一种基于脑MR图像的三层头模型自动分割方法

A method for automated segmentation of three-layer brain model from brain MR images

在求解脑电图EEG及脑磁图MEG正问题的过程中,常需要对人体组织中介电常数差距大的地方进行分层建模,以便更加准确的计算其传导矩阵。在人体脑部的各个组织中,颅骨和周围脑组织的介电常数差别尤其大。如何从脑MR图像中自动、准确地分割出颅骨部分,成为精确计算MEG/EEG正向传导矩阵的关键问题。而在磁共振图像(MRI)中,虽然软组织能被很清晰的成像,但颅骨却因为缺少氢而在图像中成像模糊,用传统的分割算法很难自动分割出准确的结果。为解决上述问题,本文提出一种结合脑组织平滑的先验信息和基于形变的曲面演化自动分割算法,来分割脑MR图像中的颅骨部分。再利用基于水平集的活动轮廓方法提取出头皮组织,进而构建出EEG及MEG正问题计算所需的三层真实头模型。我们将自动分割结果与手动分割结果进行了比较,证明了本文方法的有效性。

In the process of solving EEG / MEG forward problem,it is essential to build a multi-layer brain model which distinguishes different tissues,so as to calculate a more accurate lead field matrix. The skull and surrounding tissues are different in human brain. How to segment skull from MR images automatically and accurately is a key problem for calculating the MEG / EEG lead field matrix. Soft tissues are very clear in MR image,but the intensity of skull is uncertain due to the shortage of hydrogen in the skull tissue,consequently it is difficult to segment skull in MR image using traditional segmentation algorithm. In order to solve this problem,a segmentation algorithm in combination with deformable triangular surface with a prior information of the smoothness of brain tissues was proposed to segment the skull from MR images. Furthermore,through using level-set-based active contour method to extract the head skin,a three-layer realistic head model for calculating EEG and MEG forward problem was constructed. The experiment comparing the results of automated segmentation with that of manual segmentation proves the validity of our proposed method.