., Carr, O., Pimentel, M.A.F., Mahdi, A., Vos, M.D.: Comparing feature-based classifiers and convolutional neural networks to detect arrhythmia from short segments of ECG. Comput. Cardiol. 44, 1 (2017)Google Scholar
- 24.Krizhevsky, A., Sutskever, I., Hinton, G.E.: ImageNet classification with deep convolutional neural networks. In: NIPS, pp. 1097–1105 (2012)Google Scholar
- 25.Coşkun, M., Uçar, A., Yıldırım, Ö., et al.: Face recognition based on convolutional neural network. In: IEEE MEES, pp. 376–379 (2017)Google Scholar
- 31.Xie, J., Wang, Z., Yu, Z., Guo, B.: Enabling efficient stroke prediction by exploring sleep related features. In: IEEE UIC, pp. 452–461 (2018)Google Scholar