Studies On Key Techniques For Automatic Analysis Of Electrocardiogram

The research of ECG automated analysis and diagnosis has been focus in signal processing field in the last years.It'll be a great fruit to put the search into practice with a reliable performance,not only in medical field,but also in modem signal processing field and artificial intelligence field.The ECG automated analysis and diagnosis is an interdisciplinary field,involving a large number of basic theories and key technologies. Up to now,there are still lots of limitation and deficiencies existing in the issue.And it starves for improvement and innovation both in academic research and in practical application.Based on the status,this dissertation carries on the study in four main aspects,i.e.preprocessing,waveform detection and fiducial point location,feature extraction and selection,classification and recognition of ECG signals.1.Study on the ECG Signal Preprocessing:In view of practical requirement of ECG signal processing and automated analysis,a wavelet filter,a morphological filter and an adaptive filter are individually are designed and applied to ECG signal preprocessing,and modification is made in the operation of morphological filter method for elimination of ECG baseline wander.For the respective limitations of three filters above,a novel adaptive filter method based on wavelet transform and morphological operation is put forward.The model simulation and real data experiment proved the validity of the new filter,which exceeds both wavelet filter and morphological filter in performance.2.Study on the ECG Waveform Detection and Fidueial Point Location:This part is the foundation of the ECG automated analysis and diagnosis.A new method for QRS detection was proposed based on Marr wavelet transform and morphological operation, which integrates wavelet modulus maxima detecting method and morphological peak-valley detecting method.The detection accuracy and location precision of the method is quite high,and it provides an exact analysis instrument for automated ECG analysis.A Hilbert transform based algorithm for R-wave detection is also proposed to overcome the deficiencies of complex computational quantity and low work speed in wavelet detection method.The algorithm fully utilizes the properties of the Hilbert transform and the fast computational speed and high detection accuracy contributes to its real-time application.A P-wave detecting algorithm is brought forward by an adaptive QRS-T cancellation technique based on wavelet transforms.The method can reduce the effect of high frequency noise and baseline wander efficiently,and is non-sensitivity to the variations of ECG morphology and the occurrence of cardiac arrhythmia.A preferable result for P wave detection is acquired,and this method can be analogously applied to the detection and location of T wave.The validity of the methods mentioned above for waveform detection and fiducial point location has all been proved using either entire data or partial data of MIT-BIH database.3.Study on the ECG Feature Extraction and Selection:The effective extraction and selection of ECG feature can provide a powerful support for ECG automated classification and recognition.The time-domain feathers,wavelet transform domain characteristics and higher-order cumulants of ECG signal are defined and extracted as three types of ECG features,according to the distribution of ECG signal in time-frequency domain and characteristics of cardiac arrhythrnia signals.The basic theories and application on principal component analysis(PCA)and multiple discriminant analysis(MDA)in the issue are also studied in feature selection to achieve dimension reduction.4.Study on ECG Classification and Recognition:A novel approach is put forward for classifying cardiac an'hythmia signals based on ECG waveform features and support vector machine(SVM).SVM algorithm,which is commonly used for two-class classification problem,is generalized and used for the multi-class classification of ECG in the study.The deduction procedure of basic theory is presented for the realization of SVM multi-class classification in one time.The SVM multi-class classification method is used in ECG for the first time,and an elementary study is also carried out on the effect of parameter setting,feather extraction and selection on SVM classification performance.SVM based ECG classification is also carried out respectively with three different types of feature sets,including higher-order cumulants,wavelet transform characteristics and multi-domain fusion features.The classification method is validated with the 13 types of cardiac arrhythmia signals obtained from the total 48 ECG recordings in MIT-BIH database.PCA method and MDA method,used to select ECG feature,can greatly reduce the feature dimensions and improve the efficiency of SVM classifier while keeping the performance maintained mostly or descended very slightly.