Research On Physiological Data Based On Persistent Homology Algorithm

Cardiovascular disease is a disease of the heart or blood vessels.Cardiovascular diseases continue to bring people more and more heavy burden.The number of deaths from cardiovascular and cerebrovascular diseases worldwide exceeds 15 million annually,ranking first among all diseases.Therefore,a simple,fast,and automated system is needed that can analyze a large number of physiological signals in real-time and quickly,in order to prevent cardiovascular diseases in advance.Traditional large-scale devices can help identify symptoms or causes of cardiovascular disease,as well as identify abnormal heart rhythms.However,these devices are bulky and difficult to operate,and cannot achieve real-time monitoring.With the development of artificial intelligence,wearable physiological signal detection devices have been improved.These wireless transmission devices have small size,simple operation,low price,and can record physiological signals for a long time without damage.However,people are often in a state of motion and uncontrollable factors such as equipment and environment,resulting in significant noise in the recorded signal quality,which will affect the performance of computer-aided analysis technology.24-hour real-time monitoring will generate a massive amount of data.If the data is directly handed over to doctors,it will bring a heavy workload to doctors.The automatic processing and analysis of wearable physiological signals is the key to the application of wearable devices in real-time detection of clinical cardiovascular systems.How to select high-quality signals from massive data,extract temporal and spatial features of high-dimensional physiological signals,and achieve real-time analysis of wearable physiological signals are key issues that urgently need to be solved for real-time monitoring of cardiovascular systems.Persistent homology is an important research direction in topological data analysis.Its core idea is to study the topological structure of filtering scales and analyze the changes in Betty numbers in various dimensions.During the process of filtering scale changes,the topological features with shorter durations are noise,while the features with longer durations are used to represent their intrinsic characteristics.Topological data analysis has been applied to different fields,such as computational geometry,bioinformatics and medicine,but it has not yet been applied to the analysis of wearable physiological data.This study applies the persistent homology method to wearable dynamic electrocardiogram signal quality assessment,arrhythmia classification,and classification of heart sound signal diseases.The research process of this experiment: Firstly,it is necessary to construct a point cloud matrix of physiological signals through sliding windows,and then generate persistent barcodes and diagram through complex filtering.Finally,experiments are carried out based on Goog Le Net transfer learning classification model.This study used four datasets to classify wearable physiological signals,and the results showed that this method has high accuracy and operability.In the 2011 dataset,the 12 lead dataset and the single lead dataset constructed in this study showed good performance.The highest classification performance was achieved in the 12 lead dataset with m Acc=98.04%,F1=98.40%,Se=97.15%,Sp=98.93%,and in the single lead dataset with m Acc=98.55%,F1=98.62%,Se=98.37%,Sp=98.85%.In order to investigate the practicality of the persistent homology method,this study will validate the performance of the method in the 2017 Physio Net/Cin C Challenge Cup,the 2016 Physio Net/Cin C Challenge Cup heart sound dataset,and a publicly available heart sound signal disease dataset.In the 2017 dataset,the highest classification performance was achieved with m Acc=98.55%,F1=98.62%,Se=98.37%,Sp=98.85%.In the 2016 heart sound dataset,the highest classification performance was achieved with m Acc=99.68%,Se=99.81%,Sp=99.55%,and F1=99.84%.The highest classification performance of m Acc=98.6%,m Se=98.66%,m(+P)=98.6%,and m F1=98.6% was obtained in the publicly available heart sound dataset.Based on the analysis and comparison of these data,the analysis results show that the classification performance of the persistent homology model is better than previous algorithms.Compared to methods such as principal component analysis and clustering analysis,this method not only effectively captures the topological information of large-scale high-dimensional data network layer space,but also excels at discovering information that may not be discovered using known traditional methods.In the study,the corresponding physiological data are classified using persistent homology algorithm combined with artificial intelligence methods such as Transfer learning,which provides a new way for the analysis of physiological data.