Beatwise ECG Classification for the Detection of Atrial Fibrillation with Deep Learning

Atrial fibrillation (AF) is the most common, sustained cardiac arrhythmia. Early intervention and treatment could have a much higher chance of reversing AF. An electro-cardiogram (ECG) is widely used to check the heart's rhythm and electrical activity in clinics. The current manual processing of ECGs and clinical classification of AF types (paroxysmal, persistent and permanent AF) is ill-founded and does not truly reflect the seriousness of the disease. In this paper, we proposed a new machine learning method for beat-wise classification of ECGs to estimate AF burden, which was defined by the percentage of AF beats found in the total recording time. Both morphological and temporal features for categorizing AF were extracted via two combined classifiers: a 1D U-Net that evaluates fiducial points and segmentation to locate each heartbeat; and the other Recurrent Neural Network (RNN) to enhance the temporal classification of an individual heartbeat. The output of the classifiers had four target classes: Normal Sinus Rhythm (SN), AF, Noises (NO), and Others (OT). The approach was trained and validated on the Icentia11k dataset, with 1001 and 250 patients' ECGs, respectively. The testing accuracy for the four classes was found to be 0.86, 0.81, 0.79, and 0.75, respectively. Our study demonstrated the feasibility and superior performance of combing U-net and RNN to conduct a beat-wise classification of ECGs for AF burden. However, further investigation is warranted to validate this deep learning approach.