Incorporating artificial intelligence into kidney segmentation is pivotal. It expedites precise detection and quantification of kidney conditions, thereby enhancing the efficacy of intelligent kidney disease diagnosis. This study introduces an innovative kidney segmentation technique for CT images, based on the deformable large kernel attention mechanism (D-LKA), aiming to enhance the precision of segmentation for various tissue structures including kidneys, tumors, arteries, and veins. By integrating D-LKA, the model adapts flexibly to changes in the shape of objects in the image, effectively capturing both global and local features. Experimental results demonstrate that the model outperforms current advanced models on the KiPA22 dataset. We employ diverse evaluation metrics to evaluate our model. In the segmentation of kidneys, tumors, arteries, and veins, the model achieves 95.8, 85.4, 87.2 and 84.1 in Dice Similarity Coefficient (DSC), 17.4, 10.04, 16.73 and 13.10 in Hausdorff Distance (HD), 0.47, 2.06, 0.47 and 0.80 in Average Surface Distance (AVD). In comparison with other segmentation approaches, our model achieves better segmentation results.