Detection of Atrial fibrillation, the most common cardiac arrhythmia, is a huge challenge for engineers. The databases available online are not sufficient to create reliable algorithms. Due to Physionet 2017 Challenge, researchers have an opportunity to create and benchmark their algorithms on relatively big dataset, annotated with recordings from many different patients. Presented system is an ensemble made of 2 models, that try to complement each other weaknesses. First model is sequential Recurrent Neural Network (RNN) classifier, that is fed by lengths of intervals between following R peaks. Achieved probabilities for each class are combined with hand-designed features and used as an input for Gradient Boosting Machine (GBM) classifier. 36 features were designed in attempt to comprehend entire variability of ECG signals. They can be divided into 5 categories: statistical features, QRS morphology features, RR-interval features, noise features, and frequency-based features.
[1]
Luca T. Mainardi,et al.
A Support Vector Machine approach for reliable detection of atrial fibrillation events
,
2013,
Computing in Cardiology 2013.
[2]
Piotr Augustyniak.
THE USE OF SHAPE FACTORS FOR HEART BEATS CLASSIFICATION
,
2001
.
[3]
Jürgen Schmidhuber,et al.
Long Short-Term Memory
,
1997,
Neural Computation.
[4]
Gari D Clifford,et al.
Impact of the presence of noise on RR interval-based atrial fibrillation detection.
,
2015,
Journal of electrocardiology.
[5]
J. Friedman.
Greedy function approximation: A gradient boosting machine.
,
2001
.
[6]
Qiao Li,et al.
AF classification from a short single lead ECG recording: The PhysioNet/computing in cardiology challenge 2017
,
2017,
2017 Computing in Cardiology (CinC).