基于VV-Net的三维医学图像配准

3D medical image registration based on VV-Net

三维医学图像配准算法被广泛应用于科学研究和随访等医学场景,提高其配准精度具有重要的意义。针对医学图像配准问题,提出一种基于V-Net的V形网络(VV-Net),该配准模型可以通过堆叠V-Net进行端到端的训练。具体的说,移动图像经过两个V-Net依次进行扭曲,使用额外的V-Net为前两个V-Net提供补充信息,共同构成V形网络,使移动图像与固定图像更好的对齐。同时,对提出的模型增加深度监督辅助分支防止过拟合。采用上述渐进配准与信息补充提高配准对之间的配准精度。通过ADNI、ABIDE、ADHD200和OASIS四个数据集评估模型性能。以ADNI数据集为例,所提出的配准方法与仿射变换(Affine)、对称归一化(SyN)以及体素变形(VoxelMorph)比较分别获得24.7%、13.2%以及1.3%的精度提升。实验结果表明,VV-Net在医学图像配准领域取得了良好效果。

3D medical image registration algorithm is widely used in scientific research, follow-up and other medical fields, so it is of great significance to improve its registration accuracy. Aiming at the problem of medical image registration, a V-shape network based on V-Net(VV-Net) is proposed. The registration model can be trained end-to-end by stacking V-Net. Specifically, the moving image is distorted by two V-Nets in turn, and the additional V-Net is used to provide supplementary information for the first two V-Nets to form a V-shaped network, so that the moving image can be better aligned with the fixed image. At the same time, the depth supervision auxiliary branch is added to the proposed model to prevent over fitting. The accuracy of registration is improved by using the progressive registration and information supplement. The performance of the model is evaluated by ADNI、ABIDE、ADHD200 and OASIS data sets. Compared with Affine transformation, symmetric normalization(SyN) and VoxelMorph, the proposed registration method achieves 24.7%, 13.2% and 1.3% accuracy improvements, respectively. The experimental results show that VV-Net has achieved good results in the field of medical image registration.