多分辨率水平集算法的乳腺MR图像分割

Segmentation of breast MR images based on multiresolution level set algorithm

针对乳腺MR图像信息量大、灰度不均匀、边界模糊、难分割的特点,提出一种多分辨率水平集乳腺MR图像分割算法.算法的核心是首先利用小波多尺度分解对图像进行多尺度空间分析,得到粗尺度图像;然后对粗尺度图像利用改进CV模型进行分割.为了去除乳腺MR图像中灰度偏移场对分割效果的影响,算法中引入局部拟合项,并用核函数进一步改进CV模型,进而对粗尺度分割效果进行优化处理.仿真和临床数据分割结果表明,所提算法分割灰度不均匀图像具有较高的分割精度和鲁棒性,能够有效的实现乳腺MR图像的分割.

This paper proposes a novel multiresolution level set algorithm to segment breast MR images, which have a large amount of information, intensity inhomogeneities, and weak boundary. The core of the algorithm is to get the coarse scale image by analyzing the image in multi-scale space with wavelet multiscale decomposition. Then, to segment the analysed results in terms of improved CV model. In order to deal with the effect of bias field on the global images, the algorithm introduces a local fitting term into the improved CV model and optimizes the coarse-scale segmentation result by using the Kernel function to further improve the CV model. Experimental results on both synthetic and real breast MR images demonstrate that the proposed algorithm can segment the images with intensity inhomogeneity effectively and efficiently, also it can segment the images far more accurately, computationally efficiently, and much less sensitively to the initial contour.