Authors:
Gennaro Laudato
1
;
Franco Boldi
2
;
Angela Rita Colavita
3
;
Giovanni Rosa
1
;
Simone Scalabrino
1
;
Paolo Torchitti
2
;
Aldo Lazich
4
and
Rocco Oliveto
1
Affiliations:
1
STAKE Lab, University of Molise, Pesche (IS), Italy
;
2
XEOS, Roncadelle (BS), Italy
;
3
ASREM, Campobasso (CB), Italy
;
4
Ministero della Difesa, Roma (RM), Italy
Keyword(s):
ECG Analysis, Atrial Fibrillation, Arrhythmia, Decision Support System, Machine Learning.
Abstract:
Atrial Fibrillation (AF) is a common cardiac disease which can be diagnosed by analyzing a full electrocardiogram (ECG) layout. The main features that cardiologists observe in the process of AF diagnosis are (i) the morphology of heart beats and (ii) a simultaneous arrhythmia. In the last decades, a lot of effort has been devoted for the definition of approaches aiming to automatic detect such a pathology. The majority of AF detection approaches focus on R-R Intervals (RRI) analysis, neglecting the other side of the coin, i.e., the morphology of heart beats. In this paper, we aim at bridging this gap. First, we present some novel features that can be extracted from an ECG. Then, we combine such features with other classical rhythmic and morphological features in a machine learning based approach to improve the detection accuracy of AF events. The proposed approach, namely MORPHYTHM, has been validated on the Physionet MIT-BIH AF Database. The results of our experiment show that MORPH
YTHM improves the classification accuracy of AF events by correctly classifying about 4,400 additional instances compared to the best state of the art approach.
(More)