Portable ECG Signal Acquisition And Intelligent Analysis System Based On Machine Learning

Recently,the mortality rate of cardiovascular disease is the highest among all diseases.Cardiovascular disease can lead to abnormal heart rate signals.Therefore,the analysis of heart rate signal and electrocardiogram(ECG)has become the main method to detect cardiovascular diseases.At present,the ECG signal acquisition instrument applied in the market is not portable,and the data processing adopts manual analysis,which cannot solve the problem of ECG signal analysis under large data stream.To solve these problems,the ECG acquisition and analysis system based on machine learning is considered in this study.The ECG acquisition system is designed and manufactured to improve the portability of the current acquisition scheme,and AI-assisted ECG analysis system is utilized to analyze large amounts of ECG data.The research content of this thesis consists of two parts: the ECG acquisition system responsible for collecting ECG signals and the ECG analysis system responsible for classifying ECG signals.The ECG acquisition system is responsible for the acquisition of ECG signals,including the acquisition front-end part and the analog-to-digital conversion part.The main function is to collect,process and transmit ECG signals.The front-end of the acquisition adopts AD8232 chip as the signal conditioning module,which is used to amplify and filter ECG signals.The analog-to-digital conversion part is responsible for converting ECG analog signals into digital signals.STM32F103 is served as the control chip to collect signals,perform analog-to-digital conversion and realize digital filtering.Finally,the circuit communicates with the computer terminal via bluetooth,which collects data,reads and analyzes data.The ECG analysis system responsible for classification of ECG signals includes data preprocessing part and machine learning model part.The data preprocessing part needs to realize the function of removing noise,locating the QRS waveform and dividing the heart beat,constructing the feature engineering,and dividing the training set and test set.Wavelet transform is utilized to de-noising the data.Double slope method is served as QRS waveform extraction,and morphological feature method and wavelet coefficient method are mainly used in feature engineering.The MIT-BIH database and the collected ECG data are randomly shuffled.The models mainly consider traditional models,including logistic regression,support vector machine and XGBOOST,and deep learning models including convolutional neural network and long short-term memory network.After training,the models are fused with stacking method.The parameter optimization is adjusted according to the experimental results of 5-fold cross validation.The classification accuracy of the final fusion model is 99.13%.Meanwhile,a new CNN model is constructed to realize the classification of ECG signals under the AAMI standard,and the final accuracy is 99.16%.In this study,the portability of ECG signal acquisition is improved through a self-made embedded system,and signal preprocessing and signal classification are performed on ECG signals by combining filtering algorithm and feature extraction algorithm with machine learning models.An ECG signal acquisition and analysis system based on machine learning is successfully implemented to collect,process and classify ECG signals.