Research On Intelligent Recognition Algorithms Of Various Arrhythmias Based On Single Lead ECG

Arrhythmia is one of the most common cardiovascular diseases.Its performance on the electrocardiogram is sudden,transient and concealed.If it cannot be diagnosed in time,will lead to aggravation of the patient’s condition,and even pay the price of life in severe cases.The QRS complex of Electrocardiograph(ECG)signals can reflect some pathological features of the heart.Especially when some arrhythmias occur,the changes in the peak and width of the QRS complex are quite different from the normal ones.Most of the current methods for detecting the onset and ends of QRS complexes have reached millisecond accuracy.However,the recognition of QRS complexes morphology,especially abnormal patterns,are still a challenging problem.What’s more,with the popularity of portable ECG acquisition devices and the development of remote ECG monitoring,algorithms based on computer automatic diagnosis of arrhythmia will face challenges.How to develop a powerful algorithm with high accuracy and generalization from the massive ECG data of many arrhythmias will have very important practical significance.The machine learning method based on feature extraction has good feature interpretability and good classification performance.Deep learning-based methods for identifying arrhythmia usually have a stronger generalization ability.Combining these two to realize the detection of arrhythmia will be a good attempt.Therefore,this paper firstly used the revised Douglas-Peucker algorithm to develop a method that can identify the onset and ends and detecte peaks of the QRS complex of the ECG signal.sing this localization result as a partial feature,some other traditional feature extraction is performed on the ECG signal.Combining those traditional features and the deep feature which from the residual block network and the bidirectional long-term and short-term memory network,several kinds of arrhythmias are classified.The reasearch contents are as follows(1)Research on identification algorithm of peaks and onset-ends of QRS complex.Firstly,we used the revised Douglas-Peucker algorithm to obtain the corner feature points of each heart beat.And then,based on the conditions between the angle,slope,amplitude difference of the feature points and the direction of the feature points,the potential candidate points of the onset-ends and the peaks were distinguished.Finally,according to the standard naming rules of QRS complexes,we identified the selectedpoints as the actually characteristic peaks.We verified out method on QT database,the result showed that the error of the method for the recognition of the onset and end of the QRS complex is 2.76±7.6ms and 1.3±8.8ms respectively,and the overall sensitivity to the peak recognition of the QRS complex can reach 91%,positive prediction rate can reach 95%.Compared with similar related research,the accuracy of this method in the identification of QRS complex start and ends can reach the level of most methods,and the identification of each peak in QRS complex is slightly better than the existing methods.(2)Research on intelligent classification algorithms for multiple arrhythmias based on single-lead ECG.Firstly,the frequency domain,time domain and morphology features were extracted for 7 rhythms ECG,which include atrial fibrillation,atrial premature beats,ventricular premature beats,first degree atrioventricular block,left bundle branch block and right bundle branch block and normal rhythms.As a result,a total of 120 features were extracted above.Then a residual block network and bi-directional long-term and short-term memory network are built to extract deep features.Finally these features were fused in the network for classification of different kinds of arrhythmias.An ECG database provided by the 2018 China Physiological Signal Challenge was used to verify the our method firstly.The results showed that the F1 score of the 7 rhythm categories reached 0.855,which was better than the existing algorithm.Further,we tested our method on the ECG data provided by the PhysioNet Challenge 2017 competition which collected by the portable device,the experimental results show the algorithm’s excellent generalization ability.