Research On Atrial Fibrillation Detection And Identification Method Based On Photoplethysmography

Atrial fibrillation(AF),abbreviated as atrial fibrillation,is a serious and common arrhythmic disease,which can cause dizziness,blackness,and syncope when patients onset,which in turn causes serious damage to heart function and brain function,and even takes the patient’s life.The diagnosis of atrial fibrillation mainly relies on observing the patient’s electrocardiogram for confirmation,and it takes a professional doctor to spend a lot of time to check the electrocardiogram to be able to correctly assess the patient’s condition.ECG machine operation is difficult,inconvenient to carry,high cost,pulse wave signal and ECG signal has a strong correlation,and pulse wave signal collection is simple and convenient,so pulse wave signal is more suitable for early identification and monitoring of atrial fibrillation.Based on the above reasons,this paper proposes a method for atrial fibrillation recognition based on photoplethysmography(PPG)signal,and constructs two convolutional neural network models with different inputs based on deep learning methods,both of which achieve good classification effects,the main contents of this article are as follows.(1)A set of pulse wave signal acquisition device is designed,which is mainly used for the collection of PPG signals on human fingers.By designing experiments,volunteers were asked to wear a watch that detects atrial fibrillation when collecting finger pulse waves,and label the data with the results of the watch’s detection.The pulse wave signal from the MIMIC-III database and the pulse wave signal collected by the pulse wave acquisition device are preprocessed to form the PPG-AF dataset,and the pulse wave signal segment is converted into a gram angle field map,so as to retain the complete information of the pulse wave signal and the dependence on time.(2)A Res Net-CBAM model that integrates attention mechanism and deep learning model for atrial fibrillation detection classification is proposed,and the gram angle field picture is used as input,which is trained,verified and tested,and finally realizes the classification detection of atrial fibrillation.The experimental results show that the classification accuracy,sensitivity,specificity and F1 scores of the model are 95.63%,92.17%,95.50%,93.81%,respectively,and the area under the subject’s working characteristic curve is 0.989.(3)A two-channel convolutional neural network based on multi-feature fusion is proposed,the model combines the CNN convolutional neural network with the Res Net-CBAM network by parallel connection,the one-dimensional pulse wave signal and the corresponding gram angle field map as input,the model is tuned,after many experiments,the superiority of the model is confirmed,the experimental results show that the classification accuracy,sensitivity,specificity,F1 score of the model,respectively,98.12%,97.30%,97.30%,97.30%,the area under the receiver operating characteristic curve was 0.991.In summary,the dual-channel convolutional neural network based on multi-feature fusion proposed in this study can quickly and accurately complete the automatic detection of atrial fibrillation,which can not only help doctors diagnose atrial fibrillation,but also provide new ideas for early prevention and auxiliary diagnosis of cardiovascular diseases through portable wearable devices.