Multi-scale Hybrid CNN-GNN Network with Attention-Guided Fusion for Echocardiography Segmentation

Medical image segmentation methods typically utilize convolutional neural network (CNN) with pooling layers to expand the receptive field for capturing high-level features. Due to the inherent locality of convolutional operations, CNN often exhibits limitations in explicitly modeling long-distance dependencies. In this paper, we propose a novel hybrid multi-scale graph neural network (HMGN) model for echocardiography segmentation that combines the CNN and graph neural network to capture both local and non-local image features. Specifically, in order to capture features over a large receptive field, we propose the patch graph attention (PGAT) module. Furthermore, the size and shape of the heart can vary significantly across different frames in various positions, we utilize both CNN and PGAT modules at multi-scale to capture rich information. Experimental results on the CAMUS dataset demonstrate that the proposed method obtains improved segmentation performance compared to the state-of-the-art methods, achieving the average Dice coefficient of 93.54% and Specificity of 99.29%.