Abstract
目的
建立一种基于牙颌面畸形患者三维颅面特征的相似性度量模型，并通过专家相似度评分对度量模型的有效性进行检测。
方法
选取2020年1月至2022年12月在北京大学口腔医院行双颌手术及术前正畸治疗的骨性Ⅲ类牙颌面畸形患者52例，其中男性26例，女性26例，根据性别分为两组。每组各随机设置1例患者作为参考样本，该组内其余患者均为测试样本。由3位专家对测试样本与参考样本的相似度进行主观评分，评分范围为1~10分，其中1分为完全不同，10分为完全相同，设定7.5分为临床可接受的相似性结果。提取患者术前锥形束计算机断层扫描(cone beam computed tomography, CBCT) 和三维面部图像的三维硬、软组织颅面特征，包括距离、角度和三维点云特征等，采用特征选择算法和线性回归模型，并与专家相似度评分结果进行拟合，建立相似性度量模型。为验证模型的可靠性，选取14例新患者进行相似度匹配，并由专家评价匹配结果的相似度，以评价相似性度量模型的可靠性。
结果
相似性度量模型显示，面中、下颅面特征是影响颅面相似度的主要特征，包括前鼻棘点-颏下点(anterior nasal spine-menton，ANS-Me)距离、右上尖牙点至眶耳平面(right canine-Frankfurt horizontal plane，U3RH)距离、左髁顶点-左下颌角点(left superior point of condyle-left gonion, CoL-GoL)距离、左髁顶点-颏下点(left condyle-menton, CoL-Me)距离、颏前点至正中矢状面垂直(pogonion-midsagittal plane, Pog-MSP)距离、右鼻翼点-左鼻翼点(right alar base-left alar base, AlR-AlL)距离、鼻尖点-软组织颏前点-下唇点(pronasale-soft tissue pogonion-labrale inferius, Pn-Pog’-Li)交角、发际点-右侧耳屏点(trichion-right tragus, Tri-TraR)距离、左外眦点-左鼻翼点(left exocanthion-left alar base, ExL-AlL)距离、骨性面下1/3、骨性面中下2/3及软组织上唇区域等。在模型可靠性测试中，14例相似性匹配案例的平均相似度评分为(7.627±0.711)分，与7.5分差异无统计学意义。
结论
本研究使用的相似性度量模型寻找的相似案例与专家主观评价匹配度高，可用于骨性Ⅲ类患者的相似案例检索。
Keywords: 牙颌面畸形, 头颅相似性, 三维形态特征
Abstract
Objective
To establish a similarity measurement model for patients with dentofacial deformity based on 3D craniofacial features and to validate the similarity results with quantifying subjective expert scoring.
Methods
In the study, 52 cases of patients with skeletal Class Ⅲ malocclusions who underwent bimaxillary surgery and preoperative orthodontic treatment at Peking University School and Hospital of Stomatology from January 2020 to December 2022, including 26 males and 26 females, were selected and divided into 2 groups by sex. One patient in each group was randomly selected as a reference sample, and the others were set as test samples. Three senior surgeons rated the similarity scores between the test samples and the reference sample. Similarity scores ranged from 1 to 10, where 1 was completely different, and 10 was exactly the same. Scores larger than 7.5 was considered as clinically similar. Preoperative cone beam computed tomography (CBCT) and 3D facial images of the patients were collected. The three-dimensional hard and soft tissue features, including distances, angles and 3D point cloud features were extracted. The similarity measurement model was then established to fit with the experts' similarity scoring by feature selection algorithm and linear regression model. To verify the reliability of the model, 14 new patients were selected and input to similarity measurement model for finding similar cases. The similarity scoring of these similar cases were rated by experts, and used to evaluate the reliability of the model.
Results
The similarity metric models indicated that the features of the middle and lower craniofacial features were the main features to influence the craniofacial similarity. The main features that were related to the expert' s similarity scoring included distance of anterior nasal spine-menton (ANS-Me), distance of right upper canion point-Frankfurt horizontal plane (U3RH), distance of left superior point of the condyle-left gonion (CoL-GoL), distance of left gonion-menton (CoL-Me), distance of pogonion-midsagittal plane (Pog-MSP), distance of right alar base-left alar base (AlR-AlL), angle of pronasale-soft tissue pogonion-labrale inferius (Pn-Pog' -Li), distance of trichion-right tragus (Tri-TraR), distance of left exocanthion-left alar base (ExL-AlL), lower 1/3 of skeletal face, middle and lower 2/3 of skeletal face and upper lip region of soft tissue. Fourteen new patients were chosen to evaluate the model. The similar cases selected by the model had an average experts' similarity scoring of 7.627± 0.711, which was not significantly different with 7.5.
