Research On Automatic Detection Algorithm Of Atrial Fibrillation Based On Feature Fusion

In today’s world,cardiovascular diseases have ranked first among the causes of death in the global population,and they have to arouse widespread attention and great attention from the scientific community and society.As the most common arrhythmia,atrial fibrillation has often become a hot topic for researchers because of the high incidence and risk of atrial fibrillation.The diagnosis of atrial fibrillation based on electrocardiogram has problems such as low accuracy and huge human consumption.The diagnosis of atrial fibrillation is mainly based on the clinical diagnosis of specialists and highly dependent on expert experience.The automatic classification of atrial fibrillation has always been an urgent problem to be solved.The main methods currently used by Chinese and foreign researchers include traditional feature classification and deep learning classification methods.Both of these methods rely on the high computing power of today’s computers to achieve classification purposes through continuous learning.The classification accuracy has been greatly improved compared to the past,but both have their own shortcomings.This article aims at the characteristics of atrial fibrillation signals,combined with existing research methods at home and abroad,and proposes a ECG signal atrial fibrillation classification method based on convolutional neural networks and feature fusion,which is verified in a series of subsequent experiments The reliability of the algorithm.The main research contents of this paper are as follows:First,this paper introduces the research background of ECG signal classification of atrial fibrillation in detail,and the current research status in China and in abroad,and introduces the composition of ECG signal and common ECG database.Secondly,this article uses digital signal processing methods and improved Pan-tompkin algorithm to analyze the ECG signals that are prone to abnormal baseline drift,fundamental frequency interference,and ECG signal inversion during ECG acquisition.By preprocessing,the noise signal doped in the ECG signal is removed,and the inversion phenomenon is corrected.Third,this article uses the public training set of the 2017 PhysioNet/CinC ECG competition as a sample,and collects 234 different features from the ECG signal.These features include the parameters of the RR interval of the ECG signal and the parameters of the heart rate variability(HRV),And the distribution of other waveforms of the ECG signal,etc.,also include some non-linear characteristics.We make these features into a feature matrix,and train and verify them in the machine learning algorithm model.In the 5-fold crossover experiment,the accuracy and F1 scores were close to the results of the 2017 race.Fourth,this article uses the MIT-BIH AFDB atrial fibrillation database as a sample,and the extracted ECG signals are made into a variety of two-dimensional images,including time-domain enhanced images,STFT transform time-frequency images,S-transform time-frequency images,Gram Angle figure etc.Then,the ECG signals of atrial fibrillation were classified by two-dimensional convolutional neural network model(the model is composed of 6 convolutional layers,3 pooling layers,two fully connected layers and a softmax layer,using Relu function as the activation function).Fifth,based on the idea of manually extracting feature classification and deep learning classification,this paper combines the two methods,modifies the output of the deep neural network model,and finally forms an atrial fibrillation feature fusion classification algorithm.The algorithm combines manual features and deep learning features,and has the advantages of two features.We use technical indicators such as accuracy,sensitivity,and specificity to verify the performance of the algorithm.The test result in the atrial fibrillation database reached 99.2%,which was significantly higher than using feature classification methods or deep learning methods alone.In conclusion,the automatic detection algorithm of ECG signal atrial fibrillation based on feature fusion proposed in this paper has the advantages of high accuracy and convenient use.In the future,portable ECG monitoring device can be used to realize real-time monitoring of atrial fibrillation,improve the quality of life of patients and reduce the risk of sudden death.