Multi-organ medical image segmentation provides critical anatomical information for clinical decision-making and supports physicians in diagnosis. To address challenges in precise localization within complex regions, such as low-contrast organ boundaries, as well as insufficient scale awareness and redundant feature encoding caused by large inter-organ scale variations, this study proposes a medical image segmentation model, SC-TransUNet, based on spatial-channel collaborative modeling. In the attention interaction stage of the Transformer encoder, a hybrid deformable neighborhood attention (HDNA) module is introduced to improve boundary detail localization through adaptive offset resampling and local neighborhood interactions. In the channel mixing stage, a multi-scale channel-adaptive convolutional gated unit (MSCA-CGU) is designed to enhance scale-aware local representations, while explicit channel modulation is employed to suppress highly correlated redundant channels and improve fusion stability. Experimental results show that SC-TransUNet achieves a mean DSC score of 81.73% with a Hausdorff distance (HD) of 18.14 mm on the Synapse dataset, and 92.18% DSC with an HD of 2.83 mm on the ACDC dataset, consistently outperforming the baseline TransUNet. These results verify the effectiveness and robustness of the proposed method in handling complex boundaries and scale variations in multi-organ segmentation tasks.