Study On Heart Rate Variability Based Atrial Fibrillation Detection Method And Lightweight Algorithm

Atrial Fibrillation is a kind of cardiac arrhythmias commonly seen in the clinic,which is closely related to coronary artery disease and heart failure.Heart rate variability(HRV)is an important clinical indicator for monitoring the status of cardiac activity and can be extracted from signals collected by physiological signal acquisition devices such as electrocardiogram and photoplethysmography.RR intervals can be obtained from R wave detection using physiological signals which reflecting heart activity,and heart rate variability features can be extracted and analyzed from RR interval.The time series composed by intervals of heartbeats is an efficient measure for atrial fibrillation monitoring and preventing.Compared to atrial fibrillation detection algorithms based on physiological signals such as electrocardiogram,the algorithm which only relying on HRV features to detect atrial fibrillation has the advantages of low acquisition cost and high model generalization.The main contents of this documents are as follows:(1)Electrocardiogram signal dataset for atrial fibrillation from CPSC2021 challenge is processed,and short time RR interval time serries dataset is built by extracting R wave spikes.Next,the atrial fibrillation classification results of popular mainstreaming models are verified in the newly built RR interval dataset of this model,and the classification accuracy of deep learning models such as bidirectional LSTM reached more than 98%.The classification effect of atrial fibrillation using only HRV was close to that of atrial fibrillation classification using the complete electrocardiogram waveform,demonstrating the feasibility of using the HRV signal instead of the complete waveform signal such as electrocardiogram and photoplethysmography for atrial fibrillation detection screening.Extraction of R wave spikes.(2)This paper investigates the lightweight and practical improvement of the HRVbased RR interval time-series AF detection model.To address the problem of short time paroxysmal atrial fibrillation,this paper shortens the interval used to classify RR intervals to 30 R wave spikes,which reduces the time required for acquisition to less than half a minute and solves the problem.In order to calculate R waves quickly on portable devices,in locating R-wave spikes this paper takes suitable for deployment to embedded devices.To reduce the performance consumption of AF detection,this paper proposes lightweight model based on improved time series forest algorithm for AF detection that substantially reduces the performance requirements of AF detection without losing a significant amount of accuracy compared with models such as the bidirectional LSTM.