Dual-branch encoding with deformable decoding for abdominal medical image segmentation

To address the limitations of Convolutional Neural Network (CNN) in local feature extraction and Transformer in global modeling for medical image segmentation, this paper proposes the CTDG-Net model. The network uses a dual-branch encoder during the encoding phase and introduces the Deep Deformable Convolutional Module in the decoding phase to adaptively adjust receptive fields for capturing irregular organ boundaries. Combined with the Efficient Gated Upsampling Module, CTDG-Net dynamically fuses local details and global semantic features from the dual encoder branches. Experiments on the Synapse dataset demonstrate that the proposed method achieves a mean Dice coefficient of 83.76%, significantly outperforming existing segmentation networks.