1. Real-time Detection of Atrial Fibrillation
from Short time single lead ECG traces
using Recurrent neural networks
Sujadevi VG*, Soman KP and Vinayakumar R
1Centre for Computational Engineering and
Networking (CEN), Amrita School of Engineering,
Coimbatore, Amrita Vishwa Vidyapeetham, Amrita
University, India.
2. Outline
• Introduction
• Background information / Related works
• Proposed Method – Deep Learning
• Description of the data set and Results
• Summary
• Future Work
• References
2
3. Introduction
• Atrial fibrillation (AF) is a disorder of the
functioning of the heart’s electrical system that is
characterized by the irregular beating of the
heart [1].
• Atrial fibrillation (AF) is the predominant type of
cardiac arrhythmia affecting more than 45 Million
individuals globally.
• It is one of the leading contributors of strokes and
hence detecting them in real-time is of
paramount importance for early intervention.
3
4. Background information / Related works
• Machine learning methods are used to
identify the pathological ECG from the normal
sinus rhythm.
• Machine learning methods relies on the
feature engineering and deposing
mechanisms.
• Deep learning is a new filed of machine
learning which can learn the patterns by
taking the raw input ECG signals.
4
5. Proposed Method
Figure 1. Architecture of proposed system for normal sinus
rhythm and atrial fibrillation.
5
6. Description of the data set and Results
We used the publically available raw signals of
Atrial fibrillation (AF) and normal
sinus rhythm (NSR) from MITBIH Physionet; MIT-
BIH Atrial Fibrillation Database
and MIT-BIH Normal Sinus Rhythm Database [2].
6
7. ECG signal
Figure (a) A single lead ECG wave form of
normal sinus rhythm,(b) A single lead ECG wave
form with atrial fibrillation
7
9. Summary and Future work
• Deep learning based mechanism such as RNN, LSTM and GRU
architecture is proposed to distinguish AF and NSR on a single
lead ECG.
• All the deep learning methods have performed well, mostly
LSTM and GRU outperformed RNN and GRU takes less training
cost in comparison to LSTM.
• The proposed method is considered as more accurate in real-
time ECG classification because it doesn’t rely on any feature
engineering mechanisms.
• Though the deep network methods showed significant results,
we lack in showing the inner mechanics of the deep models.
This can be achieved by transforming the non-linearity to
linearized form, thereby computing the Eigen values and
Eigen vectors on them across time-steps [3].
9
10. References
[1] Go AS, Hylek EM, Phillips KA, Chang Y, Henault LE, Selby JV,
Singer DE. Prevalence of diagnosed atrial fibrillation in adults:
national implications for rhythm management and stroke
prevention: the AnTicoagulation and Risk Factors in Atrial
Fibrillation (ATRIA) Study. JAMA. 2001 May 9;285(18):2370-5.
[2] Goldberger AL, Amaral LAN, Glass L, Hausdorff JM, Ivanov
PCh, Mark RG, Mietus JE, Moody GB, Peng C-K, Stanley HE.
PhysioBank, PhysioToolkit, and PhysioNet: Components of a New
Research Resource for Complex Physiologic Signals. Circulation
101(23):e215-e220
[3] Moazzezi, R. Change-based population coding. PhD
thesis,UCL (University College London), 2011.
10