Research On ECG Diagnosis Of Content-Based Retrieval

The ECG disease has become the top killer of threat to human’s health. Theprimary means to reduce the ECG disease mortality is early detection and earlytreatment. Diagnosis of the ECG disease mainly depends on the judgment of thedoctors on ECG. However, this method not only consumes the time and laboriousbut also the different doctors may provide the different judgment with the sameECG. Therefore, more and more experts and scholars begin to study the ECGautomatic analysis techniques. Automatic analysis can help doctors improve thespeed and accuracy of diagnosis.Based on the analysis of the previous ECG automatic diagnostic techniques,this paper presents an in-depth study at the critical issues of ECG automaticdiagnostic technology. Mainly studied the following three aspects: the ECGdenoising algorithm, the detection of the ECG signal feature point, the retrievalmethods and diagnostic strategy based on the retrieving content ECG diagnostictechniques:For the pre-processing of the ECG signal, according to the differentcharacteristics of the major ECG noise just like baseline drift and high-frequencynoise in ECG signal. This paper uses mathematical morphology and wavelettransform method to filter out the noise. Introduce the principle andimplementation steps of the two methods, and has performed the correspondingexperiment. The experimental results show the denoising better.For the characteristics of QRS wave, P wave and T wave, we studied thedetection method suitable for each waveform has been studied, use wavelettransform modulus maxima method to detect the QRS wave. And proposescorresponding method for multiple detection and miss detection problem. Thedifficulty to detect P wave is to judge whether heartbeat contain P wave, thisthesis puts forward a method which based on multi-feature to solve this problem, and the method combines wavelet transform and neural network. T wavedetection difficulty lies in the determination of T wave morphology. This thesisby uses the relation between T wave morphology and maxima modulus pair todetermine T wave morphology, and then position the T wave. This thesis carriesout the detection with the above three kinds of waveform. Experiment resultsshow that feature points which detected by the proposed method are nearly thesame with the points marked by experts, and have proved the proposed methodget lower false detection rate, higher sensitivity by carrying on the experiment ofmultiple ECG signals.The study focus in two main categories about the ECG diagnosis ofcontent-based retrieval, that is retrieval algorithms and diagnostic strategies,retrieval can implement through calculating the similarity of the feature of heartbeat. This thesis selects the features according to the predecessors’ experienceand the relationship between the heart beat type and waveform, this thesiscalculates the similarity using the Euclidean distance, the diagnostic strategy ofthe heartbeat,we develop a new algorithm which selects the heart beat as the typeof the tested that the heart beat’s type have the highest emergence frequency inthose heart beats of the larger similarity, and this thesis shows the accuracy of theretrieval and diagnosis using the above algorithm. This thesis designs theinterface of the diagnostic system. The interface can achieve selecting the ECG,extracting the ECG feature points, selecting the heartbeat, and the retrieval anddiagnosis of the heartbeat.