A Novel Recurrent Neural Network Architecture for Classification of Atrial Fibrillation Using Single-lead ECG. Banerjee, R., Ghose, A., & Khandelwal, S. In 2019 27th European Signal Processing Conference (EUSIPCO), pages 1-5, Sep., 2019.
Paper doi abstract bibtex Atrial Fibrillation (AF) is a type of abnormal heart rhythm which may lead to a stroke or cardiac arrest. In spite of numerous research works, developing an automatic mechanism for accurate detection of AF remains a popular yet unsolved problem. In this paper, we propose a deep neural network architecture for classification of AF using single-lead Electrocardiogram (ECG) signals of short duration. We define a novel Recurrent Neural Network (RNN) structure, comprising two Long-Short Term Memory (LSTM) networks for temporal analysis of RR intervals and PR intervals in an ECG recording. Output states of the two LSTMs are merged at the dense layer along with a set of hand-crafted statistical features, related to the measurement of heart rate variability (HRV). The proposed architecture is proven on the open access PhysioNet Challenge 2017 dataset, containing more than 8500 single-lead ECG recordings. Results show that our methodology yields sensitivity of 0.93, specificity of 0.98 and F1-score of 0.89 in classifying AF, which is better than the existing accuracy scores, reported on the dataset.
@InProceedings{8902936,
author = {R. Banerjee and A. Ghose and S. Khandelwal},
booktitle = {2019 27th European Signal Processing Conference (EUSIPCO)},
title = {A Novel Recurrent Neural Network Architecture for Classification of Atrial Fibrillation Using Single-lead ECG},
year = {2019},
pages = {1-5},
abstract = {Atrial Fibrillation (AF) is a type of abnormal heart rhythm which may lead to a stroke or cardiac arrest. In spite of numerous research works, developing an automatic mechanism for accurate detection of AF remains a popular yet unsolved problem. In this paper, we propose a deep neural network architecture for classification of AF using single-lead Electrocardiogram (ECG) signals of short duration. We define a novel Recurrent Neural Network (RNN) structure, comprising two Long-Short Term Memory (LSTM) networks for temporal analysis of RR intervals and PR intervals in an ECG recording. Output states of the two LSTMs are merged at the dense layer along with a set of hand-crafted statistical features, related to the measurement of heart rate variability (HRV). The proposed architecture is proven on the open access PhysioNet Challenge 2017 dataset, containing more than 8500 single-lead ECG recordings. Results show that our methodology yields sensitivity of 0.93, specificity of 0.98 and F1-score of 0.89 in classifying AF, which is better than the existing accuracy scores, reported on the dataset.},
keywords = {bioelectric potentials;electrocardiography;medical disorders;medical signal detection;medical signal processing;recurrent neural nets;signal classification;recurrent neural network architecture;atrial fibrillation classification;abnormal heart rhythm;stroke;deep neural network architecture;single-lead electrocardiogram signals;recurrent neural network structure;long-short term memory networks;temporal analysis;RR intervals;PR intervals;hand-crafted statistical features;heart rate variability;open access PhysioNet Challenge 2017 dataset;single-lead ECG recordings;cardiac arrest;Electrocardiography;Heart rate variability;Time series analysis;Signal processing;Recurrent neural networks;Training;Atrial Fibrillation;Electrocardiogram;Long-Short Term Memory;Classification},
doi = {10.23919/EUSIPCO.2019.8902936},
issn = {2076-1465},
month = {Sep.},
url = {https://www.eurasip.org/proceedings/eusipco/eusipco2019/proceedings/papers/1570529299.pdf},
}
Downloads: 0
{"_id":"wcmCtyhGrLC8MaYxj","bibbaseid":"banerjee-ghose-khandelwal-anovelrecurrentneuralnetworkarchitectureforclassificationofatrialfibrillationusingsingleleadecg-2019","authorIDs":[],"author_short":["Banerjee, R.","Ghose, A.","Khandelwal, S."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","author":[{"firstnames":["R."],"propositions":[],"lastnames":["Banerjee"],"suffixes":[]},{"firstnames":["A."],"propositions":[],"lastnames":["Ghose"],"suffixes":[]},{"firstnames":["S."],"propositions":[],"lastnames":["Khandelwal"],"suffixes":[]}],"booktitle":"2019 27th European Signal Processing Conference (EUSIPCO)","title":"A Novel Recurrent Neural Network Architecture for Classification of Atrial Fibrillation Using Single-lead ECG","year":"2019","pages":"1-5","abstract":"Atrial Fibrillation (AF) is a type of abnormal heart rhythm which may lead to a stroke or cardiac arrest. In spite of numerous research works, developing an automatic mechanism for accurate detection of AF remains a popular yet unsolved problem. In this paper, we propose a deep neural network architecture for classification of AF using single-lead Electrocardiogram (ECG) signals of short duration. We define a novel Recurrent Neural Network (RNN) structure, comprising two Long-Short Term Memory (LSTM) networks for temporal analysis of RR intervals and PR intervals in an ECG recording. Output states of the two LSTMs are merged at the dense layer along with a set of hand-crafted statistical features, related to the measurement of heart rate variability (HRV). The proposed architecture is proven on the open access PhysioNet Challenge 2017 dataset, containing more than 8500 single-lead ECG recordings. Results show that our methodology yields sensitivity of 0.93, specificity of 0.98 and F1-score of 0.89 in classifying AF, which is better than the existing accuracy scores, reported on the dataset.","keywords":"bioelectric potentials;electrocardiography;medical disorders;medical signal detection;medical signal