Conv-RGNN: An efficient Convolutional Residual Graph Neural Network for ECG classification

Yupeng Qiang, Xunde Dong, Xiuling Liu, Yang Yang, Yihai Fang, Jianhong Dou

Abstract: Highlights•We present the Conv-RGNN model. It incorporates the speciallydesigned Residual GNN (RGNN) blocks, which efficiently aggre-gate feature information across leads while integrating spatialrelationships, providing a more comprehensive perspective for ECG diagnosis.•The Conv blocks capture temporal features and address GNNs’ local feature extraction limitations, while the Pos-Atten module selectively enhances similar semantic features across positions by gathering contextual information.•Inter-patient experiments on two multi-label datasets and one single-label dataset demonstrate that Conv-RGNN surpasses other models in parameter efficiency, inference speed, performance, and robustness.