Attention mechanisms with UNet3+ on brain tumor MRI segmentation
Yuan Zhong · AIP conference proceedings · 2022
Brain tumor is one of the most dangerous disease of human brain. Often treated by radiotherapy, chemotherapy and surgery, timely detection and diagnosis is crucial to patient survival and prognosis. To speed up diagnosis process and improve overall accuracy, researchers keep on finding new techniques that can be implemented in the field of medical image analysis. Recently, deep learning image segmentation models are widely used in brain tumor image analysis, and contests such as BraTS are held annually to facilitate research in this field. In this paper, three attention mechanisms, convolutional block attention module, attention gate and efficient channel attention were combined with full scaled skip connection Unet, and transfer learning from pre-trained Unet's effect was also explored.