2E-Net: Unet-like Double Encoder for Medical Image Segmentation

As an important branch in the field of medical image segmentation, abdominal multi-organ segmentation plays a crucial role in disease diagnosis and treatment supervision. Through long-term research in the field of medical image segmentation, a series of medical image segmentation models represented by U-Net have made significant progress. There are many types of abdominal organs with different sizes and blurred boundaries, so the existing models still fail to achieve satisfactory segmentation results. Most existing abdominal multi organ segmentation methods use pure CNN, pure Transformer, or a combination of both to construct the entire network structure, which cannot better leverage the advantages of both. In order to solve these problems, we propose a double encoder medical image segmentation method called 2E-Net, which improves the MISSFormer method based on Transformer by introducing ResNet50. 2E-Net introduces an Encoder Fusion Module (EFM) to merge the two encoders for feature extraction. Extensive experiments on the Synapse dataset show that our proposed method outperforms the MISSFormer method, achieving a Dice Similarity Coefficient (DSC) score of 82.13%.