ECG Classification Based on Unfixed-Length Segmentation of Heartbeat

An automatic ECG classification system is of great significance. Deep learning (DL) methods, e.g. convolutional neural network (CNN), are proved very useful for ECG classification. Due to that DL methods require the input data with fixed length, current work segments beat with fixed length. But obviously, the length of beat is varying with the changing temporal, personal, or contextual conditions. We solve the contradiction by segmenting beat with unfixed length and then reshape the beat to that with fixed length. Based on this, we extract inner-beat, beat and inter- beat representation. We validate our method on MIT-BIH dataset and the accuracy reaches 95.45 %, better than related work.