快速脑白质损伤分割

本文提出了一种快速鲁棒的脑白质损伤分割算法。该算法能可靠地分割白质损伤区域。与大多数基于学习的算法不同，他们大多以专家标记的映射图为基础来学习分类器。本文方法基于如下事实：白质损伤区域的体素相对于正常体素具有较高的亮度。该方法只需要用户标出一些包含白质损伤的兴趣区域（ROI），然后算法首先执行一个K-means聚类，做初步的白质损伤区域标记。对标记的区域做进一步的学习获得一幅损伤区域的概率图，根据概率图产生一张置信图。最后引入SIFT算法中的多尺度处理的方法。在初始的标签的基础上，将依据邻近的多层切片的概率分布做进一步的标签的确认，最后得到较好的分割结果。

White Matter Lesion Segmentation

In this paper, we proposed a fast and robust WM lesion segmentation algorithm. It is reliable to WM lesion segmentation for MR brain images. It is different from most proposed algorithms, which exploit export labeled lesion to train a classifier. The algorithm we proposed only needs users to label some regions of interest （ROI） based on the truth that lesion voxels have higher intensity than those normal ones. We first apply k-means clustering for a roughly labels. This initial le- sion label is used to further refine the probability distribution estimation for the final lesion segmentation. Combining the ad- vantage of the multi-scale in sift algorithm, we use multi-slices＇ probability map to refine the final lesion segmentation.