Application And Research Of Deep Learning In Long-Term ECG Monitoring And Atrial Fibrillation Analysis

Atrial Fibrillation(AF),which characterized by rapid and irregular activation of atrial,is one of the arrhythmia that is most common and serious in clinic.People may have no obvious feelings,and may also be confused,numb and uncomfortable when AF occurs.Particularly,AF patients have five times risk of stroke than normal people,and AF is also associated with heart failure,disability and death.Therefore,people’s quantity of life will severely affect once suffering AF.Inaddition,the number of people with AF is extremely large because the prevalence of AF increase with age.AF is rare before the age of 50,but about 10% of people are diagnosed with AF by the age of 80.With the aging of the population,the diagnosis and treatment of AF is very important.AF is often diagnosed by electrocardiogram(ECG).The ECG characters of AF patient is sigficantly abnormal when AF occurs,such as the R-R intervals is irregular and P-wave transform into rapid atrial wave.With the increase in the number of AF patients and the development of intelligence collection equipment,the need of long-term ECG mornitoring are continues to increase,and the pressure on clinicians are also huge.Therefore,method of AF automatic detection by computer is hot topic.It is great significance to detect AF by computer automaticly,such as giving the doctor a primary screening effect,monitoring people out-of-hospital after treating AF and monitoring healthy people in real time in order to find symptoms in time for medical treatment.In this paper,three works were done based on long-term ECG monitoring and deep learning technology:1.Long-term ECG monitoring and data collecting from clinic AF patient was completed with intelligence wearable device,and noise filtered and R-wave located was carried out for AF data.Especially further noise cleaned was carried out for AF data that easily bring noise in daily life with long-term wearable devices,2.Proposed a novel method based on convolutional neural network(CNN)and ECG atrial and ventricular conduction characters characters for AF automatic detection.That is,combined with the electrophysiological mechanism of atrial fibrillation,R-R intervals and F-wave frequency spectrum were extracted as the input of CNN to classify AF automaticlly.Performance evaluation was carried out with international standard ECG database and clinic AF data,and the sensitivity,specificity and accuracy of 97.4%,97.2% and 97.3%.the result shows that the performance of AF diagnose with atrial and ventricular conduction characters characters is higher than the performance that diagnose with either one character,respectively.And the result shows that the two characters can complement each other in performance and effectively improve the accuracy of AF automatic detection.3.Initially,discussed that fine-tune the convolutional neural network with a few patient-specific ECG data to improve the performance of AF detection in long-term monitoring.Additionally,heart rate variability parameters were combined to analysis different types AF to provide technical guidance for radiofrequency ablation in preoperative screening and postoperative follow-up.Overall,an acceptable result of AF automatic detection was obtained by the novel nethod with CNN and atrial and ventricular conduction characters,and it confirmed that atrial and ventricular electrophysiological conduction mechanism and the corresponding ECG characteristics are independence diagnosis basis.The two characters can complement each other in performance and effectively improve the accuracy of AF automatic detection.However,limited by the time of the study and the shortage of existing collection equipment and number of patients in long-term clinic data analysis,only preliminary exploratory analysis was carried out.In addition,large clinic data test and statistical evaluation are required in future for further conclusion.