基于灰度分布匹配的多模态脑部MR图像肿瘤分割算法

针对多模态核磁共振(MR)脑肿瘤图像的分割问题,提出一种基于灰度分布匹配的分割算法。首先,学习图像灰度强度的非参数模型分布,来描述当前图像的正常区域。然后,计算肿瘤图像中各区域之间的全局相似性,从中寻找灰度分布与学习模型最匹配的子区域,同时利用平滑操作来避免存在孤立区域。最后,对FLAIR模态图像进行处理,以分离出脑水肿区域,最终获取脑肿瘤区域的准确边界。在多模态脑肿瘤图像数据库BraTS 2012上进行实验,结果表明该算法能够准确且完整的分割出肿瘤区域。

Tumor Segmentation on Multi-Modality Brain MR Image Based on Grayscale Distribution Matching

For the issue that the tumor segmentation on multi-modality brain magnetic resonance (MR) image, a tumor segmentation algorithm based on grayscale distribution matching is proposed. Firstly, the distribution of non parametric model of image grayscale intensity is used to describe the normal region of the current image. Then, the global similarity between tumor image regions is calculated to find the sub-regions whose gray-level distribution is most consistent with the learning model. At the same time, the smooth operation is used to avoid the existence of the isolated region. Finally, the FLAIR modal images are processed to isolate the brain edema regions, and finally to obtain the accurate boundaries of brain tumor regions. Experiments are performed on multi-modality brain tumor image database: BraTS 2012, and the results show that the algorithm can accurately and completely separate the tumor region.