Exploration Of Intelligent Detection Of Atrial Fibrillation Based On Wearable ECG

Atrial fibrillation(AF)is one of the most common cardiac arrhythmia diseases,with a high incidence in the elderly and a gradual trend of younger age.The incidence of atrial fibrillation is increasing year by year due to factors such as population aging and changes in dietary habits.Atrial fibrillation not only has a high prevalence and incidence,but also poses a great threat to human life and health.Early detection,early intervention,and early treatment are the most effective ways to reduce the risk of atrial fibrillation.With the development of Internet of Things technology,the application of wearable medical devices is becoming more and more extensive.Wearable electrocardiogram(ECG)monitoring devices can collect and analyze patients’ ECG data in real time,bringing great convenience to atrial fibrillation detection.This paper mainly focuses on the systematic analysis of wearable ECG signals,and the key technical in atrial fibrillation monitoring.We designed a wearable ECG signal-based detection algorithm for atrial fibrillation and paroxysmal atrial fibrillation,and developed a complete atrial fibrillation monitoring system.We aimed to enable continuous and paroxysmal AF screening for a broad range of potential patients,so as to promote the early detection,early intervention and early treatment of atrial fibrillation patients.And effectively improve the level of atrial fibrillation diagnosis and treatment.The main research contents and achievements of the paper are as follows:(1)In this paper,we firstly studied different types of noise and their effects on ECG in the wearable environment,and designed filters and preprocessing methods for different noise characteristics to reduce the noise in the wearable ECG signal.The designed pre-processing method can effectively reduce the impact of noise on the detection of QRS,especially for impulse noise.(2)We study the noise characteristics in the wearable ECG and the classification criteria of ECG signal quality assessment.To overcome the challenges brought by the uncertainty of the wearable noise to ECG signal quality algorithms,a deep learning model for signal quality classification is designed.By introducing the short connection mechanism of the residual network and the shared weight mechanism of the recurrent residual network,the model improved the depth of the model and reduced the training parameters.The model extracted features of different receptive fields by setting convolution modules with different size convolution kernels.The designed model can screen out the ECG signals that meet the standard of atrial fibrillation analysis,with an accuracy rate of 98.72%.(3)In order to improve the detection accuracy of atrial fibrillation,a deep learning-based atrial fibrillation classification model was designed.Considering that deep learning models are prone to overfitting,we proposed two training strategies to improve the robustness and generalization ability of the algorithm.The designed model achieves 97.71% accuracy of atrial fibrillation detection,and also maintain its detection performance when tested on independent datasets.(4)In order to obtain the related disease information of patients with paroxysmal atrial fibrillation,a paroxysmal atrial fibrillation localization algorithm based on RR interval rhythm was designed.Since that ECG of patients with paroxysmal atrial fibrillation often contains premature beats,premature beats screening algorithm was designed to reduce the influence of premature beats on the detection of paroxysmal atrial fibrillation.The designed premature beats screening algorithm can screen out premature beats with an accuracy rate of 96.83%,and the screening rate for single premature beat reaches 99.5%.Which effectively improves the accuracy of paroxysmal atrial fibrillation localization and the accuracy of atrial fibrillation burden calculation.