基于ITK多图谱配准的脑部皮质下核团的分割研究

Multi-atlas Segmentation of Brain Subcortical Structures Based on ITK Registration Framework

大量研究表明,阿尔茨海默症(AD)的病变与大脑皮质下核团的萎缩息息相关,某些核团的萎缩(如海马)可能成为AD疾病早期诊断的标志,而皮质下核团的分割是研究核团萎缩模式的重要前提。基于AD患者和正常人各30例3DT1W-MR图像,先结合直方图分析和三维形态学分析方法对图像进行脑组织提取,后采用ITK配准算法将10个脑图谱图像经两阶段分别配准到提取脑组织后的图像空间。第一阶段实现基于均方差的仿射配准,第二阶段实现基于互信息的B样条形变配准,两阶段的配准均采用线性插值法和梯度下降的优化搜索方法。最后采用STAPLE融合算法,对配准后得到的10个目标图像进行图像融合,得到最终的分割结果。结果表明:除尾状核外,分割得到的其余6对核团的体积与常用的FSL-FIRST算法的分割结果无统计学差别(P>0.05);AD患者的右侧伏核和双侧海马发生萎缩(P<0.05)。因此,基于ITK配准框架的多图谱配准分割方法能有效分割MR图像上边界不明确的皮质下核团。

Numerous studies indicated that the process of Alzheimer＇s disease （AD） was closely correlated with atrophy of subcortical structures. Atrophy of some subcortical structures like hippocampi, could be a biomarker of early diagnosis of AD, which makes the segmentation of subcortical structures very important. Based on 3DT1W-MR images of 30 AD and 30 normal controls, we firstly performed brain tissue extraction combining histogram analysis and three-dimensional morphological processing method, then registered 10 brain atlas to＇ the preprocessed image based on ITK registration framework. A two-stage images registration was applied to register multiple atlas on the subject MR scan. First, we applied affine registration with mean squares as metric. Second, B-spline transform based on mutual information model was applied to further registration. In the two- stage registration method, linear interpolation model and the optimizer named regular step gradient descent were used. After registration, STAPLE algorithm was achieved to perform image fusion on the obtained 10 registered images to get the final segmented image. The results show that all volumes of subcortical structures except caudate show no statistical differences between our method and FSL-FIRST algorithm （P 〉 0.05） ; bilateral hippocampus and right nucleus accumbens suffer atrophy in AD （P 〈 0.05）. Our experimental data shows the validity of the multi-atlas-based segmentation method on segmentation of the brain subcortical structures.