Conclusion
The similarity measurement model established by this model could find the similar cases which highly matched experts' subjective similarity scoring. The study could be further used for similar cases retrieval in skeletal Ⅲ malocclusion patients.
Keywords: Dentofacial deformities, Craniofacial similarity, 3D morphological features
正颌外科手术通过改变上、下颌骨之间的相对位置以及它们相对于颅骨的位置能够改善患者咬合、气道等功能，同时改善面形。近年来，随着牙颌面畸形患者对正颌外科术后面形改善的需求不断增高[1]，颅骨、软组织形态改善成为正颌外科设计的关键依据。在缺少软组织细节评价指标和精准预测方法的情况下，先验相似病例的实际术后效果能够为医生提供细节、真实的手术方案和术后效果参考。
基于正常人头颅的手术设计已被用于骨折复位、创伤继发畸形矫治以及颅颌面缺损修复重建等疾病的诊疗中[2-5]。该方法的核心是利用三维形貌数据检索的方法，先在正常人数据库中寻找与患者相似的头颅样本，再以该样本为依据对患者进行骨折复位、缺损重建等。然而，该治疗思路并不适用于骨性牙颌面畸形患者的治疗设计，首先，牙颌面畸形患者除了异常的骨性形态外，其软组织也与正常人有所差异，以正常人骨骼结构为模板进行手术设计获得的轮廓效果可能并不尽如人意，以骨性Ⅲ类牙颌面畸形患者为例，相似的面中部骨组织形态前提下，软组织可能出现发育不足或代偿性发育过度两种表现；其次，牙颌面畸形患者的面上、面中结构也存在异常，面下1/3的手术设计可能需要妥协或者做骨骼容积补偿，达到全脸和谐整体美观，因此，寻找软硬组织形态均相似的牙颌面畸形患者作为设计依据比正常人头颅更为合理。
相似度评价是寻找相似头颅的重要工具，既往研究基于正常人头颅骨骼结构建立检索算法模型[3, 5-8]，缺乏软组织的相似度匹配，不满足牙颌面畸形患者相似度检索需求，因此，本研究设计一种基于软硬组织标记点和专家主观测评相结合的相似度评价模型，以获取更适宜临床使用、满足患者需求的手术方案，并通过专家主观测评，检验该模型筛选相似头颅的可靠性。
1. 资料与方法
本研究经过北京大学口腔医院生物医学伦理委员会批准(批准号：PKUSSIRB-202499065)，患者均签署知情同意书。
1.1. 资料获取
纳入2020年1月至2022年12月在北京大学口腔医院行双颌手术及术前正畸治疗的骨性Ⅲ类牙颌面畸形患者52例，其中男性26例，女性26例；根据性别将所有患者分为2组，每组随机选定1例为参考样本，其余25例为测试样本。纳入标准：(1)年龄>18岁；(2)骨性Ⅲ类牙颌面畸形；(3)术前锥形束计算机断层扫描(cone beam computed tomography, CBCT)和三维面部图像数据完整；(4)术后效果美观。排除标准：(1)有颌面部手术史或外伤史；(2)存在唇腭裂等先天畸形；(3)三维面部图像质量不佳。
采用CBCT扫描仪(Cefla公司，意大利) 采集颅颌面数据，曝光参数为110 kV，2~3 mA，视野24 cm×19 cm，重建体素精度0.30 mm，图像以DICOM格式储存。采用三维立体摄影扫描仪(3dMD公司, 美国)采集面部三维数据，保留从左耳到右耳、发际线到颏下的三维颜面图像，并进行标准化处理，数据以OBJ格式保存。
1.2. 特征选择
1.2.1. 标志点相关特征将CBCT
数据导入Ivs-plan软件，重建颅骨的三维形态，并以STL格式保存。标定所有模型的鼻根点、眶下点、左右耳点，使用两个参考平面，即法兰克福水平面(Frankfurt horizontal plane, FHP)和正中矢状面(midsagittal plane, MSP)，重新定义图像坐标系，以鼻根点作为坐标原点，定义向右、向下、向前为正轴。将全部颅骨STL图像进行坐标系对齐，根据双侧耳点之间的距离将所有图像缩放为统一大小。在CBCT数据和STL图像上标注47个标志点[9]，根据标志点坐标计算牙颌面畸形相关距离及角度作为硬组织标志点相关特征，作为特征集1，详见表 1。利用深度学习模型[10-11]在三维面部图像上自动标注25个面部标志点，所有标志点均由专家进行确认，以确保其准确性。根据软组织标志点坐标计算牙颌面相关距离及角度作为软组织标志点相关特征，作为特征集2，详见表 2。
表 1.
