Detection of Cardiac Arrhythmias From Varied Length Multichannel Electrocardiogram Recordings Using Deep Convolutional Neural Networks

Automatic identification of different arrhythmias helps cardiologists better diagnose patients with cardiovascular diseases. Deep learning algorithms are used for the classification of multichannel ECG signals into different heart rhythms. The study dataset includes a cohort of 43101 12-lead ECG recordings with various lengths. Two options are tested to standardize the recordings length: zero padding and signal repetition. Downsampling the recordings to 100 Hz allow handling the problem of different sampling frequencies of data coming from different sources. We design a deep one-dimensional convolutional neural network (CNN) called VGG-ECG, a 13-layer fully CNN for multilabel classification. Our team is called MIndS and our approach achieved a challenge validation score of 0.368, and full test score of -0.128, placing us 38 out of 41 in the official ranking.