【摘要】 脑肿瘤MRI图像分割是脑肿瘤诊断和治疗的重要环节。对于脑肿瘤MRI医学图像存在难以精确分割的问题，在U-Net网络分割方法基础上进行了改进，于编码路径-解码路径的长连接中引入注意力模块，使网络模型关注需要分割区域的特征，避免信息冗余，以达到脑肿瘤图像精准分割的目的。此外，还提出一种基于Dice损失和焦点损失的混合损失函数，用以解决类不平衡问题，提高对肿瘤核心区域的分割效果。将改进模型及改进混合损失函数在BraTS2018和BraTS2019上进行实验。通过分析表明，与传统的U-Net相比，提出的分割方法在脑肿瘤不同区域的Dice值、精准率、敏感度均有提升，拥有更好的性能。更多还原

【Abstract】 MRI image segmentation of brain tumors is an important step in the diagnosis and treatment of brain tumors. The difficulty of accurate segmentation is a widespread problem in the medical image segmentation field of brain tumor MRI. In order to solve this problem, it is improved on the basis of U-Net network segmentation method, attention modules are introduced into the long connection between coding path and decoding path of U-Net network, so that the network model can pay attention to the characteristics of the region to be segmented and avoid information redundancy. In addition, a mixed loss function based on Dice loss and focal loss is proposed to solve the problem of sample imbalance and improve the segmentation effect of tumor core region. The improved model and the improved mixed loss function are tested on brats 2018 and brats 2019. The experimental analysis shows that compared with the traditional U-Net, the dice value, accuracy and sensitivity of the proposed segmentation method in different regions of brain tumors are improved, and the proposed method has better performance.更多还原