基于3D-CSC的核磁共振脑组织图像分割方法

Segmentation of MR Brain Image Based on 3D-CSC

为了实现核磁共振脑组织图像的快速自动识别,同时克服灰度不均匀性和噪声影响,本文提出一种改进的基于3D Cell Structure Code(3D-CSC)技术的三维核磁共振脑组织图像的全自动分割方法.首先通过3D Kuwahara滤波器过滤图像中的噪声,利用基于图像灰度和空间信息的非参数偏场纠正方法抑制灰度不均匀性,然后利用多级并行的区域增长技术—3D Cell Structure Code(3D-CSC),对图像进行快速自动分割,并通过直方图分析方法实现了不同脑组织的分类,进而利用数学形态学运算进行后处理,提高了脑组织识别的准确度.通过在8个真实T1-加权MR图像和10组模拟T1-加权MR图像上的实验,验证了该方法的有效性.

To realize a fast and automatic brain tissue recognition, this paper presents an advanced method for 3 dimensional （3D） automatic segmentation of MR brain image, and the method is constructed based on the 3D Cell Structure Code （3D-CSC） and copes with noise and intensity inhomogeneity. Firstly, a 3D Kuwahara filter is applied to noise reduction, and a non-parametric bias correction method based on intensity features and spatial image features is used to compensate intensity inhomogeneity. Next, a 3D hierarchical parallel region growing technique, called 3D-CSC, is utilized for 3D automated segmentation of MR brain image. In order to recognize brain tissue, a histogram-based classification is applied. Finally, by taking morphological operations for postprocessing, the accuracy of brain recognition is further improved. Experiments implemented on 8 real and 10 simulated Tl-weighted MR brain images substantiate the effectiveness of our method.