【作者】 黄恒；

【导师】 宋恩民；

【作者基本信息】 华中科技大学 ， 计算机技术， 2021， 硕士

【摘要】 部分接受全髋关节置换术的患者,在术后一段时间出现病理情况需要进行二次翻修。在翻修术前,需要通过患者CT图像获取关键解剖点和解剖轴的参数信息,辅助医生进行术前规划,提高手术准确度。根据临床需求,确定课题任务,即定位髋部假体球头和臼窝中心、拟合假体柄轴线和颈轴线。鉴于要定位的点、线并非解剖结构上的显著点,且病例图像较少的情况,充分利用髋关节假体的特征,制定了基于传统图像处理的定位方案。算法主要分为三个步骤:数据预处理、基于球头和髋臼窝表面点拟合球头和髋臼窝中心、基于柄轴和颈轴中心点拟合柄轴线和颈轴线。在预处理阶段,针对数据中存在金属伪影、分辨率不统一、假体分布情况多样的问题,对数据进行金属伪影校正、等方形重采样和左右侧切分数据。采用阈值分割法对假体进行分割,并使用连通区域标记来获取臼窝和股骨假体ROI范围和去除小连通域。在定位球头中心时,针对假体形态特征确定球头上半球在Z轴方向上的上界和下界,并利用边界跟踪算法提取每个球头上半部分切面的边界。在定位髋臼中心时,充分利用髋臼窝和假体的相对位置关系,利用球头中心发出平行线寻找髋臼窝内表面点,最后用最小二乘法将得到的球体表面点进行球体拟合。在进行假体柄轴线和颈轴线拟合时,计算假体掩码内与目标轴线垂直的法平面质心作为直线拟合的观测点。对于柄轴线,其法平面在假体内的体现为股骨柄下半部分每个Z轴切面,计算这些切面的质心作为柄轴线拟合的样本点。对于颈轴线,在原图中以球头球心为棱中点,裁取出小矩形,并在小矩形中以球头球心为球心生成不同半径空心球体,以这些球体与股骨颈的交面作为法平面并计算其质心点。最后用随机抽样一致性算法将得到的直线样本点进行直线拟合。用已有数据集对定位算法进行测试,实验结果表明能得到误差较小的中心定位结果和符合解剖定义且偏差较小的轴线拟合结果。更多还原

【Abstract】 Some patients who have undergone total hip arthroplasty have pathological conditions that require a second revision.Before revision,it is necessary to obtain the parameter information of key anatomical points and anatomical axes from patients’ CT images to assist doctors in preoperative planning and improve surgical accuracy.According to the clinical needs,the task was determined,which was to locate the bulb and the center of the socket of the hip prosthesis,and to fit the axis of the stem and the axis of the neck of the prosthesis.In view of the fact that the points and lines to be located are not significant points on the anatomical structure and there are few case images,a localization scheme based on traditional image processing was developed by making full use of the features of the hip prosthesis.The algorithm can be divided into three steps: data preprocessing;fitting the center of the bulb and the acetabular fossa based on the surface points of the bulb and the acetabular fossa;fitting the axis of the stem and the axis of the journal based on the center points of the stem and the cervical shaft.In the pre-processing stage,in view of the problems of metal artifacts in the data,the resolution is not uniform,and the distribution of the prosthesis is varied,the data are corrected by metal artifacts,resampling by equal square,and segmented by left and right sides.Threshold segmentation method was used to segment the prosthesis,and connected region markers were used to obtain the ROI range of the socket and femoral prosthesis and remove the small connected domain.When locating the center of the ball head,the upper and lower bounds of the hemisphere on the Z-axis were determined according to the morphological characteristics of the prosthesis,and the boundary tracking algorithm was used to extract the boundary of the half section of each hemisphere.When locating the center of the acetabulum,the relative position relationship between the acetabulum fossa and the prosthesis was made full use of,and parallel lines were sent out from the center of the ball head to find the internal surface points of the acetabulum fossa.Finally,the spherical surface points obtained were fitted by the least square method.The centroid of the normal plane perpendicular to the target axis in the prosthesis mask was calculated as the observation point of straight line fitting during the fitting of the stem axis and the neck axis.For the axis of the stem,its normal plane in the prosthesis is reflected by each Z-axis section of the lower part of the femoral stem,and the centroid of these sections is calculated as the sample point for the axis fitting of the stem.For the cervical axis,the ball head center was taken as the midpoint of the edges in the original image,and the small rectangle was cut out.In the small rectangle,the ball head center was taken as the center of the ball to generate hollow spheres of different radii.The intersection planes between these spheres and the femoral neck were taken as the normal plane and the centroid points were calculated.Finally,the line sample points are fitted by the random sampling consistency algorithm.The existing data sets are used to test the location algorithm,and the experimental results show that the center location results with small errors and the axis fitting results with small deviations can be obtained.更多还原