Real time ECG characteristic point detection with randomly selected signal pair difference (RSSPD) feature and random forest classifier

Detection of ECG characteristic points serves as a first step in many automated ECG analysis techniques. We propose a novel statistical scheme for ECG recognition. We design an effective feature, i.e. the randomly selected signal pair difference (RSSPD) feature, to effectively represent ECG morphology. The RSSPD generates a feature pool of signal pairs, and then a multi-class random forest is adopted in this work to train a classification model to detect all the ECG characteristic points simultaneously. After the post processing stage, the final output is generated. Our work provides robust and accurate detection performance in ECG datasets. The evaluation results on QT database show better detection accuracy compared with other studies.