Research On Cardiovascular Disease Recognition Based On Deep Learning

With aging populations,changing diets and increasing the number of mental worker in our country,the morbidity and mortality of cardiovascular disease,which has been become a major cause of harming human health,increased year after year.Some physiological signals are changed in the early phase of cardiovascular disease.And if the changed physiological signals were recognized,the occurrence of cardiovascular disease can be early warned and patient's treatment time is saved.However,many people in our country are reluctant to have regular medical examinations and ignore themselves early lesions caused by cardiovascular disease to miss the best time for early treatment,because of unevenly distributed medical resources and difficulty and high cost to access medical service in most areas.To cure the above problems,the model of cardiovascular disease recognition based on deep learning in unsupervised environment outside the hospital is built by using patient's own attributes,the ECG and pulse signals that are being widely used in portable medical devices.The main work is as follows.1)The research background of cardiovascular disease recognition was outlined.And the problems of the cardiovascular disease diagnosis methods of function,medical imaging and physiological signal in research at home and abroad were pointed out.Thus the importance of cardiovascular disease recognition in unsupervised environment outside the hospital was highlighted.In addition,the feasibility of cardiovascular disease recognition based on deep learning in unsupervised environment outside the hospital is analyzed through introducing the research status about deep learning in medical and other fields.2)The rationality of the physiological signals are used to cardiovascular disease recognition was demonstrated.For one thing,from the bioelectrical and hemodynamic principles of the cardiovascular system and the pathogenesis of cardiovascular diseases,the rationality of ECG and pulse signal are used to the cardiovascular diseases recognition was demonstrated.For another,the analysis of heart rate variability,pulse rate variability and pulse transit time variability for ECG and pulse signals were introduced to lay a theoretical foundation of cardiovascular disease in unsupervised environment outside the hospital.3)The wavelet analysis to extract the peak of R wave of ECG and the difference-wavelet analysis to extract the peak of P wave of pulse signal were put forward.Firstly,the flaw of existing peak detection algorithm was explained based on the normal and pathological features of ECG and pulse signal.Then,for the needs of cardiovascular disease,corresponding peak extraction algorithm was put forward.Finally,the accuracy of the algorithm and the feasibility of these algorithm are used to physiological signal variability extraction were verified by used of ECG and pulse signal in different environment.4)For the quasi-periodic,non-stationary,non-linear features of physiological signal variability,the analysis of time,Poincare plot and entropy were chosen to analyze the physiological signal variability.And the significance and feasibility of the features in cardiovascular disease diagnosis were demonstrated.Final,feature vectors(total 50),are formed by the 48 features(16 kinds)and age and sex of patient,were used to cardiovascular disease recognition.5)The model of cardiovascular disease recognition based on deep learning in unsupervised environment outside the hospital was built.Firstly,the deep learning algorithm was given rise through explaining the gradient descent and its flaw of BP algorithm.And the model structure and training process of deep belief network(DBN)were introduced.Then,the Setting mechanism of DBN's parameters was analyzed from the theory of DBN.And DBN's parameters were set according to empirical formula and the characteristic of cardiovascular disease data.Finally,the model of cardiovascular disease recognition based on deep learning in unsupervised environment outside the hospital was built.The usability of model and the rationality of parameter selection were verified by used of public data set.