结合灰度直方图和细胞自动机的多模态MRI脑胶质瘤分割

为了解决脑胶质瘤边界模糊、复杂而导致的分割不准确问题，提出了一种将灰度直方图（gray level histogram，GLH）与改进的细胞自动机相结合的脑胶质瘤分割算法。首先，对脑胶质瘤的T2加权图像和液体衰减反转（FLAIR）图像进行融合；然后，利用灰度直方图特性增强脑胶质瘤区域；最后，以加权距离为特征向量用改进的细胞自动机进行分割，并得到脑胶质瘤各组织分割结果。在20组BraTS2015（brain tumor segmentation）数据库数据和10组临床脑胶质瘤数据上进行分割实验，整个肿瘤区域及核心肿瘤区域的平均分割准确率分别达到90.76%和89.73%。实验结果表明，相对于对比方法，所提算法不仅能更好地分割出对比度明显的胶质瘤区域，还在一定程度上解决了模糊胶质瘤区域分割不准确的问题。该算法在保持不增加算法复杂度的同时，亦提高了算法分割的准确性和鲁棒性。

Brain glioma segmentation for multi-modality MR images based on gray level histogram and cellular automata

The fuzzy and complex glioma boundary can cause inaccurate segmentation of the glioma. In order to solve this problem, this paper proposed a new glioma segmentation algorithm combining GLH with improved cellular automaton. Firstly, this method fused T2-weighted and fluid attenuated inversion recovery MR images of brain glioma. Then, it used the histogram feature to enhance glioma region. And, it calculated the weighted distance eigenvector of glioma images. Finally, it utilized the improved algorithm of cellular automata to obtain the segmentation result of glioma tissues. It separately segmented twenty groups of brain tumor segmentation database data and ten groups of clinical glioma data. The average segmentation accuracy rate of the entire tumor area and core tumor area reached to 90.76% and 89.73% respectively. The experimental results show that compared with the contrast method, the proposed algorithm can better segment the glioma region with obvious contrast. And it solves the problem of inaccurate segmentation due to the fuzzy glioma region to some extent. While, it also improves the accuracy and robustness without increasing the complexity.