MAFUNet: Multi-Attention Fusion Network for Medical Image Segmentation

The purpose of medical image segmentation is to identify target organs, tissues or lesion areas at the pixel level to help doctors evaluate and prevent diseases. Therefore, the model is required to have higher accuracy and robust representation. At present, the proposed models focus on the improvement of performance, and ignore the number of trainable parameters. This paper proposes a lightweight ECA-Residual module to build a model encoder, which can effectively extract features while reducing the number of parameters. The feature fusion of encoder and decoder using simple skip connections will produce semantic differences. Therefore, in this paper a spatial attention gating module is designed to solve this problem. The module suppresses the image irrelevant area and improves the performance of the model while ensuring the computational efficiency. The experiment is carried out on the Synapse dataset, and the results are superior to the current advanced models in terms of accuracy and parameter quantity with 84.21 % dice coefficient and 12.79 HD95. In addition, the accuracy of the model on the ACDC dataset is also better than the current advanced model, which proves the robustness and generalization of the model.