Research On The Classification Methods Of Heart Rate Abnormalities Based On BCG Signal With LSTM

Cardiovascular disease is the "number one killer" that affects human health,and it has the characteristics of high fatality rate and high incidence rate.And heart rate is one of the most commonly used parameters for clinical evaluation and description of physiology and pathology,meanwhile it has be widely used as indicators for evaluating cardiovascular status.Therefore,early diagnosis is of great importance in cardiac abnormalities,which can decrease the risk and incidence of cardiovascular.Arrhythmia is a group of condition in which the heartbeat is irregular.Currently,conventional techniques of diagnosis is to perform an electrocardiogram test on the patient,then make judgments relied on the doctor’s professional knowledge and experience.However,the ECG signal varies from person to person,it is quite difficult to detect early period of irregular cardiac rhythm,misjudgments and missed judgments would be happen.Using algorithms to mine the inherent characteristics of heart rate signals and automatically classify heart rate abnormalities is of great significance for clinical medical diagnosis and self-monitoring.Based on the above analysis,this article mainly carried out the following work:1.Preprocessing of BCG.In this paper,using the heart rate detection method of BCG,35 elderly people had been monitored and collected heart rate signals for a period of one year.The collected heart rate data was smooth preprocessed,and the heart rate data was marked with binary classification and multiple classification.2.Based on some existing machine learning methods for heart rate abnormalities classification.Regardless of the influence of time on the sample data,three traditional machine learning methods,such as support vector machines,K-value proximity,and multi-layer perceptrons,are used to conduct binary classification and multiple classification experiments on the collected ECG data.Research and experimental results show that the classification accuracy rates of support vector machine,K-value proximity,and Multilayer Perceptron are 83.47%,92.54% and94.07% on the binary classification problem,respectively.On the multiple classification problem,the classification accuracy rates of the three methods are61.74%,69.59% and 73.1%.3.Heart rate abnormal classification method based on LSTM.Due to the time continuity of the collected BCG,considering the inherent relationship of the sample data in time,a long short-term memory network(LSTM)is proposed to classify abnormal heart rate data.The experiment explored the classification effect of LSTM on the binary classification and multiple classification problems,and compared it with the classification effect of three traditional machine learning methods.The experimental results show that LSTM has certain advantages in both the binary classification and the multiple classification problems.