基于证据推理的区域合并用于交互式的医学图像分割

Evidential Reasoning-based Region Merging for Interactive Medical Image Segmentation

在CT图像中,精确地分割脂肪组织对治疗诸如癌症等疾病具有非常重要的作用。与此同时,如果医学图像分割中允许医师的介入,将会得到更好的分割结果。基于此,本文提出一种新的基于证据推理的区域合并方法,用于交互式的医学图像分割。该方法在初始化后,目标区域与其邻接区域的相似性利用证据推理方法计算得到。如果目标区域与某一个邻接区域的相似性最大,那么这2个区域将合并成为一个区域。实验结果表明所提算法在视觉和定量分析上均能取得好的分割性能。

When preoperatively diagnosing the gastric cancer,extracting regions which include lymph nodes in CT images accurately is very important for doctors. Meanwhile,it is better to allow direct intervention of doctors to make necessary corrections when needed. Due to above problems,a novel evidential reasoning based region merging（ ERRM） method for interactive image segmentation is proposed. ERRM not only can extract object regions effectively,but also allow direct intervention of doctors. After initial segmentation,the similarity between the target region and its adjacent regions will be calculated by evidential reasoning（ ER）. Two regions will be merged if they have the highest similarity. The experimental results show that ERRM can obtain better performance than other methods in both qualitative and quantitative analysis.