Research Article
A Two-Stream Model Combining ResNet and Bi-LSTM Networks for Non-contact Dynamic Electrocardiogram Signal Quality Assessment
@INPROCEEDINGS{10.1007/978-3-031-06368-8_21, author={Guoqiang Zhu and Yang Li and Yonglin Wu and Zhikun Lie and Chen Chen and Wei Chen}, title={A Two-Stream Model Combining ResNet and Bi-LSTM Networks for Non-contact Dynamic Electrocardiogram Signal Quality Assessment}, proceedings={Wireless Mobile Communication and Healthcare. 10th EAI International Conference, MobiHealth 2021, Virtual Event, November 13--14, 2021, Proceedings}, proceedings_a={MOBIHEALTH}, year={2022}, month={6}, keywords={Non-contact ECG Signal quality assessment Convolutional neural network (CNN) Recurrent neural network (RNN)}, doi={10.1007/978-3-031-06368-8_21} }- Guoqiang Zhu
Yang Li
Yonglin Wu
Zhikun Lie
Chen Chen
Wei Chen
Year: 2022
A Two-Stream Model Combining ResNet and Bi-LSTM Networks for Non-contact Dynamic Electrocardiogram Signal Quality Assessment
MOBIHEALTH
Springer
DOI: 10.1007/978-3-031-06368-8_21
Abstract
With the rapid advancement of monitoring without contact electrocardiogram (ECG), dynamic and real-time signal quality assessment (SQA) becoming a practical problem. In this paper, a two-stream structure that combines residual network (ResNet) and bidirectional long short-term memory (Bi-LSTM) for dynamic ECG (dECG) signals quality assessment is proposed. The ResNet stream is dedicated to extracting spatial features using time-frequency spectrum images as inputs. Meanwhile, the Bi-LSTM stream is devoted to exploring the temporal features using the ECG time series as the input. Then, these two streams are fused using a decision fusion mechanism and the performance is significantly boosted. As compared to the single stream-based approach, the proposed structure can compensate for either temporal or spatial information effectively. The overall accuracy of 99.69% can be achieved in distinguishing the dECG signal quality into three categories, namely clear ECG signal with clear R waves, blurry ECG signal without clear R waves, and noisy ECG signal with motion artifact. Experimental results show that this method demonstrates superior accuracy in determining the quality of dECG signals measured by the noncontact device. Therefore, the proposed model is expected to be a promising solution for non-contact dECG signal quality assessment in a practical ECG diagnosis.
