Research On Automatic ECG Detection And Diagnosis Method

With the development of our society, cardiac diseases have been the majorhealth concerns. Cardiac diseases keep puzzling patients and doctors for theirchronicity and burstiness. In order to find the heart abnormalities, we wish to find asimple and effective way to monitor the activities of heart. Theelectrocardiogram(ECG) as a kind of electrophysiological signals collected by anon-invasive way has got widely applied. In the area of automatic detection anddiagnosis method of dynamic ECG signals, a lot of researches are ongoing.To implement an intelligent24-hour ECG monitoring system, a set of effectivedetection and diagnosis algorithms is needed. This paper gives a series of researchon the detection of five characteristic ECG waves and classification of severalcommon ECG abnormalities.Firstly, a detection algorithm of QRS complex based on wavelet transform andK-means clustering is proposed. The wavelet transform is used as analyzing tool forthe preprocessing. Then, based on the slope information, a QRS area is separated byusing K-means clustering. In this QRS area, the R-wave is detected, and then, theQ-wave and S-wave get detected too. Then, the detection of P-wave and T-wave isbased on the distribution information of relative positions in R-R interval. Afterthese, based on the detection result of these five characteristic waves, waveinformation such as time interval and wave height are extracted. In the second partof this paper, a SVM-based classification method is proposed to classify the normalsinus rhythm and four abnormal beats: left bundle branch block beat(LBBBB), rightbundle branch block beat(RBBBB), atrial premature contraction(APC) andpremature ventricular contraction(PVC).The MIT-BIH Arrhythmia Database is selected as the experimental dataset forthis paper. The annotation files have included the information of R-wave positionand abnormal ECG categories. Comparing the result of the proposed detectionalgorithm with annotations given by MIT-BIH Arrhythmia Database, QRS detectionsensitivity of99.72%is proved. And comparing the result of P-wave and T-wavedetection with the annotation made by our researchers, P-wave detection sensitivityof93.11%and T-wave detection sensitivity of94.11%are got. Besides, an averageclassification accuracy of95.06%of the proposed classification method is got in theend.