A deep learning approach for arrhythmia detection and classification based on IoT wearable technology

The development of internet of things (IoT) technology linked with the healthcare industry and the recent growth in health wearable devices have allowed researchers to investigate in developing continuous monitoring systems to continuously analyze the enormous electrocardiogram (ECG) signals to detect and classify cardiac abnormalities and arrhythmia. However, researches often ignore the importance of recognizing arrhythmia in a timely manner to prevent and avoid any sudden and unexpected deaths of the subject. Thus, a high classification accuracy of the minority classes of ECG signals is extremely important for diagnosing arrhythmia. The heart’s operation is deemed to have long periods of normal activity and short periods of arrhythmic activity. So, achieving high accuracy is quite easy. Nevertheless, the difficulty relies in achieving high performance on other metrics such as F1 score, precision and recall, identifying the small cardiac abnormalities, the S and V beats. In this paper, a continuous ECG monitoring system tailored for athletes is proposed, the incentive is to analyze the collected data by the sensors at a Fog layer between the devices and the cloud layers of the proposed system architecture, where effective decisions can be taken immediately once something serious occur. To detect and classify arrhythmia; we propose an ECG arrhythmia classification system based on deep learning algorithms that comprises two major steps: 1) a deep learning model to classify the ECG signals into two different main beats, normal and abnormal; 2) a transfer learning algorithm is proposed to further classify the S and V beats in order to enhance their accuracy.