Brain Tumor Classification Based on Multi-Level Attention

Over the previous few years, the incidence of brain tumors has been increasing, and it is now among the top 10 most prevalent tumors worldwide. Early detection and diagnosis of brain tumors can assist doctors in creating targeted treatment regimens. The method of deep learning to classify brain tumor magnetic resonance (MR) images can increase the speed and precision of diagnosis. This paper proposes a multi-level attention model to categorize brain tumors. This model uses EfficientNetV2 as the backbone. The Efficient Triplet Attention (ETA) module and the Pyramid Spilt Attention (PSA) module improve the model's capacity to retrieve meaningful information from MR images and decrease the impact of noisy information. The conclusion shows that our model can effectively increase the classification accuracy of brain tumors, with a rate of 98.53%.