Quest Journals
Journal of Electronics and Communication Engineering Research
ISSN:2321-5941 Volume 2 ~ Issue 1 (2014) pp: 01-06
www.questjournals.org
*Corresponding Author: Chee Teck Phua 1 | Page
School of Engineering, Nanyang Polytechnic, Singapore
Research Paper
A novel approach to detect Atrial Fibrillation efficiently and
accurately from 48 hours of ECG data
Jing Niu, Victoria Gokana, Shirley Goh,
Gaelle Lissorgues, Chee Teck Phua
*
Received 03 January, 2014; Accepted 10 January, 2014 © The author(s) 2013. Published with open
access at www.questjournals.org
ABSTRACT:- An algorithm for automatic detection of Atrial Fibrillation (AF) using R peak to R peak (RR)
interval from electrocardiograph (ECG) signals has been developed and evaluated. The algorithm consists of a
preprocessing which is based on Pan-Tompkins method to extract reliable QRS complex thus RR interval,
followed by a 3 steps approach to detect AF using RR interval in ECG. The first step computes the Root Mean
Square of the Successive Differences of the RR intervals extracted from a 48 hours ECG recording to identify
the presence of arrhythmia. The second step defines the precise starting and stopping time of an arrhythmia
episode through applying autocorrelation on the squared ECG signal within an arrhythmia window detected by
the first step. The last step discerns AF from other type of arrhythmias through computing the Shannon Entropy
(ShE) for the entire range of arrhythmia episode defined in step 2. The evaluation of the proposed algorithm
was carried out utilizing normal sinus rhythm, AF and the arrhythmia ECGs provided in MIT-BIH database.
The result demonstrates a 100% reliability in the detection of AF with accuracy of 99.5% in the identification of
the start and stop time of an AF episode. In addition, the relationship between ShE threshold value of AF and
duration of AF episode is studied with a formula proposed based on the results. The total processing time for
the 48 hours ECG data is approximately 8 minutes, demonstrating the high efficiency of this algorithm.
Keywords:- Atrial Fibrillation, Electrocardiogram, Tom-Pankins, Autocorrelation, Shanon Entropy
I. INTRODUCTION
Atrial Fibrillation (AF) is the most common arrhythmia, particularly in the elderly and those with heart
disease.[1,2] It significantly increases the risk of cognitive dysfunction, thrombo-embolism and heart failure
leading to mortality and morbidity [3-8]. Approximately 6 million Europeans and approximately 3 million
Americans are affected by AF [1, 9]. The risk of developing AF increases with age, at age of 55 the risk is
approximately 23% [1]. As the populations aging, it is highly possible that AF will become a public health
burden [10]. Therefore, timely and accurate diagnosis of AF is crucial.
In an AF episode, an uncoordinated electrical and physical activity in the atria of the heart occurs. This
will lead to disrupted electrical pathways thus quivery of the atria and irregular heart-beats with significant beat-
to-beat variability and complexity. Such abnormally in heart activities can be detected through analysis of
electrocardiogram (ECG). The ECG of AF episodes have irregular heart beat interval (RR) and/or low P-wave
amplitude in the QRS complex. Different algorithms had been explored to distinguish AF from other kind of
arrhythmias using RR. Examples of such algorithms are the Root Mean Square Successive Differences
(RMSSD), Sample Entropy and Fast Fourier transform [11, 12]. However, the patients may not be aware of the
occurrence of AF as its occurrence is usually unpredictable [13,14]. The prevalence of asymptomatic AF
diagnosed parenthetically during clinical visit is approximately 20% [15,16]. Hence, Holter-monitoring is
commonly utilized to trace the ECG signal of a patient for 24 to 48 hours [17]. It is time consuming and
inefficient for a trained specialist to examine through such huge amount of data to search for abnormal episode
which might be only less than one minute in length. This makes the automatic detection of AF episodes with
various duration and random location from the recorded ECG data essential.
In this paper, a computationally efficient algorithm that is able to accurately recognize AF episode
from ECG data recorded for up to 48 hours is proposed. The algorithm analyzes the dynamics of RR interval
obtained from an ECG signal. The diagnosis is made by examining the variability and randomness of the RR
interval.