Research And Implementation Of Arrhythmia Classification Based On Neural Networ

Cardiovascular risk factors have significant impacts on human health,and the incidence of cardiovascular diseases is continually increasing.Arrhythmia is considered to be the precursor of cardiovascular disease,or often accompanied by cardiovascular disease.Therefore,arrhythmia is the key to the prevention and treatment of cardiovascular disease.The electrocardiogram(ECG)is an intuitive reflection of the electrical activity of the human heart and is currently an important basis for the diagnosis of cardiovascular diseases.With the continuous development of artificial intelligence and cloud computing technology,the research of ECG monitoring mobile devices based on wearable technology and deep learning method has become a research hotspot.The work of this thesis is carried out in combination with the research contents and research objectives of Tianjin Key Research and Development Program "ECG monitoring popular science application development based on flexible intelligent wearable technology" and Tianjin Science and Technology Program "ECG sensing and diagnosis research based on capacitive coupling fabric electrode".Taking arrhythmia classification and diagnosis as the research object,a single lead arrhythmia classification algorithm suitable for Wearable ECG detection system is designed based on convolution neural network(CNN)and bidirectional long short-term memory(BiLSTM)neural network theory.On this basis,the interaction between signal segmentation length and arrhythmia classification diagnosis results is analyzed,and the influence of deep learning model parameters on classification diagnosis results is explored and tested.The research results have potential application prospects for the prevention and diagnosis of heart diseases.The innovative research results obtained in this thesis are as follows:1.An arrhythmia classification algorithm based on CNN-BiLSTM hybrid model is proposed.Based on the deep learning theory,the algorithm comprehensively utilizes the respective advantages of CNN and BiLSTM models to extract the multilevel features of ECG signals,and realizes the high-accuracy arrhythmia classification of 8 ECG signals: normal beat,left bundle branch block,right bundle branch block,atrial premature beat,ventricular premature beat,atrial fibrillation,sinus bradycardia and ventricular tachycardia,the accuracy can reach 98.86%.The classification effects of CNN-BiLSTM with support vector machine,k-nearest neighbor network,decision classification tree,LSTM and BiLSTM models are analyzed and compared.The results show that the designed CNN-BiLSTM has greatly improved the accuracy,precision and sensitivity,and is more suitable for the classification and diagnosis of arrhythmias.2.An optimization algorithm of BiLSTM and CNN-BiLSTM models based on Bayesian optimization is proposed.Based on Bayesian theory,the algorithm explores the effects of initialization learning rate and the number of BiLSTM hidden layer units on the classification performance of the model.The classification accuracy of optimized BiLSTM and CNN-BiLSTM is improved by 0.48% and 0.23% respectively compared with that before optimization.The experimental results show that the optimization algorithm improves the performance of the model and avoids the blindness of parameter setting,which leads to the suboptimal state of the model.On this basis,the effects of different ECG signal lengths on the arrhythmia classification performance of CNN-BiLSTM are studied.The results show that the classification effect of the model is the best when the length of ECG signal is 800.3.A wearable ECG monitoring system based on "acquisition terminal-mobile app-cloud platform" is designed.The CNN-BiLSTM arrhythmia classification algorithm proposed in this thesis is used to realize the remote auxiliary diagnosis of cardiovascular diseases.The accuracy of the simulator and the physical are 90.65%and 90% respectively.The experimental results show that the arrhythmia auxiliary diagnosis algorithm can better complete the classification and diagnosis of ECG signals,and further verify the feasibility of the algorithm.