A Machine Learning approach for Atrial Fibrillation detection in telemonitored patients

PEDRO BARRERA; LORENZA VECINO SCHANDY; MARÍA PAULA BONOMINI; CRISTIAN MATEOS; MATIAS HIRSCH; LUCAS GRANA; SERGIO LIBERCZUK

CABA

Congreso; XXIV Congreso de Bioingeniería XIII Jornada de Ingeniería Clínica; 2023

Universidad Nacional Arturo Jauretche y Sociedad Argentina de Bioingeniería

Atrial fibrillation (AF) is the most common type ofcardiac arrhythmia [1]. As it is typically asymptomatic, it oftengoes undiagnosed until major complications arise, such as stroke[2]. Therefore, the development of rapid, economical, and widelyaccessible diagnostic tools for detecting AF at an early stageis crucial. Telemonitoring with machine learning-assisted de-vices shows promise in achieving this goal. This paper presentsan algorithm that automatically detects AF in signals obtainedby portable electrocardiographs connected to a telemonitoringplatform via smartphones. The algorithm consists of two stages:a noise detection stage and an AF detection stage. The noise de-tection stage involves analyzing the ECG signals using 5-secondwindows with a 1-second shift. A support vector machine (SVM)classifier predicts the presence or absence of noise in each win-dow, allowing for the detection of noisy and non-noisy segmentsof the signal. In the AF detection stage, the non-noisy segmentsare processed using a Pan-Tompkins algorithm to find the Rpeaks of the signal, and the corresponding RR interval series.This RR series is used to calculate five indices, which serve asinput features of a logistic regression classifier that predicts thepresence or absence of AF in the ECG signal. The algorithmwas trained and tested using the Physionet Short Single-LeadAF Database (SSLAFDB) and achieved an accuracy of 87.9%and an F1-score of 87.7%. Further validation was performedby an external partner using two other databases, reporting anaccuracy of 93.1% and 89.5% on two different databases