Research On ECG Morphological Features Recognition And Its Effect To Classification

The analysis of electrocardiogram (ECG) is a non-invasive, effective, simple and low-cost technique to detect the electrical heart activity. To detect and classify ECG diseases by computer is a classic pattern recognition task. Morphology features, which influence ECG diagnosis results, are introduced according to physicians'experiences and advice. Fail to utilize them should be one of the most important reasons for the underperformance of automatic ECG classification.Standard ECG databases are created for validating and comparing different algorithms on feature detection and disease classification. At present, there are four frequently used standard databases:MIT-BIH arrhythmia database, QT database, CSE multi-lead database and AHA database. With the development in equipment and diagnosis approach, these databases can not meet the requirements of further R&D works. Therefore Chinese Cardiovascular Disease Database (CCDD or CCD database), which contains 12-Lead ECG data, detailed annotation features and beat diagnosis result is constructed. It is advanced for not only improving the raw ECG data's technical parameters, but also introducing valuable morphology features which are utilized by experienced cardiologists effectively. All test works in this paper are based on CCDD.After that, two methods are presented for ECG morphology features recognition. The first one is based on 1 nearest neighbor classifier and dynamic time warping (INN-DTW). INN-DTW is a strong time series matching algorithm and used to compare the ECG segments with the templates stored in the system. Template selection and reduction is applied to accelerate the classification speed and cut down the templates volume as well. With the help of new proposed template reduction algorithm, an accuracy of 90.71% is acquired by using a small portion of the original template set. The second one is a real time algorithm focused on higher speed by simulating cardiologists' recognition procedure while using DTW algorithm for double checking. The accuracy of this method is 91.07%.In order to answer the "why" and "how" challenges of ECG morphological features' real utility, four kinds of representation method are proposed for morphological features, and experiments results are compared on 5 kinds of widely used classifiers between using morphological features and without morphological features. By utilizing feature selection algorithms, the performance is improved once again.The study of morphology features recognition on ECG should be investigated further to meet the requirements of clinical classification.