Research On The Classification Of ECG Arrhythmia Based On Wavelet Analysis And Neural Network

The heart is one of the most significant organs in our body.Electrocardiogram (ECG) has already become a means of measuring and testingfor a healthy heart since we began to detect ECG signals. With the research ofelectro-cardiology and the development of computer science， the automaticclassification of ECG signal has been increasingly becoming an important topic.In this paper， it combines wavelet transform with neural network technique tomake a further research on the topic of ECG automatic classification. Also， itproves that neural network technique is a significant role in the computerautomatic diagnosis.At the beginning， the paper firstly gives us today’s development andresearch status of ECG signals’ automatic classification and diagnosis. Afterthat， the paper briefly introduces the mechanism and the basic knowledge ofECG. It interprets several typical ECG arrhythmia signals， which includes thecriteria of diagnosis and the characteristic of arrhythmia wave. Then it analyzesthe difficulties of the ECG signal classification and diagnosis which are powerline interference， electromyographic noise and baseline drift. These kinds of noise make the computer difficult to classify the ECG signals precisely， andalso interferes the machine to extract appropriate features. Based on this status，the paper compared several approaches aimed at ECG signals de-noising， andfinally combines the wavelet threshold de-nosing with the waveletdecomposition and reconstruction filter to filter the power line interference andelectromyographic noise， to effectively correct the baseline drift. All of thispreprocessing work is for the following wave detection and classification.The paper introduces the QRS wave detection method based on wavelettransform theory. It mainly illustrates the use of two orthogonal quadraticb-spline wavelet detection algorithms. The result of simulation experimentsshows that this algorithm can do an accurately positioning to ECG data， whichcan reach an as high as above99%detection rate for QRS wave. Therefore， itmeans that this approach can accurately extract the waveform characteristicsparameters as input set for post processing.At last， based on the current clinical classification criteria， this paperdesigned and trained an improved BP neural network usingLevenberg-Marquardt algorithm. The Levenberg-Marquardt algorithm at thesame time has the advantages of the steepest gradient descent method andNewton’s method， which not only is able to overcome the disadvantage of theslow convergence speed of BP network， but also solves the low trainingprecision problem.In this paper， the experimental training sample sets are taken from the internationally popular MIT-BIH standard arrhythmia database. The results ofthe simulation experiment shows that the classification by the improved BPnetwork model gives a good performance on the identification of heartpremature symptoms. Besides， when the network is changed by takingdifferent features as the parameter， it still can classify a variety of arrhythmiasymptoms. And it also performs well as the result shows.