Research On Atrial Fibrillation Detection Algorithm Based On Short Single Lead ECG Recordings

Car^diovascular disease is the most deadly disease at present and two-fifths of the world's population die from it.Motor vehicle drivers are more prone to heart health problems due to long-term exposure to adverse conditions such as traffic j ams and noise pollution,as well as occupational features such as long-term overload work.Atrial fibrillation is the most common type of arrhythmia in cardiovascular disease and is significantly associated with morbidity and mortality in many heart diseases.Atrial fibrillation can be diagnosed by electrocardiogram.Therefore,the use of electrocardiogram to automatically monitor the risk of atrial fibrillation in real time has become an urgent need and research hotspot in the current society.Most existing atrial fibrillation detection algorithms focus on two categories of normal and AF rhythms,ignoring other situations that may occur in reality,or just trained and verified on small data sets containing a small number of subjects,resulting in the atrial fibrillation detection algorithm is still limited in practicality.Therefore,the algorithm proposed in this paper using machine learning algorithm and deep convolutional neural network for atrial fibrillation detection fully considers the possible situations by classifying the rhythms as:normal rhythm,atrial fibrillation rhythm,other rhythms and noise.The model then trained and validated on a collection of 8,528 ECG data from different subj ects.The machine learning-based atrial fibrillation detection algorithm proposed in this paper considers both ventricular activity and atrial activity.After denoising and reference point recognition of ECG signals,a total of 38 features,including time-frequency features,entropy features and statistical features,are extracted.Then four traditional machine learning algorithms including support vector machine,K-nearest neighbor,decision tree and random forest are used to classify the features and perform parameter tuning.The highest accuracy rate 83.83%and the maximum average F1 score 82.23 were achieved using the random f-orest algorithm.The proposed convolutional neural network consists of an input layer,four convolutional blocks,a fully connected layer,and an output layer,wherein each convolution block is composed of two convolutional layers.The algorithm first converts the ECG signal into a logarithmic spectrum as the input then the convolutional neural network can automatically learn the complex features in the data,avoiding the manual design of complex feature extraction rules.The model is trained on the GPU and achieved a better classification accuracy than the machine learning method with the overall accuracy 84.81%,and the average F1 score 82.41%.Iin this paper,we designed and implemented the atrial fibrillation detection algorithm based on short single lead ECG recordings with high accuracy and practicability.Runing the algorithm on a wearable device or mobile device that can collect ECG signals,the motor vehicle drivers could monitor his or her heart health in real time,thus avoiding the occurrence of major traffic accidents,the logistics company can also reduce the logistics cost due to proper monitoring,while at the same time for the society,can achieve a high degree of sharing of medical resources and reduce public medical expenses,which has great practical significance.