Early diagnosis of brain tumors enables earlier, more targeted treatment, significantly improving the patient’s chances of recovery. As deep learning rapidly advances in medical image processing, brain tumor classification algorithms based on it have gained significant attention. Convolutional neural network (CNN), known for their powerful feature learning abilities, can automatically extract tumor-related features from brain images for classification by training on large-scale medical image datasets. While VGG’s successful application in the ImageNet competition is well-known, simply deepening the network structure does not always result in significant performance improvements for medical image tasks. As the network depth increases, the risk of overfitting and extracting irrelevant features may grow, resulting in a lack of robustness in the model’s learned features. Networks that are excessively deep may experience gradients that become either too small or too large during backpropagation, leading to instability in weight updates. This paper addresses the above issues by using the ResNet network as the backbone for brain tumor image classification and proposes a way based on the dual attention mechanism: MAMF-ResNet (Multi-Attention Mechanisms Fusion ResNet). The structure of the ResNet network is optimized and adapted as a way to improve the classification accuracy. This chapter uses the publicly available BraTS 2018 and BraTS 2019 datasets for training and testing. Experimental results show that the improved model proposed in this chapter outperforms the base model in all evaluation metrics.