Research On Diagnosis Model Of Coronary Heart Disease Based On Physiological Time Series

Coronary heart disease is one of the major diseases that threaten human health.Clinical diagnosis of coronary heart disease mainly depends on indirect diagnosis methods such as manual interpretation and biochemical indicators.This method has low diagnosis efficiency and strong subjective factors.So the study of the diagnosis model of coronary heart disease based on physiological time series is of great significance and can assist doctors in diagnosis.However,the existing research has many deficiencies,such as signal analysis relies on human judgment,the use of data for coronary heart disease classification models is mostly single-lead electrocardiogram(ECG)signals and clinical use of multi-lead ECG,and there is currently no good quantitative diagnostic indicator that can compare the development of the disease.In summary,this paper uses signal analysis technology,deep learning and dynamic complexity to study the diagnosis model of coronary heart disease based on physiological time series.The main research contents are as follows:First,an empirical mode decomposition convolutional neural network(EMDNet)is proposed based on single-lead ECG signals to achieve coronary heart disease classification.Because the ECG of healthy people and patients with coronary heart disease are significantly different in the time-frequency domain,this paper uses empirical mode decomposition to obtain the time-varying rhythm in different frequency bands,and uses the attention mechanism to learn the more critical time-frequency domain representation.The two together constitute a learning mechanism for time-frequency domain feature representation;then use convolutional neural networks for classification.Therefore,the model can adaptively learn the signal feature representation without human interference.Secondly,based on the multi-lead ECG signal,a multivariable empirical mode decomposition convolutional neural network(MEMDNet)is proposed to classify coronary heart disease.Considering that clinically used multi-lead ECG contains more information than single-lead ECG,this paper uses multivariable EMD to solve the problem of non-corresponding spectrum bands of signal decomposition,and combines multivariable EMD with the attention mechanism to form a learning mechanism of time-frequency-space feature representation.The mechanism can dig deeper into the more critical feature representations of multi-lead ECG in space,time and frequency dimensions.Therefore,the model can be extended to clinical use in the future.Thirdly,the diagnostic indicators for the development of coronary heart disease based on dynamic complexity theory were constructed: mean multi-scale entropy (?) and complexity-activity correlation coefficient.This paper designs (?) that based on the dynamic complexity theory,and the larger the value is,the higher the health level is;this paper constructs the complexity-activity correlation coefficient which combined with the amount of human activity reflected by the triaxial acceleration,and the smaller the absolute value is,the stronger the adaptability of human body to different activity levels is.Therefore,the constructed evaluation index can dynamically reflect the changes of the patient’s overall health level.Finally,experiments verify the constructed models.On the one hand,the EMDNet,MEMDNet and existing models proposed in this paper are compared and analyzed in public data sets.The results show that the proposed models have achieved accuracy rates of 99.51% and 99.80% respectively,and the classification accuracy rates of the seven ECGs exceed 99.44%.They are both higher than existing models.On the other hand,the constructed (?) and complexity-activity correlation coefficient are applied to the collected clinical data and compared with biochemical indicators and doctor evaluations.The results show that there are strong correlations between the constructed indicators and the biochemical indicators of clinical judgment(The absolute values of the pearson correlation coefficients both exceed 0.8),and the evaluation results based on hourly particle size and day particle size are consistent with the clinician’s evaluation.Therefore,the models proposed in this paper can provide new ideas for the related research on the diagnosis of coronary heart disease and the diagnosis of disease development in the future.