CNN Based Heart Rate Classification Using ECG Signal Without R-peak Detection for Rhythm-Aware Health and Emotion Monitoring

Heart rate is an important vital sign in health and wellness monitoring to identify various kinds of cardiac arrhythmias, such as sick sinus syndrome with slow heart rates (bradyarrhythmias) and atrial fibrillation with fast heart rates (tachyarrhythmias). Therefore, in this paper, we present a one-dimensional convolutional neural network (CNN) based heart rate classification (HRC) method using the electrocardiogram (ECG) waveform without R-peak detection with the major objective of developing heart rhythm-aware health and emotion monitoring systems. The proposed CNN-ECG-based HRC method consists of two major stages: preprocessing and CNN architecture for directly classifying the ECG signal into normal rate ECG, slow rate ECG, and fast rate ECG signals without the use of R-peak detection. The trained CNN model is obtained using the ECG signals (normal, slow, and fast heart rates) taken from the Massachusetts Institute of Technology Beth Israel Hospital arrhythmia (MIT-BIHA) database. On different kinds of untrained ECG signal databases, the CNN-based method achieves an overall accuracy of 89.18%, 92.74%, and 88.98% for the Apnea-ECG (APNEA-ECG) database, the MIT-BIH polysomnographic (MIT-BIH SLP) database, and the MIT-BIH atrial fibrillation (MIT-BIH AFib) database, respectively. Evaluation results show that the deep ECG waveform-based HRC method achieved promising results for the ECG signals having similar ECG morphological patterns within 10 -second ECG signals.