Bimodal classification algorithm for atrial fibrillation detection from m-health ECG recordings

• Published in: Computers in biology and medicine Vol. 104; pp. 310 - 318
• Main Authors: Kruger, Grant H., Latchamsetty, Rakesh, Langhals, Nicholas B., Yokokawa, Miki, Chugh, Aman, Morady, Fred, Oral, Hakan, Berenfeld, Omer
• Format: Journal Article
• Language: English
• Published: United States Elsevier Ltd 01.01.2019; Elsevier Limited
• Subjects: Age; Algorithms; Arrhythmia; Atrial fibrillation; Atrial Fibrillation - diagnosis; Atrial Fibrillation - physiopathology; Atrial Fibrillation - therapy; Audio frequencies; Automation; Cardiac arrhythmia; Classification; Correlation analysis; Databases, Factual; Datasets; Diagnosis, Computer-Assisted; E-health; EKG; Electrocardiography; Ethics; Female; Fibrillation; Frequency domain; Humans; Laboratories; Male; Monitoring, Physiologic; Patients; Physicians; Recording; Rhythm classification; Risk analysis; Statistical analysis; Stroke; Telemedicine; Thromboembolism; Time domain; Waveforms; United States > US; Frequency domain; Time domain; E-health; Rhythm classification; Atrial fibrillation
• ISSN: 0010-4825; 1879-0534; 1879-0534
• DOI: 10.1016/j.compbiomed.2018.11.016

Atrial Fibrillation (AF) is the most common cardiac arrhythmia, presenting a significant independent risk factor for stroke and thromboembolism. With the emergence of m-Health devices, the importance of automatic detection of AF in an off-clinic setting is growing. This study demonstrates the performance of a bimodal classifier for distinguishing AF from sinus rhythm (SR) that could be used for automated detection of AF episodes. Surface recordings from a hand-held research device and standard electrocardiograms (ECG) were collected and analyzed from 68 subjects. An additional 48 subjects from the MIT-BIH Arrythmia Database were also analyzed. All ECGs were blindly reviewed by physicians independently of the bimodal algorithm analysis. The algorithm selects an artifact-free 6-s ECG segment out of a 20-s long recording and computes a spectral Frequency Dispersion Metric (FDM) and a temporal R-R interval variability (VRR) index. Scatter plots of the VRR and FDM indices revealed two distinct clusters. The bimodal scattering of the indices revealed a linear classification boundary that could be employed to differentiate the SR from AF waveforms. The selected classification boundary was able to correctly differentiate all the subjects from both datasets into either SR or AF groups, except for 3 SR subjects from the MIT-BIH dataset. Our bimodal classification algorithm was demonstrated to successfully acquire, analyze and interpret ECGs for the presence of AF indicating its potential to support m-Health diagnosis, monitoring, and management of therapy in AF patients. •A personal ECG recorder was developed and used to collect data from 68-subjects.•Heart rate variability versus a frequency domain-based metric was computed.•Distinct clusters for sinus rhythm and atrial fibrillation groups were obtained.•It was found that a linear classification boundary could be employed.•Perfect separation of the groups was obtained for the collected data.