颅颌面硬组织标志点相关特征
Landmark-based features for craniofacial hard tissue
| Feature | Definition |
| A, subspinale; B, supramental; UI, upper incisor; LI, lower incisor; FHP, Frankfurt horizontal plane; MSP, midsagittal plane. | |
| SNA | Angle sella-nasion-A point |
| SNB | Angle sella-nasion-B point |
| N-A-Pog | Angle nasion-A point-pogonion |
| N-Pog_FHP | Angle formed by FHP and nasion-pogonion line |
| S-Gn_FHP | Angle formed by FHP and sella-gnathion line |
| UI-UIapex_FHP | Angle formed by FHP and UI-UIapex line |
| LI-LIapex_MP | Angle formed by mandible plane and LI-LIapex line |
| MP_FHP | Angle formed by FHP and mandible plane |
| OP_FHP | Angle formed by FHP and occlusion plane |
| CoR-GoR-Me | Angle right superior point of the condyle -right inferior gonion-menton |
| CoL-GoL-Me | Angle left superior point of the condyle -left inferior gonion-menton |
| GoR-Me-GoL | Angle right inferior gonion-menton-left inferior gonion |
| CoR-Me-CoL | Angle right superior point of condyle-menton-left superior point of the condyle |
| Ba-S-N | Angle basion-sella-nasion |
| Ba-N | Distance between basion and nasion |
| Ba-S | Distance between basion and sella |
| S-N | Distance between sella and nasion |
| PoR-PoL | Distance between right porion and left porion |
| ZyR-ZyL | Distance between right zygoma point and left zygoma point |
| JR-JL | Distance between right jugale and left jugale |
| N-Me | Distance between nasion and menton |
| N-ANS | Distance between nasion and anterior nasal spine |
| ANS-Me | Distance between anterior nasal spine and menton |
| S-GoRL | Distance from sella to the line right inferior gonion-left inferior gonion |
| MxR-MxL | Distance between right maxillary basal point and left maxillary basal point |
| UI-APNS | Distance from UI point to the line anterior nasal spine-posterior nasal spine |
| ANS-PNS | Distance between anterior nasal spine and posterior nasal spine |
| UMcuspR-UMcuspL | Distance between right upper first molar point and left upper first molar point |
| UIH | Distance from upper incisor point to FHP |
| U3RH | Distance from right upper canion point to FHP |
| U3LH | Distance from left upper canion point to FHP |
| U6RH | Distance from right upper first molar point to FHP |
| U6LH | Distance from left upper first molar point to FHP |
| LMcuspR-LMcuspL | Distance between right lower first molar point and left lower first molar point |
| CoR-CoL | Distance between right superior point of the condyle and left superior point of the condyle |
| CpR-CpL | Distance between right coracoid process and left coracoid process |
| GoR-GoL | Distance between right gonion and left gonion |
| AgR-AgL | Distance between right antegonion and left antegonion |
| CoR-GoR | Distance between right superior point of the condyle and right gonion |
| CoL-GoL | Distance between left superior point of the condyle and left gonion |
| GoR-Me | Distance between right gonion and menton |
| GoL-Me | Distance between left gonion and menton |
| CoR-Me | Distance between right superior point of the condyle and menton |
| CoL-Me | Distance between left superior point of the condyle and menton |
| UI-MSP | Distance from UI point to MSP |
| LI-MSP | Distance from LI point to MSP |
| Pog-MSP | Distance from pogonion to MSP |
| Overbite | Vertical distance from UI to LI |
| Overjet | Sagittal distance from UI to LI |
表 2.
