Automatic Identification of Abnormalities in 12-Lead ECGs Using Expert Features and Convolutional Neural Networks

Automatic identification of the rhythm/morphology abnormalities in ECGs has gained growing attention in various areas and remains a challenge. We propose an algorithm to classify 12-lead ECGs into 9 categories. We extracted expert features including generic features and specific features with statistics and physiology significance. Then a 17-layer Convolutional Neural Network (CNN) was proposed to detect deep features in ECGs. With these features we trained ensemble classifiers to predict labels. Experiment on the training set (5-fold cross-validation) reports 0.81 accuracy score.