基于可变形模板匹配的三维骨骼测量方法  

3D Bone Measurement Method Based on Deformable Template Matching

作　　者：王菲 崔青山 徐楷文 WANG Fei;CUI Qingshan;XU Kaiwen(College of Internet of Things Engineering,Hohai University,Changzhou 213000)

机构地区：[1]河海大学物联网工程学院,常州213000

出　　处：《计算机与数字工程》2025年第3期894-900,共7页Computer & Digital Engineering

摘　　要：计算机辅助术前规划中,正确提取骨骼医学语义可以帮助医生进行骨骼参数精确测量,进而判断骨骼是否畸形,以及确定其畸形程度。传统骨骼参数测量方法不能较好满足医生术前规划实际需求,费事费力且存在个体差异,难点在于如何准确提取骨骼语义。提出一种基于可变形模板匹配的三维骨骼测量方法,骨骼语义分割基于可变形模板匹配,匹配难点在于计算对应关系。利用神经网络预测骨骼模板与目标骨骼之间对应语义区域,提取语义区域中心点,构建特征点的稀疏对应关系,根据稀疏对应关系将骨骼模板拟合至目标骨骼;再将预标注在骨骼模板的语义区域映射至目标骨骼;最后,根据目标骨骼语义区域提取轴线,测量骨骼参数。该方法得到的语义区域与手工分割语义区域进行比较,实验结果表明,该方法得到的语义区域Dice系数即Dice Similarity Coefficient精度达到95%;通过该方法进行股骨角度测量,颈干角平均误差为1.4°,前倾角平均误差为0.9°。该方法能满足骨骼参数测量实际需求,有效提高骨骼畸形术前诊断与规划效率。In computer aided preoperative planning,the correct extraction of bone medical semantics can help doctors accurately measure bone parameters,and then judge whether there exist deformities,and determine the degree of deformity.Traditional bone parameter measurement methods can not meet the actual requirements of preoperative planning,which is laborious and individual differences.The difficulty lies in how to extract bone semantics accurately.This paper proposes a 3D bone measurement method based on deformable template matching.The semantic segmentation of bones is based on deformable template matching.The difficulty of matching is to calculate the corresponding relationship.The neural network is used to predict the corresponding semantic region between the bone template and the target bone,extract the center point of the semantic region,construct the sparse correspondence of feature points,and fit the bone template to the target bone according to the sparse correspondence,and then the semantic region of the pre-labeled bone template is mapped to the target bone.Finally,the axis is extracted according to the semantic region of the target bone,and the bone parameters are measured.The experimental results show that the accuracy of Dice coefficient of semantic region obtained by this method is 94%.the average error of neck shaft angle is 3.17°and the average error of anteversion angle is 2.62°by this method.This method can meet the actual needs of bone parameter measurement,and effectively improve the efficiency of preoperative diagnosis and planning of bone deformity.

关 键 词：计算机辅助术前规划 三维骨骼测量 医学语义分割 模板匹配 迭代拟合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]