ECG quality assessment based on multi-feature fusion

This paper proposes a new method for ECG quality classification based on multi-feature fusion. Lots of features, including waveform attributes, power spectrum, R-wave detection, etc., are given and each feature is evaluated independently. For the best performance, different combinations of features are tested. Rule-based method and learning-based method are considered for classification. The database from PhysioNet/Computing in Cardiology Challenge 2011 is used for performance evaluation and 92.8% and 90.4% classification accuracy are obtained in the training and test collection respectively using the rule-based method, and the average processing time of each ECG recording is 0.78s. Furthermore, learning-based method gets higher classification accuracy, and 94.0% and 91.6% are achieved in the training and test collection respectively, but the time cost is a little larger than rule-based method, and it is 2.03s.