12-lead ECG reconstruction using convolutional neural network

The tradition reconstruction of the convolutional 12-lead electrocardiography(ECG) is Dower transform, but the model is too simple that the performance about detail is poor. To improve the performance, we raise a new way for reconstruction based on convolutional neural networks(CNN). At first, in order to get the high semantic of the Frank XYZ-lead ECG, we train one auto-encoder model (AE) based on convolutional neural networks. Then we train an CNN model for reconstructing 12-lead of which the input layer is the encoder layer of the AE. At last, to judge the performance of the model, R2 and percent correlation are adopted for evaluating the results. The result proves that CNN model is quite suitable and outstanding.