TY - JOUR
T1 - Visualization deep learning model for automatic arrhythmias classification
AU - Jiang, Mingfeng
AU - Qiu, Yujie
AU - Zhang, Wei
AU - Zhang, Jucheng
AU - Wang, Zhefeng
AU - Ke, Wei
AU - Wu, Yongquan
AU - Wang, Zhikang
N1 - Publisher Copyright:
© 2022 Institute of Physics and Engineering in Medicine.
PY - 2022/8/31
Y1 - 2022/8/31
N2 - Objective. With the improvement of living standards, heart disease has become one of the common diseases that threaten human health. Electrocardiography (ECG) is an effective way of diagnosing cardiovascular diseases. With the rapid growth of ECG examinations and the shortage of cardiologists, accurate and automatic arrhythmias classification has become a research hotspot. The main purpose of this paper is to improve accuracy in detecting abnormal ECG patterns. Approach. A hybrid 1D Resnet-GRU method, consisting of the Resnet and gated recurrent unit (GRU) modules, is proposed to implement classification of arrhythmias from 12-lead ECG recordings. In addition, the focal Loss function is used to solve the problem of unbalanced datasets. Based on the proposed 1D Resnet-GRU model, we use class-discriminative visualization to improve interpretability and transparency as an additional step. In this paper, the Grad-CAM++ mechanism has been employed to the trained network model and generate thermal images superimposed on raw signals to explore underlying explanations of various ECG segments. Main results. The experimental results show that the proposed method can achieve a high score of 0.821 (F1-score) in classifying 9 kinds of arrythmias, and Grad-CAM++ not only provides insight into the predictive power of the model, but is also consistent with the diagnostic approach of the arrhythmia classification. Significance. The proposed method can effectively select and integrate ECG features to achieve the goal of end-to-end arrhythmia classification by using 12-lead ECG signals, which can serve a promising and useful way for automatic arrhythmia classification, and can provide an explainable deep leaning model for clinical diagnosis.
AB - Objective. With the improvement of living standards, heart disease has become one of the common diseases that threaten human health. Electrocardiography (ECG) is an effective way of diagnosing cardiovascular diseases. With the rapid growth of ECG examinations and the shortage of cardiologists, accurate and automatic arrhythmias classification has become a research hotspot. The main purpose of this paper is to improve accuracy in detecting abnormal ECG patterns. Approach. A hybrid 1D Resnet-GRU method, consisting of the Resnet and gated recurrent unit (GRU) modules, is proposed to implement classification of arrhythmias from 12-lead ECG recordings. In addition, the focal Loss function is used to solve the problem of unbalanced datasets. Based on the proposed 1D Resnet-GRU model, we use class-discriminative visualization to improve interpretability and transparency as an additional step. In this paper, the Grad-CAM++ mechanism has been employed to the trained network model and generate thermal images superimposed on raw signals to explore underlying explanations of various ECG segments. Main results. The experimental results show that the proposed method can achieve a high score of 0.821 (F1-score) in classifying 9 kinds of arrythmias, and Grad-CAM++ not only provides insight into the predictive power of the model, but is also consistent with the diagnostic approach of the arrhythmia classification. Significance. The proposed method can effectively select and integrate ECG features to achieve the goal of end-to-end arrhythmia classification by using 12-lead ECG signals, which can serve a promising and useful way for automatic arrhythmia classification, and can provide an explainable deep leaning model for clinical diagnosis.
KW - arrhythmia classification
KW - gated recurrent unit
KW - interpretability
KW - res-net
UR - http://www.scopus.com/inward/record.url?scp=85136242400&partnerID=8YFLogxK
U2 - 10.1088/1361-6579/ac8469
DO - 10.1088/1361-6579/ac8469
M3 - Article
C2 - 35882225
AN - SCOPUS:85136242400
SN - 0967-3334
VL - 43
JO - Physiological Measurement
JF - Physiological Measurement
IS - 8
M1 - 085003
ER -