基于双支路特征融合的MRI颅脑肿瘤图像分割研究

Research on MRI brain tumor image segmentation based on dual-branch feature fusion

针对磁共振成像(magnetic resonance imaging, MRI)颅脑肿瘤区域误识别与分割网络空间信息丢失问题,提出一种基于双支路特征融合的MRI脑肿瘤图像分割方法。首先通过主支路的重构VGG与注意力模型(re-parameterization visual geometry group and attention model, RVAM)提取网络的上下文信息,然后使用可变形卷积与金字塔池化模型(deformable convolution and pyramid pooling model, DCPM)在副支路获取丰富的空间信息,之后使用特征融合模块对两支路的特征信息进行融合。最后引入注意力模型,在上采样过程中加强分割目标在解码时的权重。提出的方法在Kaggle_3m数据集和BraTS2019数据集上进行了实验验证,实验结果表明该方法具有良好的脑肿瘤分割性能,其中在Kaggle_3m上,Dice相似系数、杰卡德系数分别达到了91.45%和85.19%。

To address the problem of MRI brain tumor region misidentification and spatial information loss of segmentation network, an MRI brain tumor image segmentation method based on dual-branch feature fusion is proposed.First, the contextual information of the network is extracted by structurally the re-parameterization visual geometry group and attention model(RVAM) in the primary branch, and then the rich spatial information is obtained in the secondary branch using deformable convolution and pyramid pooling model(DCPM),after which the feature fusion module is used to fuse the feature information of the two branches.Finally, the attention model is introduced to strengthen the weight of segmented targets in the up-sampling process at decoding.The proposed method has been experimentally validated on the Kaggle_3 m and BraTS2019 datasets, and the experimental results show that our method has good brain tumor segmentation performance, where the Dice similarity coefficient and Jaccard coefficient reach 91.45% and 85.19% on Kaggle_3 m, respectively.