基于生成模型约束的graph cuts标签融合算法

Generating model constraints graph cuts label fusion algorithm

人脑磁共振图像(magnetic resonance imaging,MRI)分割算法易受图谱错误标签的影响,为减少错误标签对标签融合准确度的影响,提高人脑磁共振图像的分割精度,在图谱预选阶段采用梯度信息和互信息筛选出相似性较大的图谱图像,对标签融合阶段提出利用生成模型(generative model,GM)约束的graph cuts标签融合方法,快速准确地分割出了人脑海马体结构。与其他的标签融合方法对比,提出的算法具有更高的分割精度。

The segmentation algorithm of brain magnetic resonance image is easily affected by error label. In order to reduce the influence of the error label to label fusion method,and improve the brain MR image segmentation accuracy,this paper proposed an atlases selecting method based on the gradient information and mutual information in the pre-selection stage. And proposed the method of GM constraints graph cuts in the label fusion stage,to rapidly and accurately segment the hippocampus.Compared with the others label fusion method,the proposed method has higher segmentation accuracy.