Research On Arrhythmia Prediction Based On Long Short-term Memory Network

Cardiovascular disease is a serious threat to human health due to high morbidity and mortality.ECG signals are widely used to measure the health of the heart and provide a wealth of information for the diagnosis and treatment of cardiovascular diseases.ECG is a standard tool for recording cardiac activity and is currently the most widely used in the diagnosis of arrhythmias.Current time-series signal processing for ECG usually focuses on the current moment to detect the arrhythmia that the patient may suffer,which lacks attention to early prediction of impending ECG signal abnormalities,which requires an automatic arrhythmia prediction algorithm to provide advanced auxiliary decision support for doctors.The purpose of this paper is to improve the accuracy of arrhythmia prediction.The key contents of the study are as follows:First,in order to effectively extract ECG signal features while reducing the complexity and computational complexity of the model,a method for arrhythmia prediction based on stacked long short-term memory networks is studied.The method makes full use of the characteristics of the long and short-term memory network with low computational complexity and high speed.The entire network structure adopts a two-layer stack,with 20 units in each layer to extract features from the ECG signal waveform,and select the Softmax function to complete arrhythmia prediction.The method achieves an overall prediction accuracy of 97.2% on the public dataset MIT-BIH.Secondly,because the traditional forecasting model is established for single-step time series forecasting,it is only suitable for short-term decision-making problems.In order to achieve effective monitoring of complex arrhythmias,multi-step prediction of the ECG signal sequence is also required.However,when a single neural network model is used for multi-step prediction of long time series signals,the prediction error will increase with the prediction step size,resulting in a large error between the prediction result and the original sequence.Aiming at this problem,an autoencoder-based arrhythmia prediction method is proposed.The method consists of a two-layer long short-term memory neural network.The first neural network layer is used as an encoder layer to extract and encode the features of the preprocessed ECG signal;the second neural network layer is used as a decoder layer to obtain the ECG signal features at subsequent moments through decoding,and then use Softmax to classify The device realizes arrhythmia prediction.The experimental results of this paper on the public dataset MIT-BIH show that the method effectively improves the prediction accuracy of various types of arrhythmias,and the overall prediction accuracy reaches 97.9%.Finally,because the data volume of ECG signals is relatively small compared to the neural network with a larger network scale,it is prone to overfitting,it means the accuracy of the training set is high,but the accuracy of the test set is low.Parameter tuning cannot improve the accuracy.Aiming at this problem,a method for predicting arrhythmia by adding dropout mechanism is proposed.The method realizes random probability inactivation of neural units by adding Dropout mechanism module,effectively solves the problem of overfitting,and achieves the goal of high accuracy of arrhythmia prediction.The experimental result of this paper on the public data set MIT-BIH shows that the method adopts the method of embedding the Dropout mechanism module to make the overall prediction accuracy of arrhythmia reach 98.4%,compared with the original arrhythmia prediction algorithm of stacked long short-term memory network model,the accuracy is improved by 1.2%.