A Study Of Personalized Health Status Recognition Based On Multivariate State Estimation Technique

Chronic diseases pose a major threat to human health,and the problem of chronic diseases in China is becoming increasingly serious.For the management and treatment of chronic diseases,the assessment of the human body’s health status is particularly important.Physiological signals such as heart rate,respiration,and body temperature contain abundant personalized health and disease information.Meanwhile,the development of wearable devices allows people to continuously collect human physiological signals with low load,and the deep mining of high-value information as well as personalized health status recognition based on continuous physiological signals has become a hot research topic.However,the human body is a system with high complexity in dynamic equilibrium.One of the major difficulties in analyzing continuous physiological signals collected in real life is that physiological signals exhibit highly individualized characteristics.Multivariate State Estimation Technique(MSET)is an individualized nonparametric modeling method based on historical data,which can largely erase individual differences in physiological signals by analyzing the estimated residuals of the observed data to achieve abnormal state alarms.The purpose of this study is to conduct deep mining of continuous physiological signals collected by wearable devices based on MSET,explore the methods of data screening and data pre-processing based on the characteristics of human physiological signals,and attempt personalized modeling for the recognition of human health status,thus providing a possible solution for the health management of chronic patients.In this paper,the researchers first validated the performance of the wearable device and developed signal quality assessment algorithms for data screening to address the problem of complex noise and motion artifacts in physiological signals,promoting the study of MSET.Next,researchers introduced the theory of MSET in detail and designed the signal preprocessing and feature processing process.A total of six features including mean heart rate,heart rate range,heart rate standard deviation,mean respiratory rate,respiratory wave interquartile range and mean activity level were extracted from the physiological signal for MSET modeling,and the residual was defined as the difference between the actual value of each observed parameter and the estimated value obtained by MSET.The high-dimensional residual series were fused into a Multivariate Health Index(MHI)using a Gaussian mixture model.Finally,the researchers designed three progressive experiments to validate the algorithm: 1.The researchers added increasing artificial noise to a section of physiological signals of heart rate to test the MSET algorithm’s ability to recognize the noise;2.24 volunteers were included to study MSET’s ability to recognize acute hypoxic stress status;3.A retrospective observational study was conducted in 17 chronic patients with coronary artery disease combined with a tendency to heart failure whose Brain Natriuretic Peptide(BNP)had changed significantly during hospitalization to investigate the association between the difference in the residual distribution of the signals and the amount of change in BNP test results.The results show that the wearable device is able to collect signals stably even in complex environments,and the signal quality assessment algorithms are able to effectively screen out the signal segments for the follow-up study.The validation experiments indicate that MSET is sensitive to abnormal fluctuations in the observed parameters of physiological signals,and even small fluctuations can be captured.When the linear noise intensity is about 9%(6 bmp),the actual value of the average heart rate is significantly outside the estimated interval of the algorithm to realize the early warning.The algorithm can effectively identify the acute hypoxic stress state in humans,and the algorithm performs better with Gaussian operator as a nonlinear operator and when L2 regularization is performed,and the median(Q1,Q3)AUROC(Area Under Receiver Operating Characteristic Curve)(%)of the model are 76.39(63.20,87.19)and 76.45(67.02,87.22).In addition,the distribution of residuals estimated by MSET had some regularity,in which the Pearson correlation coefficients between SMD(Cohen Standardized Mean Difference)and OVL(Overlapping Coefficient)of MHI and the amount of BNP outcome change reached 0.786 and 0.835,with p-values less than 0.001,indicating that there is a link between the difference in residual distribution and the amount of change in BNP examination results,and it is preliminary demonstrated that the model can reflect the degree of change in human health status to some extent.The value and innovation of this paper are to further promote the clinical application of wearable devices,explore the key technology of MSET in the field of human physiological signal information utilization,and propose a method to measure the degree of health change of patients with chronic diseases based on MSET.This paper focuses on the characteristics of physiological signals,data pre-processing methods,algorithm principles and the pattern of residual distribution,which is believed to be useful for similar research on complex continuous physiological signal information mining and human health state assessment.In summary,MSET provides a new individualized way to analyze physiological signals,which can identify small changes in physiological signals and then measure the health status of the human body,providing additional decision support information to physicians.Potential application scenarios for this algorithm include: early warning of disease deterioration,chronic disease management and health status assessment,and treatment effectiveness evaluation.