Utilization of Residual CNN-GRU with Attention Mechanism for Classification of 12-lead ECG

Petr Nejedly1, Adam Ivora1, Ivo Viscor1, Josef Halamek2, Pavel Jurak3, Filip Plesinger3
1Institute of Scientific Instruments of the Czech Academy of Science, 2Institute of Scientific Instruments, CAS, CZ, 3Institute of Scientific Instruments of the CAS


Cardiac diseases are the most common cause of death. The fully automated classification of the electrocardiogram (ECG) supports early capturing of heart disorders, and, consequently, may help to get treatment early. Here in this paper, we introduce a deep neural network for classification of human ECG into 9 groups (atrial fibrillation, 1st degree AV block, Bundle branch blocks, premature contractions, changes in the ST segment and normal rhythm). The network architecture utilizes a convolutional neural network with residual blocks, bidirectional Gated Recurrent Units, and an attention mechanism. The algorithm was trained and validated on the public dataset proposed by the PhysioNet Challenge 2020. The trained algorithm was tested using a hidden test set during the unofficial phase of the challenge and obtained the F2 score, G2 score, and geometric mean score of 0.817, 0.599 and 0.700, respectively as entries by the Medisig-MAI team. The runtime of our algorithm was 19 minutes utilizing only CPU computing.