Automated atrial fibrillation detection based on deep learning network

Aiming at the shorting of the existing atrial fibrillation (AF) detection algorithms and improve the ability of intelligent recognition and extraction of AF signals. Recently, deep learning theory with massive data has been used on image, voice and other filed widely. In this paper, a method based on the stack sparse autoencoder neural network, a instance of deep learning strategy, was proposed for AF detection. Greedy layer-wise training algorithms and massive unlabeled hotter data from a hospital were used to train the deep learning system, and Back Propagation algorithm and half of the MIT-BIH standard databases were applied to optimized the whole system. Another half of the standard data were used to evaluated the performance of this method. The autoencoder learns the high level features which can describe the necessary information better from the raw data The experimental results show that the accuracy of the algorithm based on stack sparse autoencoder is 98.309%, so this approach is of great significance on the real-time monitoring of atrial fibrillation signal in electrocardiogram.