核素肾动态显像肾脏感兴趣区域分割算法的研究

Study of a Novel Automatic Method of ROI Delineation in Renal Dynamic Radionuclide Imaging

目的提出一种新颖的核素肾动态显像肾脏感兴趣区自动勾画算法。方法选取30位行核素肾动态显像患者作为研究对象,首先对原始肾图做形态学运算、强度对增强和高斯滤波等预处理,去除非感兴趣区和提升图像对比度;接着采用最大类间差法(Otsu)自适应确定最佳阈值,完成肾脏的初步分割;然后形态学操作和边界追踪被用来提取肾脏边界。结果基于本文算法的分割结果与专家手工分割结果具有很高的相关性,平均真符合率达91%,平均假误符合率为13.4%,平均假符合率为9.3%,边界误差距离为1.6个像素。且基于本文算法的Dice相似性系数(0.9061±0.0196)和平均耗时(2.1477±0.2835)s均优于其他算法。结论基于本文的分割算法能快速准确的提取肾脏感兴趣区域,可应用于核素肾动态显像肾小球滤过率的测定。

Objective This paper aims to propose a novel method of ROI extraction which was used in renal dynamic imaging.Methods A total of30clinical dynamic renograms were introduced.The renal image was initially performed by morphological reconstruction followed by intensity-pair and Gaussian smoothing filter,which could sharpen the edge of kidney and suppress the noise.Then,adaptive Otsu threshold method was applied to segment the rough renal area.Finally,the morphological operation and boundary tracking were adopted to remove the irrelevant parts and extract the contour of renal.Results There was high correlation between physicians’manual contours and these by our approach.For area error analysis,the mean true positive area overlap,the mean false negative,the mean false positive and the boundary error were91%,13.4%,9.3%and1.6pixels,respectively.Our approach acquired larger Dice index(0.9061±0.0196)and lower computing time(2.1477±0.2835)s than other methods.Conclusion The proposed method is a feasible approach for ROI extraction in renal dynamic imaging,which can obtain more efficient,accurate and robust results.