Clinical Analysis Of Clinical ECG Data Based On Convolutional Neural Network

Cardiovascular disease is the highest morbidity and mortality in China.At the present stage,clinical ECG data are often used for examination and good results have been achieved.With the increase in the demand for real-time detection of cardiovascular diseases under large scale population conditions,this diagnostic method has relied on the problems of physicians’ personal experience and diagnosis level,long time-consuming,and low working efficiency.The automatic analysis of clinical ECG data by artificial intelligence has important theoretical significance and application value in solving the above problems.The existing ECG data automatic analysis technology is based on the characteristics of P-QRS-T waves extracted from ECG data,and then classifies and diagnoses by using prior knowledge or machine learning method.An automatic analysis method based on prior knowledge uses the clinician’s clinical experience to set classification criteria for diagnosis.This method has achieved good effect for specific types of cardiovascular diseases.The automatic analysis method based on machine learning uses the shallow neural network to classify and recognize the characteristics of the P-QRS-T wave of ECG data,which improves the generality of the method.These two methods have achieved good results for single lead or double lead experimental ECG database.With the popularity of large-scale ECG acquisition equipment and dynamic electrocardiogram,the scale of ECG data has been further increased,and the researchers have gradually established the 12 lead clinical ECG database.The clinical ECG database not only has a much larger amount of data than that of the experimental ECG database,but the uncertainty of clinical collection also greatly increases the complexity of the data.These factors lead to the difficulty in the selection of clinical ECG data,which seriously affect the clinical use of the existing ECG analysis methods.This article aims at the characteristics of large volume and high uncertainty of clinical ECG data,improves data preprocessing methods,and uses convolutional neural networks to build models and perform analysis.This method does not need feature extraction and directly classify and recognize ECG data.It can effectively retain implicit information in ECG data and improve the recognition rate of clinical ECG data.The specific research contents are as follows:(1)Preprocessing of clinical ECG data.According to the correlation between the 12 lead ECG data,the method of multi-lead de-noising is used to eliminate the exception of the single lead and improve the signal to noise ratio;the high-pass filter and the band stop filter are used to remove the baseline drift and power frequency interference;The second-order blind source separation method is used to remove myoelectric interference and solve the problem that the original wavelet transform method is difficult for the selection of wavelet basis when facing clinical ECG data.(2)Automatic analysis of single lead ECG data.When the amount of input data is large,shallow neural networks need to increase the number of hidden layer units to improve the classification accuracy,which brings about a large amount of calculation and difficult to solve the problem.The depth convolution neural network is chosen to achieve better results with smaller calculations by increasing the depth of the layer.Using the convolutional neural network to establish the model,the single lead MIT-BIH database was automatically analyzed,and the five classification accuracy rate was 98.67%.(3)Automatic analysis of clinical 12 lead ECG data.The 12 lead clinical ECG data are independent from each other.When using convolutional neural network modeling,an independent convolution unit is added to improve the convolution fusion calculation method;With the increase of the number of convolution neural networks,the error transfer gradient is reduced.The residual network should be introduced to improve the feedback effect.Using the improved convolutional neural network,150,000 sets of clinical ECG data were automatically analyzed.The accuracy of the two-category was 90.46%.