An Approach to Predict Multiple Cardiac Diseases
The First China ECG Intelligent Competition launched ECG challenge to classify 8 kinds of abnormalities from uneven 12-lead ECGs. These abnormalities can be classified into two categories according to morphology and rhythm, four in each group. In this paper, for morphology tasks neural network is applied mainly with input median wave extracted from raw data, while traditional methods are executed and promoted by machine learning to achieve rhythm classification. Non-coexistence relationship is taken into consideration to fit in clinical significance better. The final average F1 score is 0.886 on test set, which certificates these are effective methods for ECG auto detection.
KeywordsECG abnormalities Deep learning Machine learning
This work is supported by program ykj-2018-00393 of Technology foundation of Beijing University of Technology.
- 2.Muthuvel, K., Suresh, L.P., Alexander, T.J., Veni, S.H.K.: Classification of ECG signal using hybrid feature extraction and neural network classifier. In: Kamalakannan, C., Suresh, L.P., Dash, S.S., Panigrahi, B.K. (eds.) Power Electronics and Renewable Energy Systems. LNEE, vol. 326, pp. 1537–1544. Springer, New Delhi (2015). https://doi.org/10.1007/978-81-322-2119-7_150CrossRefGoogle Scholar
- 3.Sarfraz, M., Khan, A.A., Li, F.F.: Using independent component analysis to obtain feature space for reliable ECG Arrhythmia classification. In: 2016 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pp. 62–67. IEEE (2016)Google Scholar
- 6.Rajpurkar, P., Hannun, A.Y., Haghpanahi, M., et al.: Cardiologist-level arrhythmia detection with convolutional neural networks (2017)Google Scholar