A novel multi-resolution SVM (MR-SVM) algorithm to detect ECG signal anomaly in WE-CARE project

Cardiovascular disease (CVD) has become the leading cause of human deaths today. In order to combat this disease, many professionals are using mobile electrocardiogram (ECG) remote monitoring system. While using mobile ECG systems, most of the cardiac anomalies can be observed, especially when serious myocardial ischemia, heart failure, and malignant arrhythmia occur. Thus, ECG anomaly detection and analysis have attracted more and more attention in the clinical and research communities. Currently, the existing solutions of ECG automatic detection and analysis technologies are challenged by an accuracy requirement. Based on this motivation, we propose a novel Multi-Resolution Support Vector Machine (MR-SVM) algorithm to detect ECG waveform anomaly. This proposal is tested in our WE-CARE (a Wearable Efficient telecardiology system) project. Clinical trials and experimental results show that the algorithm can successfully extract original QRS complex waves and T waves regardless of noise magnitude and distinguish the ST segment morphological anomalies. Compared with European standard ST-T database, our solution can achieve the average T wave recognition accuracy rate of 97.5% and ST anomaly detection accuracy rate of 93%.