In recent years, diffusion models have achieved remarkable success in generating pixel-by-pixel semantic segmentation. In this paper, we propose a brand new end-to-end network Diff-WCSAU-Net for 3D MRI brain tumor segmentation, which integrates the Diffusion Model into WCSAU-Net to efficiently extract semantic information from the input. First, in order to extract the multi-scale features of multimodal MRI and to reduce the number of parameters of the model, a wavelet transform based on Haar wavelets is introduced in this network. Second, to incorporate conditional information into the network and to enhance the feature representation in key regions, we introduce a cross-attention mechanism in the base U-Net and the WCSAU-Net achieves feature fusion in the decoder through an improved skip connection mechanism. Finally, for the specificity of the brain tumor segmentation task, in order to instruct the model to output three different lesion regions with correct containment relationships, an asymmetric Jaccard-like coefficient as a constraint is introduced in the loss function, which improves the overall performance of the model. In comparison with the current state-of-the-art methods, our experimental results show that the proposed model can segment the multilevel anatomical structures of brain tumors with more precision, potentially facilitating the accurate diagnosis and treatment of brain tumors.