基于双高斯空间模型的人脑MRI分割方法

A Brain MR Image Segmentation Method Based on Dual Gaussian Spatial Model

针对传统的基于高斯混合模型的分割方法由于忽略了空间信息和高斯模型间的线性组合导致对强噪声或边缘模糊的人脑MRI的分割效果并不理想的缺点,运用双高斯空间模型来对人脑MRI进行分割。由于双高斯空间模型拥有优良的空间相关性,能够很好的对脑部MRI纹理和边缘进行有效划分;因为人脑MRI的灰度分布呈现高斯分布特征,用双高斯空间模型对人脑MRI建模能够很好的符合人脑MRI的灰度分布特点;同时高斯权重置信度的指数结构以及平滑系数的引入使得分割鲁棒性大大增强;运用梯度下降法对双高斯空间模型的平滑系数进行求解,根据最大后验概率准则得到图像的最终分割结果。实验结果表明,该方法对人脑MRI具有很好的分割结果,同时鲁棒性与抗噪性能大大增强。

Due to the neglect of the neighborhood information,the traditional segmentation method based on Gaussian model is not ideal for the segmentation of the human brain MRI which contains the strong noise or the edge blurring.Since the excellent spatial correlation of dual Gaussian Spatial model,it can segment brain MRI textures and edges effectively.At the same time,the grey level distribution of the human brain MRI presents the characteristics of Gaussian distribution,so the dual Gaussian spatial model can be very good in accordance with the characteristics of the grey level distribution.Gradient descent method is used to solve the smoothing factor of the model,and finally the segmentation results are obtained according to the maximum a posteriori（MAP）criterion.The experimental result has shown that the method proposed has good segmentation effects for human brain MRI and robustness and anti noise ability is greatly enhanced.