Atrial Fibrillation Detection Algorithm Based On CNN-LSTM Model

With the rapid development of China’s social economy and the gradual acceleration of the aging trend of the population,the incidence and mortality of heart disease have shown a trend of increasing.Atrial fibrillation(AF)is one of the most common heart diseases in the clinic.The disease has a short onset time and no obvious symptoms in the early stages.The treatment of AF often costs a lot of medical expenses and takes up a lot of social resources due to missing the best treatment opportunity.Therefore,the research on the intelligent detection algorithm of early atrial fibrillation has important theoretical and practical value for optimizing the detection process and improving the detection efficiency.In order to solve the shortcomings of the feature extraction in the existing atrial fibrillation detection research,the Convolutional & Long Short-Term Memory Neural Networks(CNN-LSTM)is proposed based on the P wave disappearance and heart rhythm disorder of the AF ECG signals.The CNN-LSTM model consists of two main parts: the feature extractor composed of convolution unit and LSTM unit,and the classifier composed of full connection layer and output layer.The convolution unit can excavate structural features such as the disappearance of P wave and waveform disturbance in AF signal.The LSTM unit is used to extract time series features such as heart rhythm disorders.The multidimensional features are summarized by the fully connection layer and the detection results are output through the output layer.This paper conducted several rounds of model classification performance verification experiments with ECG signals from the Physinet/Cin C Challenges 2017 dataset,clinical diagnostic features,and 169 multidimensional mathematical hybrid features,as well as four traditional machine learning algorithms of k-nearest neighbor algorithm,support vector machine,decision tree and random forest and two deep learning algorithms of convolutional neural network and long and short-term memory neural network.The results show that the accuracy,sensitivity and specificity of CNN-LSTM model reach 96.75%,97.40% and 96.10%,respectively,and the score is 0.97.The performance of the model is improved by 9.54% on average compared with other comparison algorithms.After embedding CNN-LSTM model into the mobile terminal,self-detection and auxiliary diagnosis of early AF can be realized,which can reduce the delay of patients’ illness and greatly improve the working efficiency of doctors.