Premature ventricular contraction analysis for real-time patient monitoring

Background and objective: Improvements in wearable sensor devices make it possible to constantly monitor physiological parameters such as electrocardiograph (ECG) signals for long periods. Remotely monitoring patients using wearable sensors has an important role to play in health care, particularly given the prevalence of chronic conditions such as premature ventricular contraction (PVC), one of the prominent causes of death world-wide. PVC is a serious cardiovascular condition that can lead to life-threatening conditions. The instant recognition of life-threatening cardiac arrhythmias based on a wearable ECG sensor for a few seconds is a challenging problem of clinical significance. Method: Twenty seconds of consecutive ECG beats that were identified empirically to characterise a PVC episode were analysed. Three morphological features and seven statistical features were directly extracted in real time. These features were normalized and fed into an artificial neural network (ANN) classifier for classification. The PVC detector was uploaded into a smartphone to classify each episode as either PVC or non-PVC. Results: The proposed algorithm was tested on the MIT-BIH Arrhythmia, St. Petersburg Institute of Cardiological Technics (INCART) and Shimmer3 ECG databases. The proposed method resulted in an improved sensitivity, positive predictive value and accuracy of 98.7%, 97.8% and 98.6% respectively compared to recently published methods. In addition, the proposed method is suitable for real-time patient monitoring as it is computationally simple and requires only a few seconds of ECG recording to detect a PVC rhythm. Conclusion: This study provides a better and more accurate identification of the presence of PVC beats from wearable ECG recordings/mobile environment and standard environment, leading to more timely diagnosis and treatment outcomes. (C) 2018 Elsevier Ltd. All rights reserved.