Research On Automatic Recognition Of ECG Based On Multimodal Neural Network

The electrocardiogram(ECG)is an important basis for doctors to diagnose cardiac diseases.It is a non-invasive and effective medical tool to observe the heart rhythm and state.It is a very important subj ect in the field of Cardiology to detect irregular changes of heart rhythm in human body by electrocardiogram(ECG).Traditional ECG recognition algorithms have poor accuracy and require manual design of ECG features,only then can we processed the next step to perform the classification task.But on the one hand,it requires professional medical background knowledge.On the other hand,for ECG,the features of manual design are completely based on human’s current knowledge of ECG.These features may be representative,but they are not.enough to cover all the features of ECG.With the rapid progress of in deep learning,it has shown its powerful performance in various fields.Using deep learning to solve the automatic recognition of ECG has become a research hotspot of medical researchers.Therefore,a combined model of convolution neural network and cyclic neural network is proposed to solve the classification problem of ECG.This paper focuses on ECG signals and the recognition classification of ECG signals.The main purpose is to solve the problem that feature extraction is too complex and classification effect is poor in traditional ECG recognition methods.A multi-modal neural network is proposed to complete the task of ECG classification.The process of ECG recognition is simpler and more convenient,and the recognition effect is better.Based on the analysis of traditional ECG signal recognition algorithm,this paper compares the advantages and disadvantages of single neural network in ECG signal recognition by experiments.On this basis,according to the characteristics of ECG signal,the single neural network is modified and optimized,and the following research results are obtained:Firstly,two common convolutional neural network and recurrent neural network models are designed to verify the advantages of deep neural network over traditional methods in ECG classification tasks.At the same time,we found the limitations of single convolutional and recurrent neural networks in ECG classification tasks.Secondly,a method of combining model is proposed to compensate for the disadvantage of single network,make full use of the waveforn features learnt by convolutional neural network and the temporal features learnt by recurrent neural network,and utilize the feature fusion strategy of attention mechanism to effectively fuse the wavefon-n features and temporal features to complete the task of ECG classification.The CLWA(CNN-BiLSTM With Attention)proposed in this paper is verified by experiments.CLWA has a classification accuracy of 99.54%.Compared with the traditional single convolution neural network and single recurrent neural network,it performs better in the classification sensitivity and specificity of each category of ECG.Finally,this paper designs a set of ECG automatic recognition system,which can monitor and warn real-time ECG in multi-scenarios,and has great significance for future ECG medical construction.