天津大学学报
天津大學學報
천진대학학보
Journal of Tianjin University
2015年
10期
937-946
,共10页
冯昌利%张建勋%梁睿%代煜%崔亮
馮昌利%張建勛%樑睿%代煜%崔亮
풍창리%장건훈%량예%대욱%최량
分形几何%最小凸包%肺结节%分割
分形幾何%最小凸包%肺結節%分割
분형궤하%최소철포%폐결절%분할
fractal theory%minimal convex hull%lung nodule%segmentation
在计算机辅助诊断系统中,为了缩小系统的分析范围、提高计算效率,需要将肺区域分割出来。但是通过已有方法获得的肺区域边界不准确,为此提出了一种基于分形几何和最小凸包法的肺区域分割算法。首先,根据肋骨和各组织的位置关系以及 CT 图像的上下层相似的性质,实现了对初始肺区域的自动提取。其次,利用网格线将肺区域分成小子块,并计算各子区域块的分形维数。根据肺区域边界的全局性质和局部性质,构造了最优的分形维数阈值,并根据该阈值识别需要修复的肺边界。最后,利用Jarvis步进法对肺边界进行了修复,从而在CT图像中获得了最终的肺区域。通过数值实验证明了提出的算法比传统方法更优秀,具有较高的分割准确率和较高的鲁棒性。
在計算機輔助診斷繫統中,為瞭縮小繫統的分析範圍、提高計算效率,需要將肺區域分割齣來。但是通過已有方法穫得的肺區域邊界不準確,為此提齣瞭一種基于分形幾何和最小凸包法的肺區域分割算法。首先,根據肋骨和各組織的位置關繫以及 CT 圖像的上下層相似的性質,實現瞭對初始肺區域的自動提取。其次,利用網格線將肺區域分成小子塊,併計算各子區域塊的分形維數。根據肺區域邊界的全跼性質和跼部性質,構造瞭最優的分形維數閾值,併根據該閾值識彆需要脩複的肺邊界。最後,利用Jarvis步進法對肺邊界進行瞭脩複,從而在CT圖像中穫得瞭最終的肺區域。通過數值實驗證明瞭提齣的算法比傳統方法更優秀,具有較高的分割準確率和較高的魯棒性。
재계산궤보조진단계통중,위료축소계통적분석범위、제고계산효솔,수요장폐구역분할출래。단시통과이유방법획득적폐구역변계불준학,위차제출료일충기우분형궤하화최소철포법적폐구역분할산법。수선,근거륵골화각조직적위치관계이급 CT 도상적상하층상사적성질,실현료대초시폐구역적자동제취。기차,이용망격선장폐구역분성소자괴,병계산각자구역괴적분형유수。근거폐구역변계적전국성질화국부성질,구조료최우적분형유수역치,병근거해역치식별수요수복적폐변계。최후,이용Jarvis보진법대폐변계진행료수복,종이재CT도상중획득료최종적폐구역。통과수치실험증명료제출적산법비전통방법경우수,구유교고적분할준학솔화교고적로봉성。
In the computer aided diagnosis systems,the lung region is segmented to reduce the analysis region and increase the computational efficiency. However,the boundary of the lung region obtained by the existing methods is not accurate,thus a lung region segmentation method based on the fractal theory and the minimal convex hull method was proposed. First of all,the lung region was extracted automatically according to the spatial context messages and the position relationship between the rib and other organs. Then the lung region was divided into several blocksby grid lines and the fractal dimension of each block was calculated. Besides,a fractal threshold,which was used to select the correction-needed blocks,was constructed by the global information and local information of the boundary. Fi-nally,the lung edge was corrected by using the Jarvis method. After that,the final lung region was obtainedin the CT image. The experiments demonstrate that the proposed method outperforms its traditional counterparts and it has higher segmentation accuracy rate and robustness.