Dr. Wenjie Cai, Associate Professor, School of Medical Instrument and Food Engineering, University of Shanghai for Science and Technology, China

Speech Title: Automatic Interpretation of ECG Using Deep Learning

Abstract: Electrocardiography (ECG), which can trace the electrical activity of the heart noninvasively, is widely used to assess heart health. Accurate interpretation of ECG requires significant amounts of education and training. In this work we proposed a novel architecture of deep neural networks to interpret ECG automatically. First the challenge ECGs were preprocessed to eliminate noises and transform outliers properly and were split into a training set and a validation set. The samples were then split it into 5 second segments. Morphological features were extracted by a deep convolutional neural network. Rhythmical features were assessed by a recurrent neural network. All the features were used to train the classification model. Hyperparameters were finely tuned before our codes were submitted to evaluate the hidden test data. The proposed method run fast and only took 8.1 seconds for evaluating 300 samples. It showed robustness in identifying atrial fibrillation and bundle branch block patterns. The overall F1 score was listed among top 10 in the leading board. In conclusion, deep learning architecture has promising application prospects in interpreting 12-lead ECGs.
Keywords: Convolutional neural network; Recurrent neural network; Electrocardiography interpretation