基于隐式形状配准与新型多表面图搜索的肾皮质分割的关键算法研究

The renal cortical volume and thickness have been certificated to be good biomarkers for renal function in a number of clinical situations. They can be widely applied in urological treatment decision-making and assessment of clinical outcomes of surgery. The exact renal cortex image can be useful for clinical diagonasis. However, since the boundaries between the renal cortex and its adjacent organs are weak, and the renal pelvis makes the renal cortex a nonfully closed structure,current semgnetation methods are not enough to achieve the accurate renal cortex. In order to cover the shortages,the project aims to achive the accurate renal cortex image, researches the renal cortex segmenation, and based on the actual image data, constructs the baseline of a normal feature set with mathematical statistical methods, which will be helpful for clinical diagnosis. The content includes the following parts: First, the renal cortex is roughly initialized using an implicit shape registration method, which will be useful for the automation of the following segmentation. Second, the inner and outer surfaces of renal cortex are taken with novel multiple surfaces graph searching, which is extended to use varying sampling distances and physical constraints to segment the renal cortex and the renal column. Third, a refining method for renal cortex is made to detect and reduce incorrect segmentation pixels around the renal pelvis, further improving the segmentation accuracy. In final, mathematical statistical methods are constructed to achieve successfully the features' baseline of the renal cortical features such as volume and thickness. There are good prospects of clinical application based on the research results of this project.

肾皮质体积和厚度被证明是许多临床诊断中肾功能的有效生物标记，其应用包括泌尿学的治疗决策和外科临床结果的评估。准确的肾皮质图像能为辅助临床诊断提供依据。但因肾皮质与周边器官的边界很弱且肾皮质不是一个非全闭合结构这些特殊特质，目前的肾皮质分割算法还存在不足。为弥补这些不足，本项目以获取精确的肾皮质图像为目标，研究肾皮质分割算法，并在实际数据基础上通过数学统计方法建立肾皮质正常特征集的基线范围，为计算机辅助临床诊断提供依据。其研究内容包括：1)提出一个基于隐式形状配准的方法初始化肾皮质以便肾皮质分割全自动化。2)提出采用变的采样间距和物理分离约束的新型多表面图搜索方法更好地分离肾皮质和肾柱，以抽出肾皮质的内外表面。3)提出改进的肾皮质分割方法检测和减少肾盂周围不正确的分割像素。4)构建肾皮质的数学统计模型，建立肾皮质特征如体积和厚度的基线范围。基于本项目的研究成果具有很好的临床应用前景。

对肾皮质结构的充分了解可以辅助许多临床肾功能的诊断。但因肾皮质不是全闭合的且与周边器官的边界很弱，目前的肾皮质算法还难以获取准确的肾皮质。为获取精准的肾皮质，本项目研究新的分割算法分割肾皮质，并结合临床数据对肾皮质进行分析，以辅助肾皮质的临床诊断。其研究内容为：首先提出一个基于隐式形状配准的方法全自动化分割肾皮质；然后采用新的多表面图搜索方法分离肾皮质和肾柱；接着用图割方法去掉肾盂周围不正确部分；最后结合临床数据构建肾皮质的数据统计模型以分析肾皮质。并将这些算法扩展到肺部和宫颈的研究上。项目开展三年以来，项目组已在IEEE Trans Med Imaging和Pattern Recognition等权威期刊上发表文章6篇，相关专利及更多文章会在随后一年中发表出来。本项目成员多次参加国际会议，去国外研究中心及要求其他团队的人员来华对本研究内容进行讨论与交流。总之上述工作具有较好的临床应用前景。