Journal
NEURAL COMPUTING & APPLICATIONS
Volume 32, Issue 12, Pages 8101-8113Publisher
SPRINGER LONDON LTD
DOI: 10.1007/s00521-019-04318-2
Keywords
Atrial fibrillation; Arrhythmia detection; Fully convolutional network; Residual network; Location of R peak position
Categories
Ask authors/readers for more resources
Atrial fibrillation (AF) is one of the most common arrhythmia diseases, the incidence of which is ascendant with age increase. What's more, AF is a high-risk factor for stroke, ischemia myocardial and other malignant cardiovascular diseases, which would threaten people's life significantly. Using a mobile device to screen AF segments is an effective way to reduce the mortality and morbidity of malignant cardiovascular diseases. However, most of existing AF detection methods mainly centered on clinical resting ECG signals and were incapable of processing mobile ECG signals with low signal-to-noise ratio which collected by mobile devices. In this paper, we take advantage of a fully convolutional network variant named U-Net for heart rhythmic information capturing by locating R peak positions as well as calculating RR intervals and a 34-layer residual network for waveform morphological features capturing from ECG signals. Combining both rhythmic information and waveform morphological features, two-layer fully connected networks are employed successively to discriminate AF, normal sinus rhythm , and other abnormal rhythm (other). The extensive experimental results show that our proposed AF our proposed AF screening framework named FRM-CNN can achieveF1value of 85.08 +/- 0.99% and accuracy of87.22 +/- 0.7 on identifying AF segments well without handcraft engineering. Compared with the cutting-edge AF detection methods, the FRM-CNN has more superior performance on monitoring people's health conditions with mobile devices.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available