To address the issues of insufficient global information capture and inadequate deep semantic information fusion in the U-Net model for MRI brain tumor segmentation,a novel brain tumor segmentation network,ASGU-Net was proposed. The algorithm was based on 3D U-Net,incorporating a graph convolution inference module to capture additional long-range contextual features. Additionally,dynamic snake convolution (DSConv) was introduced in the encoder-decoder to better accommodate the varied shapes of tumors,enhancing edge feature extraction and improving segmentation accuracy. Furthermore,an adaptive spatial feature fusion (ASFF) module was introduced in the decoder to enhance the feature fusion effect by integrating semantic information captured by multiple encoder blocks. The evaluation on the publicly available BraTS 2019—2021 datasets shows that the Dice values for whole tumor,tumor core and enhanced tumor are 90.70%/90.70%/91.00%,84.90%/84.00%/88.80% and 77.30%/77.40%/82.50%,respectively. The experimental results demonstrate the effectiveness of ASGU-Net in the brain tumor segmentation task. ASGU-Net can effectively addresses the issues of inadequate global information capture and feature fusion,providing effective reference for high-precision automated brain tumor segmentation.