Due to their wide variety in location, appearance, size and intensity distribution, automatic and precise brain tumor segmentation is a challenging task. To address this issue, a computer-aided brain tumor segmentation system based on an adaptive gamma correction neural network (GammaNet) is proposed in this paper. Inspired from the conventional gamma correction, an adaptive gamma correction (AGC) block is proposed to realize intensity invariance and force the network to focus on significant regions. In addition, to adaptively adjust the intensity distributions of local regions, the feature maps are divided into several proposal regions, and local image characteristics are emphasized. Furthermore, to enlarge the receptive field without information loss and improve the segmentation performance, a dense atrous spatial pyramid pooling (Dense-ASPP) module is combined with AGC blocks to construct the GammaNet. The experimental results show that the dice similarity coefficient (DSC), sensitivity and intersection of union (IoU) of GammaNet are 85.8%, 87.8% and 80.31%, respectively, the implementation of AGC blocks and the Dense-ASPP can improve the DSC by 3.69% and 1.11%, respectively, which indicates that the GammaNet can achieve state-of-the-art performance. © 2021 Elsevier GmbH