Automated Classification of Electrocardiograms Using Wavelet Analysis and Deep Learning

For the 2020 PhysioNet/Computing in Cardiology Challenge, we applied wavelet analysis to develop multiple deep learning models, creating a unique model for each lead. This approach leverages the ability of different leads, based upon their anatomical placement, to better observe different arrhythmias. A voting scheme is implemented amongst the leads, allowing for confirmation of arrhythmia diagnosis from multiple leads, thereby increasing confidence in the diagnosis while also allowing for diagnosis of multiple concurrent arrhythmias. We leverage transfer learning to simplify training our deep learning network by utilizing a modified version of SqueezeNet for training. Since SqueezeNet is designed for image classification, the ECG signals are converted to scalograms prior to training. Using this method, our team, Eagles, achieved a challenge validation score of 0.214 and a full test score of 0.205, placing us 20th out of 41 in the official ranking. While this method has shown promise, improvements are needed to improve classification accuracy in order to make it a clinically viable technique.