An improved Poincaré plot-based method to detect atrial fibrillation from short single-lead ECG. (February 2021)

• Record Type: Journal Article
• Authors: Luo, Chengsi; Li, Quanchi; Rao, Haibo; Huang, Xin; Jiang, Hongxiu; Rao, Nini
• Is Part Of: Biomedical signal processing and control. Volume 64(2021)
• Publication Date: 2021-02
• Subjects: Atrial fibrillation -- Electrocardiogram -- Feature extraction and classification -- RR interval -- Poincaré plot; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2020.102264
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A simple but efficient heuristic method for atrial fibrillation detection. A novel AF feature extraction strategy based on Poincaré plot of first-order differential RR interval. The advantage of distinguishing AF from non-AF with complicated arrhythmias. Abstract: Purpose: With the popularity of wearable mobile devices, increasing numbers of applications use single-lead electrocardiograms (ECGs), which provides information about heart rate and rhythm to detect cardiac diseases. Atrial fibrillation (AF) is the most common arrhythmia in clinical practice, it is very important to detect AF early. In this study, we propose a Poincaré plot-based method for AF detection that has the advantages of requiring few calculations and short detection segments. Method: Our method first extracts the first-order differential RR interval (ΔRRI) sequences from a segmented ECG data, and then the polar coordinate transformation is performed on the Poincaré plot of the ΔRRI to obtain the phase-based distribution. Two features, the distribution width Dw and the average distribution height Dh are extracted from the phase-based distribution to classify the AF and non-AF episodes. Five PhysioNet databases were used for the assessment, which contain 3, 843.3 hours of data from 229 subjects. Long-Term AF Database was selected as the training set, and the remaining four databases were used as the testing sets. Results: In the testing sets, the results yield good sensitivity/specificity (97.91%/99.14%) with 60 second ECG signal segments, which are better than those of the existing the Poincaré plot-based methods, and the method also performs well with 20 second ECG signal segments (sensitivity/specificity of 97.28%/98.35%). Conclusion: The detection performance of this method is good under different condition, and has the advantage of easy implementation, which makes it promising for application in the detection and monitoring of AF.