processing;recurrent neural nets;signal classification;recurrent neural network architecture;atrial fibrillation classification;abnormal heart rhythm;stroke;deep neural network architecture;single-lead electrocardiogram signals;recurrent neural network structure;long-short term memory networks;temporal analysis;RR intervals;PR intervals;hand-crafted statistical features;heart rate variability;open access PhysioNet Challenge 2017 dataset;single-lead ECG recordings;cardiac arrest;Electrocardiography;Heart rate variability;Time series analysis;Signal processing;Recurrent neural networks;Training;Atrial Fibrillation;Electrocardiogram;Long-Short Term Memory;Classification","doi":"10.23919/EUSIPCO.2019.8902936","issn":"2076-1465","month":"Sep.","url":"https://www.eurasip.org/proceedings/eusipco/eusipco2019/proceedings/papers/1570529299.pdf","bibtex":"@InProceedings{8902936,\n author = {R. Banerjee and A. Ghose and S. Khandelwal},\n booktitle = {2019 27th European Signal Processing Conference (EUSIPCO)},\n title = {A Novel Recurrent Neural Network Architecture for Classification of Atrial Fibrillation Using Single-lead ECG},\n year = {2019},\n pages = {1-5},\n abstract = {Atrial Fibrillation (AF) is a type of abnormal heart rhythm which may lead to a stroke or cardiac arrest. In spite of numerous research works, developing an automatic mechanism for accurate detection of AF remains a popular yet unsolved problem. In this paper, we propose a deep neural network architecture for classification of AF using single-lead Electrocardiogram (ECG) signals of short duration. We define a novel Recurrent Neural Network (RNN) structure, comprising two Long-Short Term Memory (LSTM) networks for temporal analysis of RR intervals and PR intervals in an ECG recording. Output states of the two LSTMs are merged at the dense layer along with a set of hand-crafted statistical features, related to the measurement of heart rate variability (HRV). The proposed architecture is proven on the open access PhysioNet Challenge 2017 dataset, containing more than 8500 single-lead ECG recordings. Results show that our methodology yields sensitivity of 0.93, specificity of 0.98 and F1-score of 0.89 in classifying AF, which is better than the existing accuracy scores, reported on the dataset.},\n keywords = {bioelectric potentials;electrocardiography;medical disorders;medical signal detection;medical signal processing;recurrent neural nets;signal classification;recurrent neural network architecture;atrial fibrillation classification;abnormal heart rhythm;stroke;deep neural network architecture;single-lead electrocardiogram signals;recurrent neural network structure;long-short term memory networks;temporal analysis;RR intervals;PR intervals;hand-crafted statistical features;heart rate variability;open access PhysioNet Challenge 2017 dataset;single-lead ECG recordings;cardiac arrest;Electrocardiography;Heart rate variability;Time series analysis;Signal processing;Recurrent neural networks;Training;Atrial Fibrillation;Electrocardiogram;Long-Short Term Memory;Classification},\n doi = {10.23919/EUSIPCO.2019.8902936},\n issn = {2076-1465},\n month = {Sep.},\n url = {https://www.eurasip.org/proceedings/eusipco/eusipco2019/proceedings/papers/1570529299.pdf},\n}\n\n","author_short":["Banerjee, R.","Ghose, A.","Khandelwal, S."],"key":"8902936","id":"8902936","bibbaseid":"banerjee-ghose-khandelwal-anovelrecurrentneuralnetworkarchitectureforclassificationofatrialfibrillationusingsingleleadecg-2019","role":"author","urls":{"Paper":"https://www.eurasip.org/proceedings/eusipco/eusipco2019/proceedings/papers/1570529299.pdf"},"keyword":["bioelectric potentials;electrocardiography;medical disorders;medical signal detection;medical signal processing;recurrent neural nets;signal classification;recurrent neural network architecture;atrial fibrillation classification;abnormal heart rhythm;stroke;deep neural network architecture;single-lead electrocardiogram signals;recurrent neural network structure;long-short term memory networks;temporal analysis;RR intervals;PR intervals;hand-crafted statistical features;heart rate variability;open access PhysioNet Challenge 2017 dataset;single-lead ECG recordings;cardiac arrest;Electrocardiography;Heart rate variability;Time series analysis;Signal processing;Recurrent neural networks;Training;Atrial Fibrillation;Electrocardiogram;Long-Short Term Memory;Classification"],"metadata":{"authorlinks":{}},"downloads":0},"bibtype":"inproceedings","biburl":"https://raw.githubusercontent.com/Roznn/EUSIPCO/main/eusipco2019url.bib","creationDate":"2021-02-11T19:15:22.048Z","downloads":0,"keywords":["bioelectric potentials;electrocardiography;medical disorders;medical signal detection;medical signal processing;recurrent neural nets;signal classification;recurrent neural network architecture;atrial fibrillation classification;abnormal heart rhythm;stroke;deep neural network architecture;single-lead electrocardiogram signals;recurrent neural network structure;long-short term memory networks;temporal analysis;rr intervals;pr intervals;hand-crafted statistical features;heart rate variability;open access physionet challenge 2017 dataset;single-lead ecg recordings;cardiac arrest;electrocardiography;heart rate variability;time series analysis;signal processing;recurrent neural networks;training;atrial fibrillation;electrocardiogram;long-short term memory;classification"],"search_terms":["novel","recurrent","neural","network","architecture","classification","atrial","fibrillation","using","single","lead","ecg","banerjee","ghose","khandelwal"],"title":"A Novel Recurrent Neural Network Architecture for Classification of Atrial Fibrillation Using Single-lead ECG","year":2019,"dataSources":["NqWTiMfRR56v86wRs","r6oz3cMyC99QfiuHW"]}