Determination of T Wave End by Multi-input Deep Learning Model

Prolonged QT intervals correspond to higher risks of malignant ventricular arrhythmias. The QT interval is the duration from the onset of QRS complex to the end of T wave (Tend). The detection of Tend is one of the most difficult problems for ECG delineation. In clinical practice, Tend location is manually determined by experienced physicians. This procedure is time-consuming, laborious and physician-specific. In this study, we propose a deep learning model for automatic Tend determination based on convolutional neural network (CNN) and long short-term memory network (LSTM). We train the model with three types of examples, one is the main input and the other two auxiliary inputs. We validate the model using Physionet QT database and a simulation ECG database, the mean±standard deviation of errors are 0.36±0.74 ms and -1.05±0.14 ms respectively.