Heart Rhythm Classification using Short-term ECG Atrial and Ventricular Activity Analysis
2017 Computing in Cardiology Conference (CinC)
As a contribution to 2017 Physionet/CinC challenge, this work aims at the classification of different ECG heart rhythms. The importance of heart rhythm classification cannot be understated, as rhythms such as atrial fibrillation have been associated with stroke, coronary artery disease and mortality. Automatic detection of heart rhythms remains a challenging task, as they can be episodic with unpredictable characteristics. In the CinC2017 challenge, the training set contains 8528 single lead
... rt-term ECGs (9-90s, 300 Hz). Recordings are categorized into four classes namely, normal rhythm, atrial fibrillation, other rhythm, and noisy. Heart rhythm classification in this work is carried out by analyzing the atrial and ventricular activities present in the ECG. First, Noisy signals are classified using a Bagging meta-algorithm, trained on a set of features extracted from short-and long-term ECG trends. Then, using a novel QRS-complex cancellation technique, atrial activity is separated and used to extract several features using phase-rectified signal averaging and complexity measures. These features are then combined with heartrate variability and average-beat analysis features, to create the final feature set. The heart rhythm type is determined by a normal vs abnormal rhythm classification (Bagging meta-algorithm), followed, if needed, by an AF vs other rhythm classification (SVM classifier). The performance on the validation set led to an average F-score of 0.91 with normal, other and AF rhythm F-score of 0.95, 0.93, 0.90. On the hidden test set, our algorithm obtained an average F-score of 0.79.