组织工程与重建外科杂志

组织工程与重建外科杂志 ›› 2026, Vol. 22 ›› Issue (1): 16-.

• • 上一篇    下一篇

下颌骨三维统计形状模型的建立及其临床应用

  

  • 出版日期:2026-01-29 发布日期:2026-03-05

Establishment of a three-dimensional statistical shape model of the mandible and its clinical applications

  • Online:2026-01-29 Published:2026-03-05

PDF (PC)

3

摘要：

目的 基于中国人群的头颅CT数据建立高精度的下颌骨三维统计形状模型（Statistical shape model，SSM），并探讨其在形态分析、缺损自动重建及基于面部形态预测下颌骨形态等方面的临床应用。方法 收集130例健康汉族成年人的头颅CT数据，其中100例用于模型训练，30例用于验证。经Mimics及Meshmixer进行三维重建、标准化和非刚性配准后，使各模型拓扑一致并进行奇异值分解（SVD）建立SSM。通过主成分分析（PCA）提取形态特征并进行性别组间比较。结合后验形状模型（Posterior shape model，PSM）算法实现下颌骨缺损的自动重建；同时，基于面部SSM与下颌骨SSM之间的相关性，利用支持向量回归（Support vector regression，SVR）建立面部形态预测下颌骨形态的模型。结果 所建立的下颌骨SSM前15个主成分可解释95.18%的形状变异，拟合测试集的平均误差为（0.55±0.08） mm，显示出良好的泛化性。性别相关差异主要集中在整体体积及下颌角角度等特征。基于PSM的自动重建在角区及体部缺损中的精度高于手工重建（RMSE分别为0.77 mm与0.67 mm，Dice系数分别为0.87与0.94）。基于面部三维形态预测下颌骨形态的平均误差为（1.50±0.34） mm，显著低于随机人群之间的差异［（2.58±0.98） mm，P<0.001］。结论 本研究首次构建了具有高精度、强泛化性且适用于中国人群的下颌骨三维SSM，并证实其在形态分析、缺损自动重建及基于面部形态预测下颌骨形态中的临床可行性。该模型为颌面外科的个体化手术设计、无放射性评估及远程医疗提供了新的技术支撑和研究基础。

关键词:

Abstract:

Objective To establish a  high-precision three-dimensional statistical shape model （SSM） of the mandible based on cranial CT data from a Chinese population， and to explore its clinical applications in morphological analysis， automated reconstruction of mandibular defects， and prediction of mandibular morphology based on facial shape. Methods Cranial CT data from 130 healthy Han Chinese adults were collected， of which 100 cases were used for model training and 30 for validation. Three-dimensional reconstruction， standardization， and non-rigid registration were performed using Mimics and Meshmixer to ensure topological consistency across models， followed by singular value decomposition （SVD） to construct the SSM. Principal component analysis （PCA） was employed to extract morphological features and compare differences between sexes. A posterior shape model （PSM） algorithm was integrated to achieve automated reconstruction of mandibular defects. In addition， leveraging the correlation between a three-dimensional facial SSM and the mandibular SSM， a support vector regression （SVR） model was developed to predict mandibular morphology based on facial shape. Results The first 15 principal components of the mandibular SSM explained 95.18% of shape variance. The average fitting error on the test set was  （0.55±0.08） mm， indicating strong generalization. Sex-related differences were primarily observed in overall volume and gonial angle characteristics. PSM-based automated reconstruction achieved higher accuracy than manual reconstruction in angle and body defects （RMSE of 0.77 mm and 0.67 mm； Dice coefficients of 0.87 and 0.94， respectively）. The mean error for predicting mandibular morphology from three-dimensional facial shape was （1.50±0.34） mm， significantly lower than inter-individual differences in a random population of （2.58±0.98） mm （P<0.001）. Conclusion This study presents the first high-precision， strongly generalizable three-dimensional mandibular SSM tailored to a Chinese population， and demonstrates its clinical feasibility for morphological analysis， automated defect reconstruction， and facial-shape-based prediction of mandibular morphology. The model provides new technical support and a research foundation for personalized maxillofacial surgical planning， radiation-free assessment， and telemedicine.

Key words: