Multi-Label Classification of 12-lead ECGs by Using Residual CNN and Class-Wise Attention

Abstract

Cardiovascular diseases have become the leading cause of illness and death worldwide. Due to their chronic nature, early screening and follow-up management will effectively improve the prevention and treatment of cardiovascular diseases, where automatic electrocardiogram (ECG) classification will play an important role. In this work, we take part in the 2020 PhysioNet - CinC Challenge (in the team ECGMaster) and propose a novel multi-label classifier of 12-lead ECG recordings which combines a residual convolutional network (residual CNN) with a class-wise attention mechanism. To deal with the problem of data imbalance between classes, we utilize a novel weighted focal loss in the training of our models. Our models were trained and tested in a 5-fold cross validation on the training data with resulting scores of 0.5501 ± 0.0223 according to the challenge metric, demonstrating a promising method for the classification of ECGs. We note that we were unable to score and rank our model on the official test data, the results were obtained on the training set only and may be over-optimistic.