A Global-Local Features Exchange and Fusion Network for Multi-Organ Segmentation

Convolutional neural network(CNN) based methods for multi-organ segmentation have achieved impressive results. However, the global feature extraction capability of CNNs is limited due to their localisation problem. In this paper, we propose a more efficient CNN and Transformer hybrid network for abdominal multi-organ segmentation. A parallel encoder is formed by the CNN and the Transformer encoder, making full use of the local and global feature extraction capabilities of both. Based on this, feature exchange modules are inserted at each scale of the encoder to enhance the features flow and alleviate the variability between different encoder features. In addition, a feature fusion module and a feature consistency loss function are proposed to couple the output features of the two encoders to ensure the consistency of the decoder input features. Experiments based on the Synapse dataset show that our approach achieves superior results compared with both CNN-based and Transformer-based state-of-the-art methods.