Brain tumor segmentation with attention-based U-Net

Brain tumors are a hazardous type of tumor, and they build pressure inside the skull when they grow, which can potentially cause brain damage or even death. Attention mechanisms are widely adopted in state-of-the-art deep learning architectures for computer vision and neural translation tasks since they enhance networks' ability to capture spatial and channel-wise relationships. We offer an attention-based image segmentation model that outlines the brain tumors in Magnetic Resonance Imaging (MRI) scans if present. In the paper, we mainly focus on integrating Squeeze-and-Excitation Block and CBAM into the commonly used segmentation model, U-Net, to resolve the problem of concatenating unnecessary information into the decoder blocks and attempt to locate the tumor boundaries. Our research clearly shows the application of the attention mechanism in U-Net, incorporates the Squeeze-and-Excitation with CBAM, and improves the performance in the brain tumor segmentation task. The model is delivered on an app with additional text to speech and chatbot features provided.