\r鲍 潮，姜 杉，张国彬，杨志永\r
\r
AuthorsHTML:\r鲍 潮，姜 杉，张国彬，杨志永\r
\r
AuthorsListE:\rBao Chao，Jiang Shan，Zhang Guobin，Yang Zhiyong\r
\r
AuthorsHTMLE:\rBao Chao，Jiang Shan，Zhang Guobin，Yang Zhiyong\r
\r
Unit:\r天津大学机械工程学院，天津 300350\r
\r
Unit_EngLish:\rSchool of Mechanical Engineering，Tianjin University，Tianjin 300350，China\r
\r
Abstract_Chinese:\r为了获得CT 引导的肺部近距离放射治疗术中实际针道位置并且辅助术中顺利进行剂量优化步骤，本文提出了一种自动针道重建算法．基于预检验模型参数评估技术、局部随机样本一致性(RANSAC)算法与主成分分析(PCA)法相结合的局部优化算法，所提出的改进的RANSAC 算法能够快速、精确、稳定地将针轴提取出来．相较于传统的RANSAC 算法，本文首先运用预检验技术加速了传统算法的迭代过程，以便于提高整体拾取效率；其次为了解决传统算法拾取精度较差的问题，先通过局部随机样本一致性算法获得一个近似较优解，而后利用PCA 法最小化所有局内点的误差以得到最终的最优解．针尖点的具体位置可以通过搜索针轴方向灰度的变化而确定. 此外，同一张切片上的多根针通过连续局内点删除算法能够一次性被分割出来．模拟数据结果表明，改进后算法的分割精度要明显高于传统的RANSAC 算法所得的结果，并且随着局内点占比下降拾取时间并没有发生较大的上升．此外，本文的算法也在实际的肺部近距离放射治疗图像中得到了验证，实际病例实验中以手动分割的结果作为算法所得的结果的参照物．实验结果表明，针轴方向和针尖点的平均误差分别被控制在1.4°和0.85 mm 以内，且平均每根针的分割时间为0.238 s．\r
\r
Abstract_English:\rTo obtain the needle placement and optimize dose distribution in computed tomography(CT)-guided lung interstitial brachytherapy，an automatic needle reconstruction algorithm is proposed. Based on the pre-test model parameters evaluation technique and local optimization involving local random sample consensus(RANSAC) algorithm and principal component analysis(PCA) technique，the proposed modified RANSAC algorithm can be used to detect the needle shaft quickly，accurately，and robustly. Compared with the general RANSAC algorithm，the pretesting technique accelerates the iteration process and consequently improves the overall needle segmentation efficiency. Moreover，to solve the problem of poor pick-up accuracy of the traditional algorithms，the local RANSAC algorithm is used to obtain an approximate optimal solution，and then the PCA technique is applied to obtain the desired model by minimizing the error estimation of the inliers. By tracing intensities along the axis，the needle tip is determined. Furthermore，multiple needles in a single slice are segmented at once using successive inliers deletion. The simulation data show that the segmentation accuracy is much higher than traditional RANSAC algorithm；however，the segmentation time did not increase significantly with decreasing inlier ratio. The application of the proposed algorithm to permanent lung brachytherapy image is also validated，where the results of manual segmentation is used as reference for the estimated needle shape. From the results，the mean errors in determining needle orientation and endpoint are regulated within 1.4° and 0.85 mm，respectively，and the average segmentation time is 0.238s per needle.\r
\r
Keyword_Chinese:肺部近距离放射治疗；CT影像；随机样本一致性算法；针道识别\r

Keywords_English:lung brachytherapy；CT image；random sample consensus algorithm；needle detection\r


PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=6210