A signal quality assessment method for mobile ECG using multiple features and fuzzy support vector machine

A signal quality assessment method for mobile ECG based on fuzzy support vector machines (FSVM) and multi-feature was proposed to help users to determine whether the ECG recordings collected using mobile phone are acceptable or not. The proposed method mainly included two modules: feature extraction and an intelligent classification approach, i.e. a FSVM classifier. First, 27 features derived from the baseline drift, the high or low amplitude, and the power spectrum of ECG were quantized and extracted to serve as the inputs of FSVM classifier. Then grid search (GS) was employed to optimize the parameters (sigma, C) for FSVM classifier. Finally, the performance of FSVM classifier was verified by comparing with the results of a kernel SVM (KSVM) classifier. Results showed that for 1,000 training mobile ECG recordings from the set A in PhysioNet/Computing in Cardiology Challenge 2011 database, the proposed FSVM classifier yielded a classification accuracy of 94.50% (vs. 93.90% for the KSVM classifier). For the 500 test mobile ECG recordings from the set B database, classification accuracies were 91.40% for the KSVM classifier vs. 92.00% for the FSVM classifier.