Research On Disease Recognition Method Based On Deep Time Series Convolutional Network And Multi-feature Physiological Data

With people’s emphasis on medical care and the rapid development of artificial intelligence,the related research work of "Artificial Intelligence(AI)+ Medical Care" has attracted more and more attention from all walks of life.The reform of sensor technology and the application of wearable devices in daily life have laid an important foundation for the study of sensor data.Among them,biosensors can more and more accurately collect physiological signal data of the human body,which also makes AI technology quickly develop in medical research.However,most of the currently popular medical detection methods are in the direction that doctors rely on their own domain knowledge,or rely on CT image recognition,gene map recognition,and artificial extract feature of signal data collected by biosensors.For large amounts of sensor data,this kind of method is difficult to mine the potential information in the data,and it is not easy to embed it into mobile devices for real-time monitoring.It is easy to fail to find abnormal conditions in time and take effective medical measures,leading to more serious physical diseases.Therefore,it is necessary to transform the traditional artificial feature extraction method into the direct analysis of the original data to achieve real-time prediction.Based on the human physiological signal data collected by the monitor,this thesis selects the three physiological signals of Arterial Blood Pressure,Electrocardiographic,and Respiration that can best represent the physical condition of the person as training data,and a hybrid method of deep convolutional layer and long short-term memory is proposed to study it and analyze the time dependence between them to achieve the purpose of disease recognition.The thesis’ s main work is as follows:(1)This thesis introduces the research background and significance of this topic,deeply analyzes the current research status of artificial intelligence in the field of disease recognition at home and abroad,and summarizes the shortcomings of existing methods,so as to put forward the main research content and innovation of this paper,focusing on the data source and relevant theoretical knowledge involved in the experiment.(2)Most problems in the field of physiological signal recognition are that researchers usually focus on a certain feature,ignore other influencing factors,and use the method of manually extracting features of time series.Considering various physiological signal data(arterial blood pressure,electrocardiogram,respiration)closely related to the human body,this thesis proposes a multi-feature physiological signal recognition method(Deep CNN-LSTM)based on deep convolutional layers(CNN)and long short-term memory networks(LSTM).This method first uses a stationary wavelet transform and a median filter to denoise and correct the baseline for low-frequency signals,then uses the sliding window method to obtain the short-time fragment sequence set,and uses CNN to obtain the feature set by studying the sequence set,finally uses LSTM to learn the time dependence between the feature sets to get the disease recognition rate.(3)The recognition effect of the Deep CNN-LSTM hybrid model is analyzed on the three kinds of human physiological signal data collected in real,and the performance of different models is compared and analyzed based on the experimental evaluation indicators.