Convolutional Recurrent Neural Network and LightGBM Ensemble Model for 12-lead ECG Classification

Abstract

Automatic abnormality detection of ECG signals is a challenging topic of great research and commercial interest. It can provide a cost-effective and accessible tool for early and accurate diagnosis, which increases the chances of successful treatment. In this study, an ensemble classifier that identifies 24 types of cardiac abnormalities is proposed, as part of the PhysioNet/Computing in Cardiology Challenge 2020. The ensemble model consists of a convolutional recurrent neural network that is able to automatically learn deep features, and LightGBM, a gradient boosting machine that relies on hand-engineered expert features. The individual models are combined using class-specific weights and thresholds, which are tuned by a genetic algorithm. Results from 5-fold cross validation on the full training set, report the Challenge metric of 0.593 that outperforms both individual models. On the full hidden test set, the proposed architecture by “AUTh Team” achieves a score of 0.281 with an official ranking of 13/41.