Atrial Fibrillation Detection with Convolutional Neural Network

We propose an algorithm for the atrial fibrillation(AF) detection a short single-lead electrocardiogram(ECG) records using convolutional neural networks. In this work, we creatively convert time series problem into space problem to solve, and several comparison experiments were conducted to adjust and optimize the parameters. Finally, the algorithm with a 52-layer convolutional neural network(CNN) could obtain the best performance. Besides, residual connections have been employed to optimize the model and a simple data augmentation was introduced to prevent overfitting during training procedure. We evaluate the performance of our algorithm on the 2017 PhysioNet/Computing in Cardiology(CinC) Challenge dataset, obtaining an accuracy of 83.8%. And the experimental results are about 1.5 percentage points higher than that obtained by CRNN which ranks top-five in the competition.