TY - GEN
T1 - Multi-Channel Lightweight Convolutional Neural Network for Remote Myocardial Infarction Monitoring
AU - Cao, Yangjie
AU - Wei, Tingting
AU - Lin, Nan
AU - Zhang, Di
AU - Rodrigues, Joel J.P.C.
N1 - Funding Information:
ACKNOWLEDGMENT This research is supported by the Nature Science Foundation of China (No. 61972092), by National Funding from the FCT-Fundac¸ão para a Ciência e a Tecnologia, through the UID/EEA/500008/2019 Project; and by Brazilian National Council for Scientific and Technological Development (CNPq) via Grant No. 309335/2017-5.
Publisher Copyright:
© 2020 IEEE.
PY - 2020/4
Y1 - 2020/4
N2 - Remote Myocardial Infarction (RMI) monitoring uses electronic devices to detect the electrocardiogram changes and inform the doctor in emergency conditions, which is an effective solution to save the patient's life. In this paper, we propose the Multi-Channel Lightweight CNN (MCL-CNN), which combines electrocardiogram signals from four leads (v2, v3, v5 and aVL) to detect the Anterior MI (AMI). Its multi-channel design allows the convolution of each lead to be independent of each other, and allowing them to find the filter that best suits them. In addition, constructing a lightweight network using different convolutional combinations in the MCL-CNN model, which makes the network has certain advantages in computing runtime parameters and more suitable for mobile devices. Meanwhile, we use balanced cross entropy to solve the problem of dataset class imbalance. These strategies make the MCL-CNN suitable for multi-lead ECG processing. Experimental results using public ECG datasets obtained from the PTB diagnostic database demonstrate that MCL-CNN's accuracy is 96.65%.
AB - Remote Myocardial Infarction (RMI) monitoring uses electronic devices to detect the electrocardiogram changes and inform the doctor in emergency conditions, which is an effective solution to save the patient's life. In this paper, we propose the Multi-Channel Lightweight CNN (MCL-CNN), which combines electrocardiogram signals from four leads (v2, v3, v5 and aVL) to detect the Anterior MI (AMI). Its multi-channel design allows the convolution of each lead to be independent of each other, and allowing them to find the filter that best suits them. In addition, constructing a lightweight network using different convolutional combinations in the MCL-CNN model, which makes the network has certain advantages in computing runtime parameters and more suitable for mobile devices. Meanwhile, we use balanced cross entropy to solve the problem of dataset class imbalance. These strategies make the MCL-CNN suitable for multi-lead ECG processing. Experimental results using public ECG datasets obtained from the PTB diagnostic database demonstrate that MCL-CNN's accuracy is 96.65%.
KW - Convolution Neural Network
KW - Deep Learning
KW - Electrocardiogram
KW - Myocardial Infarction
UR - http://www.scopus.com/inward/record.url?scp=85087913990&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85087913990&partnerID=8YFLogxK
U2 - 10.1109/WCNCW48565.2020.9124860
DO - 10.1109/WCNCW48565.2020.9124860
M3 - Conference contribution
AN - SCOPUS:85087913990
T3 - 2020 IEEE Wireless Communications and Networking Conference Workshops, WCNCW 2020 - Proceedings
BT - 2020 IEEE Wireless Communications and Networking Conference Workshops, WCNCW 2020 - Proceedings
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2020 IEEE Wireless Communications and Networking Conference Workshops, WCNCW 2020
Y2 - 25 May 2020 through 28 May 2020
ER -