Research On ECG Automatic Classification Algorithms Based On Multi-Feature Selection

In recent years, ECG automatic classification study has become one of the researchhotspots in signal processing field. It is widely used in intensive care units, wearable ECGdevices, relationship studies of diseases and heart activities, function evaluations ofpacemakers and so on. By introducing the computer-aided electrocardiogram (ECG)analysis, doctors could focus on complex ECG waveform diagnosis. In this way, it couldimprove the diagnostic efficiency and shorten the time of diagnosis, and have positivesignificances of condition guardianship and rehabilitation evaluation.1. In this paper, the key technologies of ECG automatic classification are studied,including the location of characteristic points, the extraction of feature data and theselection of classification algorithm. In the process of ECG analysis and diagnosis, we willfirstly encounter the problem of QRS complex detection, and its premise is positioning Rwaves effectively. Based on previous ECG wave detection algorithms, a Two-wayMulti-point Difference Operation Method (TMDOM) for R wave detection is proposed.Compared to traditional detection methods, TMDOM is simple and effective, and has ahigh recognition rate for R waves. Compared with Difference Operation Method (DOM),TMDOM has greater noise immunity.2. After detecting R waves, the next step is to select ECG eigenvectors. Based onanalyzing and summarizing a variety of ECG characteristics home and abroad,multi-features of ECG is extracted effectively, including R-R intervals, sample variances,wavelet coefficients, energy ratios, as well as the most important low-dimensional featuresmapped from high-dimensional ones. Among them, two important mapping characteristicsare focused on: the coefficients obtained by NMF and the characteristics gotten fromKernel Local Fisher Discriminant Analysis which have strong separability. 3. By selecting multi-features, we could obtain ECG data which represents theoriginal ECG data. At last, according to the selective characteristics, we should chooseappropriate classifier to achieve the ECG automatic classification and diagnosis. Twoclassification models have been proposed: one is based on NMF and SVM, the other isbased on KLFDA and decision tree. Experiments show that the two methods have achievedhigh recognition rate during ECG automatic classification, so they have obvious diagnosticadvantages and research significance.