Brain glioma segmentation method based on UNet network and attention mechanism

The invention provides a brain glioma segmentation method based on a UNet network and an attention mechanism, an improved brain glioma segmentation algorithm based on a UNet network model is adopted, the attention mechanism is added in the basic UNet network model, a loss function is improved, the attention mechanism is utilized to pay attention to a key area, and the segmentation accuracy of the brain glioma is improved. Key information in the image is extracted while unimportant information is ignored, the accuracy and effect of brain glioma nuclear magnetic resonance image segmentation are improved, a loss function is improved, the class imbalance problem in image segmentation is processed, and the network training process is more stable. 本发明提出基于UNet网络和注意力机制的脑胶质瘤分割方法，采用一种基于UNet网络模型改进的脑胶质瘤分割算法，在基础的UNet网络模型中添加注意力机制并改进损失函数，利用注意力机制对关键区域进行关注，提取图像中的关键信息同时忽略不重要信息，提升脑胶质瘤核磁共振图像分割的精度和效果，改进损失函数对图像分割中类别不平衡问题进行处理并使网络训练过程更加稳定。