Automatic classification of atrial fibrillation from short single-lead ECG recordings using a Hybrid Approach of Dual Support Vector Machine

• Published in: Expert systems with applications Vol. 198; p. 116848
• Main Authors: Geweid, Gamal G.N., Chen, Jiande D.Z.
• Format: Journal Article
• Language: English
• Published: New York Elsevier Ltd 15.07.2022; Elsevier BV
• Subjects: Accuracy; Atrial fibrillation (AFr); Cardiac arrhythmia; Cardiac arrythmia (CA); Cardiology; Comparative studies; Datasets; Diagnosis; Dual Support Vector Machine (DSVM); Electrocardiography; Fibrillation; Heart diseases; Hybrid Approach (HA); Performance evaluation; Segmentation; Signal processing; Signs and symptoms; Support vector machines; Atrial fibrillation (AFr); Dual Support Vector Machine (DSVM); Cardiac arrythmia (CA); Hybrid Approach (HA)
• ISSN: 0957-4174; 1873-6793
• DOI: 10.1016/j.eswa.2022.116848

Automatic Classification of Atrial Fibrillation from Short Single-Lead ECG Recordings using a Hybrid Approach of Dual Support Vector Machine. [Display omitted] •Hybrid Approach of Dual Support Vector Machine is used for the detection of AFr.•Cardiac segmentation was used to further evaluate and diagnose AFr based on the ECG.•A comparative study is made on detection of AFr using the top five-scoring methods.•The proposed algorithm saves the operation time in addition to improving accuracy. Millions of people worldwide are affected by cardiac arrythmias currently. One of the life threatening arrythmias associated with high morbidity and mortality is Atrial Fibrillation (AFr). Electrocardiogram (ECG) is regularly used in the diagnosis and evaluation of cardiac arrythmias. The goal of the ECG is to improve the outcome as well as reduce time to diagnosis in heart disease identification. Using an ECG makes the diagnosis of a cardiac arrythmia much easier. Cardiologists depend on the ECG signal to determine the appropriate treatment method of the cardiac arrythmia. In this work, cardiac segmentation was used to further evaluate and diagnose AFr based on ECG. The issue with segmentation is that it is complicated due to similarities in amplitude, time among different ECG signals as well as noise. In this paper, a comparative study is made on detection of Atrial Fibrillation using the top five-scoring methods submitted in the PhysioNet/Computing in Cardiology Challenge 2017. A new method based on a Hybrid Approach of Dual Support Vector Machine (HA-DSVM) is used for the detection of atrial fibrillation. This proposed method is accomplished by stochastic gradient descent with cross entropy loss function. The method is tested on the dataset collected from the 2017 Physionet/CinC challenge dataset with performance evaluation on training (99.27 %). Using this technique, an F1 score and accuracy of 0.95 and 99.27% can be obtained on the validation data set. One of the advantages of the proposed technique is the high reliability and accuracy which simplifies the extraction process and removal of detecting ECG signal fiducial points and removing hand-crafted features. This provides screening for a large population with a new method for diagnosing atrial fibrillation. The proposed method can be applied to many the population with symptoms of atrial fibrillation which will improve the accuracy of diagnosis and determine the required treatment.