基于纹理和几何特征融合的髁突骨微结构分割与预测  

Bone Microstructure Segmentation and Prediction of Condyle Based on Fusion of Texture and Geometric Features

作　　者：李文轩 冯斌 吴福理[1] 金琳筱 汪郡 张伟华[2] Li Wenxuan;Feng Bin;Wu Fuli;Jin Linxiao;Wang Jun;Zhang Weihua(School of Computer Science and Technology,Zhejiang University of Technology,Hangzhou 310023;Department of Oral and Maxillofacial Radiology,Stomatology Hospital,School of Stomatology,Zhejiang University School of Medicine,Hangzhou 310006)

机构地区：[1]浙江工业大学计算机科学与技术学院,杭州310023 [2]浙江大学医学院附属口腔医院口腔颌面放射科,杭州310006

出　　处：《计算机辅助设计与图形学学报》2025年第2期340-348,共9页Journal of Computer-Aided Design & Computer Graphics

摘　　要：下颌骨髁突是颞下颌关节的重要组成部分.颞下颌关节炎(TMJOA)患者常发生髁突异常改变,提示TMJOA病变进展.针对现有工作对于髁突骨微结构的研究不深入,预测髁突病变前的形态比较缺乏的问题,提出一种锥形束计算机断层扫描(CBCT)影像中髁突骨微结构分割方法.首先结合髁突的几何特征,通过数据扩充模块增加训练样本数;然后采用边缘提取模块有效地捕捉髁突边界信息,通过纹理特征提取模块获取影像中的纹理特征;最后将边界信息与解码器中得到的特征图进行融合,获得不同尺度的边缘增强区域特征图,增强模型对纹理和边缘的感知.在髁突CBCT数据集上对髁突骨微结构分割的实验结果表明,所提方法在髁突骨微结构的分割中取得了良好的效果,皮质部分的分割Dice达到0.901,松质部分的分割Dice达到0.958;在病变前皮质形态预测的实验结果表明,预测结果的Dice值为0.948,验证了该方法比文中对比方法具有更好的性能.所提方法为TMJOA病变的影像诊断和治疗提供了有力的支持和指导.The mandibular condyle is an important part of the temporomandibular joint.Temporomandibular arthritis(TMJOA)patients often have abnormal condylar changes,suggesting the progression of TMJOA lesions.Due to the lack of in-depth research on the microstructures of condylar bone in existing studies and the lack of prediction of condylar morphology before pathological changes,a method for segmenting the microstructures of condylar bone in cone-beam computed tomography(CBCT)images is proposed.First,the geometric features of the condyle are combined with a data augmentation module to increase the number of training samples;then,the edge extraction module effectively captures the boundary information of the condyle,and the texture feature ex-traction module obtains texture features from the image;finally,the boundary information is fused with the fea-ture map obtained from the decoder to obtain edge-enhanced region feature maps of different scales,enhancing the model’s perception of texture and edges.The experimental results of condylar bone microstructure segmentation on the condylar CBCT dataset show that the proposed method has achieved good results in segmenting the microstructures of the condylar bone,with a Dice coefficient of 0.901 for the cortical part and 0.958 for the cancellous part.The experimental results of morphology prediction of the cortical part before pathological changes show that the prediction result has a Dice coefficient of 0.948,verifying that the proposed method has better per-formance than the comparative methods.The proposed method provides strong support and guidance for the im-aging diagnosis and treatment of TMJOA lesions.

关 键 词：髁突锥形束计算机断层扫描 髁突骨微结构分割 髁突病变前皮质预测 边缘检测 纹理特征 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]