Advancements in medical image segmentation have enhanced diagnostic efficiency and accuracy by enabling the identification and isolation of key organs and lesions in medical scans. Pure CNN architectures, Transformer architectures, and their hybrid networks have been extensively employed in medical image segmentation, exhibiting remarkable efficacy. However, these methods still have shortcomings in terms of acquiring global and local information, and some of them depend on the supplementation of additional information. Furthermore, these methods do not consider the differences in sampling of different regions, which is not conducive to the learning of features by deep networks. This study presents a novel network architecture, LocalplusGlobalNeXt (LpGNeXt). This network utilizes a fully convolutional model to simulate visual attention at each stage by roughly focusing on the global and subsequently on local details. The introduc tion of a multi-scale global attention module enhances the network’s capacity for global information capture, respectively. Subsequently, boundary-aware local atention module is employed to more accurately delineate local boundaries at negligible cost. Furthermore, we apply the self-attention mechanism to sampling and propose window-attention based downsampling and upsampling to adaptively preserve and recover feature details. The efficacy of our approach is empirically evaluated on LiTS2017 and BraTS2021, both of which demonstrate superior performance in comparison to existing methodologies.