Analysis Of Temporal Physiological Signals For Heart Disease And Sepsis

Temporal physiological signals play an important role in human health assessment and disease diagnosis.Due to there are many types of temporal physiological signals,in order to focus on targeted research,this thesis mainly analyzes and discusses two typical temporal physiological signals and their applications:(1)Use deep convolutional neural network to analyze ECG signals and classify different heart diseases.(2)Analyze various physiological measurement data collected in a continuous time and use these data to make early prediction of sepsis.ECG signal is a kind of temporal physiological signal,which reflects the bioelectrical changes of cardiac muscle cells.Although ECG signals are widely used in clinical diagnosis of cardiovascular diseases,the diagnosis and screening of cardiovascular diseases is still a difficult problem because the diagnosis results of ECG signals are relatively subjective.While computer-aided ECG diagnosis technology is objective and easy to popularize,which is one of the solutions to this problem.This thesis proposes a ECG signal classification method based on one-dimensional convolutional neural network.The residual attention module in this network not only alleviates the problem of the vanishing gradient but also extracts the useful features of the ECG signals in the spatial domain through the attention mechanism.At the same time,in order to make full use of the time correlation of ECG signals,the network constructs a Long Short-Term Memory(LSTM)module to extract the temporal characteristics of ECG signals in the time domain.The experimental results prove that the one-dimensional convolutional neural network proposed in this thesis has achieved good results in ECG signal classification and has certain robustness.The experimental results prove that the ECG classification method based on convolutional neural network proposed in this thesis is better than other methods and has certain robustness.The physiological measurement data collected during the duration is a typical temporal physiological signal,reflecting the changes of various physiological indexes of the human body with time.It is an effective method of disease detection and diagnosis to model and analyze the continuous physiological measurement index.This thesis takes the early prediction of sepsis as an example,and uses a physiological measurement database of sepsis patients provided by Physio Net / Cin C 2019 to propose an early prediction method of sepsis based on XGBoost.Because the categories of this database are imbalanced and there are a large number of missing values,the method preprocesses the data first,and uses XGBoost to analyze the continuous physiological index of the patients,so as to predict whether patients have sepsis.In addition,this thesis also explores whether LSTM can extract temporal features from physiological signals to predict sepsis.Experiments show that the method based on XGBoost is better than the method based on LSTM.Therefore,the method for early prediction of sepsis based on XGBoost is the final work of the Physio Net / Cin C 2019 Challenge,ranking 15 th in more than one hundred teams worldwide.