Arrhythmia is a common type of cardiovascular disease.Suffering from arrhythmia,the lighter will affect the quality of life of patients,and the serious ones endanger their lives.In clinical practice,doctors use medical principles and experience to analyze patients’ electrocardiograms for diagnosis.However,due to the imbalance of medical resources,the lack of professional doctors in remote areas has made many arrhythmia diseases unable to be detected and confirmed in time.Electrocardiogram analysis using automated analysis techniques contributes to the initial screening and early diagnosis of arrhythmia diseases.In the automatic analysis of ECG signals,the traditional methods mostly use the reference point feature extracted from the ECG signal to construct a classification model to identify various arrhythmia,such as by detecting the position of the QRS complex and other reference points,and extracting the ECG.The time domain,frequency domain,and statistics of the signal are used to train the model for automatic identification.This type of method is simple and straightforward,but the recognition accuracy of features such as reference points can affect the judgment of arrhythmia diseases.In the actual application environment,due to the individual differences of ECG signals and the influence of interference factors,some reference point features are difficult to accurately extract,thus affecting the automatic recognition effect of arrhythmia.(1)In view of the poor quality of ECG signals,some reference points are difficult to identify and the band feature extraction is difficult.In this paper,the sample values of the complete single cardiac cycle waveform are used as input data,and the artificial neural network model is established to identify the arrhythmia.It avoids the trouble of extracting the feature of the reference point.Through the model structure adjustment and model parameter selection,the automatic recognition accuracy of the five typical arrhythmia diseases in this model is 98%.Because time domain features are extremely important in the recognition of arrhythmia.On the basis of the original model,after adding three RR interval features,the recognition rate of arrhythmia by artificial neural network model reached 98.38%.(2)When the sampled data value of a single cardiac cycle is used as the input of the neural network,the structure of the network model will vary depending on the sampling frequency of the device.In order to solve the inconsistency of the input layer network structure,this paper images the single-cardiac periodic waveform into grayscale and binary images,and uses the convolutional neural network to construct a classification model for arrhythmia recognition.In the MIT-BIH arrhythmia data set,five sets of experiments with different scale data sets were performed,which verified that grayscale images are more suitable for automatic recognition of arrhythmia.Finally,the accuracy of automatic identification using the convolutional neural network model is 98.11%.Compared with the artificial neural network model,the model can distinguish waveforms with large morphological differences.(3)In the convolutional neural network model based on single cardiac cycle,since the periodic characteristics of ECG signals cannot be reflected in the single-cardiac periodic waveform,this paper proposes to use the continuous ECG signal segment as the input of the convolutional neural network in the MIT-BIH arrhythmia data set.After a large number of experiments,the waveforms of the sampled values at different cardiac cycle lengths were selected to establish a model for comparison experiments.Experiments show that when the band length is 400,the overall accuracy,sensitivity,specificity and positive predictivity rate of the automatic identification of the convolutional neural network model are 99.24%,99.12%,99.81% and 99.21%,respectively.Through the comparison and verification of different experiments,it is proved that the proposed strategy based on the continuous cardiac cycle-based convolutional neural network model for automatic recognition of arrhythmia is feasible and effective. |