Background: Deep learning (DL) has shown promising results in improving atrial fibrillation (AF) detection algorithms. However, these models are often criticized because of their "black box" nature. Aim: To develop a morphology based DL model to discriminate AF from sinus rhythm (SR), and to visualize which parts of the ECG are used by the model to derive to the right classification. Methods: We pre-processed raw data of 1469 ECGs in AF or SR, of patients with a history AF. Input data was generated by normalizing all single cycles (SC) of one ECG lead to SC-ECG samples by 1) centralizing the R wave or 2) scaling from R-to-R wave. Different DL models were trained by splitting the data in a training, validation and test set. By using a DL based heat mapping technique we visualized those areas of the ECG used by the classifier to come to the correct classification.Results: The DL model with the best performance was a feedforward neural network trained by SC-ECG samples on a R-to-R wave basis of lead II, resulting in an accuracy of 0.96 and F1-score of 0.94. The onset of the QRS complex proved to be the most relevant area for the model to discriminate AF from SR. Conclusion:The morphology based DL model developed in this study was able to discriminate AF from SR with a very high accuracy. DL model visualization may help clinicians gain insights into which (unrecognized) ECG features are most sensitive to discriminate AF from SR.

Background The electrocardiogram (ECG) is commonly used, but most recent rhythm discrimination algorithms still lack both specificity and sensitivity. Deep learning techniques have shown promising results in the classification of physiological signals like ECGs. Purpose To develop and test a deep learning (DL) model to discriminate between atrial fibrillation (AF) and sinus rhythm (SR). Methods For the development of the DL model we used 1499 ECGs sampled at 500 Hz of patients diagnosed with AF. All ECGs were labeled by two experienced investigators. Only ECGs labeled as SR or AF were included in the dataset. To simplify the learning process, solely the first ECG channel was used. The ECG waveforms were preprocessed using the Fourier cosine series to correct for baseline wander. Input data was generated by normalizing and scaling all different heartbeats by centralizing the R peak, leading to 15744 single heart beat samples of 80 data points (figure A). Multiple feedforward architectures were tested with different numbers of layers, filters and activation functions. The models were trained by equally splitting the data (50%SR, 50%AF) in a training (65%), validation (25%) and test set (15%). The best performing model was chosen based on the accuracy. Results A total of 1469 ECGs (1061 (72%)SR, 408 (28%)AF) were included. The model with the best performance was a feedforward model consisting three dense layers with ReLU activation and four dense layers with Linear activation. Training of the model was performed in 32 epochs. Validation of the model resulted in an accuracy of 96% (figure B), precision of 95% and recall of 96%. Conclusions The morphology based deep learning model developed in this study was able to discriminate atrial fibrillation from sinus rhythm with a fairly high accuracy using a limited size dataset and only one lead.