Research On ECG Signal Classification Based On Curve Similarity

Electrocardiogram(ECG)heart rhythm recognition is an important method for heart disease diagnosis.At present,it is mainly relying on medical ECG equipment to detect heart disease.And doctors need to memorize many rules to determine whether the ECG waveform is abnormal or not,so as to diagnose whether the patient is sick.However,with the increase of patients and the complexity waveforms of abnormal ECG signal,the pressure of doctors is increasing.Therefore,it is of great significance to perform automatic identification research on the patient's ECG signal waveform.Based on this,this thesis proposes a disease matching recognition classification algorithm for ECG signals based on curve similarity.The major tasks of this article are as follows:(1)This thesis proposes a R peak detection method based on differentiation and adaptive threshold.The classical algorithms had many drawbacks,such as consuming numerous calculation resources,poor applicability,and poor real-time performance.To overcome the drawbacks,this thesis computes the differential of ECG signals after denoising,and set adaptive threshold to obtain the time interval where the R wave occurs.Through actual analysis,there exists a time lag between the differential signal and the original signal.Then,a new time interval after time shifting can be gained where the R wave occurs on the original signal.Compared to the official annotation information,the proposed algorithm shows a comparative result in the performance of real-time detection and acquires a great accuracy.(2)The thesis presents a matching method to recognize the ECG types based on the curve segmentation of ECG signals.It could be found that the heartbeats from the same type have similar shapes of ECG waves,while those from different types shows great differences.Therefore,the thesis uses dynamic time warping(DTW)algorithm to judge the similarity of two ECG signals.Experiments find that some segments also have partial similarities between different types of ECG,which leads to a number of mismatches in the overall DTW matching.In order to overcome this problem,the segmental matching is proposed to divide the every ECG signal to six parts.Moreover,a fitting method is further proposed to fit the six segmental parts to reduce the influence of signal noise.After segmentation and fitting,the lengths of the training and testing data may be different,because every part would be matched.The thesis also proposes an interpolating method to make the training and testing date have the same length.The experiments show that the accuracy of each segment's matching has been significantly improved after interpolating.(3)The thesis proposes a weighted Gaussian model based on majority voting rules to make final decisions on multi-segment results.The majority voting rule is used to decide the final test heartbeat's type.For the cases where the same number of segments cannot be judged,the weighted Gaussian model is used to judge the classification results.(4)This performance of this thesis has been evaluated.The ECG signals could be classified as N-type,S-type,V-type,and F-type according to the MIT-BIH arrhythmia database following the AAMI rules.Additionally,the data are also divided into a training set(DS1)and a test set(DS2).The experimental results are evaluated by three terms of sensitivity(Se),accuracy(Acc),and accuracy(+P).The results show that the best performance belongs to the method of Bessel curve fitting and equal length DTW matching.With the help of the decision rules proposed in the thesis,the proposed algorithm obtains the Acc,Se and +P of 98.69%,97.38% and 97.40%,respectively.