颅颌面软组织标志点相关特征
Landmark-based features for craniofacial soft tissue
| Feature | Definition |
| Tri-Me’ | Distance between trichion point and soft tissue menton |
| N’-Me’ | Distance between soft tissue nasion and soft tissue menton |
| N’-Me’-v | Vertical distance between soft tissue nasion and soft tissue menton |
| Tri’-N’ | Distance between trichion point and soft tissue nasion |
| Tri-N’-v | Vertical distance between trichion point and soft tissue nasion |
| N’-Sn | Distance between soft tissue nasion and subnasale |
| N’-Sn-v | Vertical distance between soft tissue nasion and subnasale |
| Sn-Me’ | Distance between subnasale and soft tissue menton |
| Sn-Me’-v | Vertical distance between subnasale and soft tissue menton |
| Sn-Stm | Distance between subnasale and stomion |
| Stm-Me’ | Distance between stomion and soft tissue menton |
| TraR-GoR’ | Distance between right tragus and right soft tissue gonion |
| TraL-GoL’ | Distance between left tragus and left soft tissue gonion |
| Tra-Go’(mean) | Average value of distance between tragus and soft tissue gonion |
| GoR’-Me’ | Distance between right soft tissue gonion and soft tissue menton |
| GoL’-Me’ | Distance between left soft tissue gonion and soft tissue menton |
| Go’-Me’ (mean) | Average value of distance between soft tissue gonion and soft tissue menton |
| ExR-ExL | Distance between right exocanthion and left exocanthion |
| TraR-TraL | Distance between right tragus and left tragus |
| GoR’-GoL’ | Distance between right soft tissue gonion and left soft tissue gonion |
| AlR-AlL | Distance between right alar base and left alar base |
| ChR-ChL | Distance between right cheilion and left cheilion |
| ULPR-ULPL | Distance between right upper lip point and left upper lip point |
| ZyR’-ZyL’ | Distance between right soft tissue zygoma point and left soft tissue zygoma point |
| Li-B’-Pog’ | Angle labrale inferius-soft tissue B point-soft tissue pogonion |
| Pn-Pog’-Ls | Angle pronasale-soft tissue pogonion-labrale superius |
| Pn-Pog’-Li | Angle pronasale-soft tissue pogonion-labrale inferius |
| Tri-G | Distance between trichion and glabella |
| G-N’ | Distance between glabella and soft tissue nasion |
| N’-Pn | Distance between soft tissue nasion and pronasale |
| Pn-Sn | Distance between pronasale and subnasale |
| Li-B’ | Distance between labrale inferius and soft tissue B point |
| B’-Pog’ | Distance between soft tissue B point and soft tissue pogonion |
| Pog’-Me’ | Distance between soft tissue pogonion and soft tissue menton |
| Tri-ExR | Distance between trichion and right exocanthion |
| Tri-ExL | Distance between trichion and left exocanthion |
| Tri-TraR | Distance between trichion and right tragus |
| Tri-TraL | Distance between trichion and left tragus |
| N’-AlR | Distance between soft tissue nasion and right alar base |
| N’-AlL | Distance between soft tissue nasion and left alar base |
| N’-GoR’ | Distance between soft tissue nasion and right soft tissue gonion |
| N’-GoL’ | Distance between soft tissue nasion and left soft tissue gonion |
| N’-Go’(mean) | Average value of distance between soft tissue nasion and soft tissue gonion |
| ExR-AlR | Distance between right exocanthion and right alar base |
| ExL-AlL | Distance between left exocanthion and left alar base |
| ExR-ChR | Distance between right exocanthion and right cheilion |
| ExL-ChL | Distance between left exocanthion and left cheilion |
| TraR-AlR | Distance between right tragus and right alar base |
| TraL-AlL | Distance between left tragus and left alar base |
1.2.2. 点云表面特征
为了评估面部形态的连续变化，将所有测试样本与参考样本进行对齐配准，并使用点云来评估面部特征。为了减少原始点云中噪声和冗余的影响，提高计算效率，采用统一的三维点云子采样，保留颅骨的主要特征和结构。为了消除位置和方向的差异，采用迭代最近点(iterative nearest point，ICP)算法实现曲面的精细配准[12]，如图 1所示，首先对原始的三维点云对进行粗配准，然后采用ICP算法进行精细配准，ICP算法是一种迭代优化方法，在每次迭代中通过寻找最近点对，调整点云的旋转和平移，使得两个点云之间的距离逐步最小化，直到算法收敛或达到预设的停止条件。计算每一测试样本相对于参考样本的区域均方根(root mean square，RMS)，以量化两个三维点云的差值。为了研究不同颅骨区域对正颌手术规划的影响，将颅颌面按眶下缘和咬合平面分为上、中、下三个区域，如图 2所示。计算这三个区域以及面中上2/3(即上部+中部)、面中下2/3(即中部+下部) 和整个颅颌面(即上部+中部+下部)最近点距离的RMS，并将其集合设为特征集3。对于软组织，基于软组织表面的标志点构建面部分区[13]。将经过双侧外眦点(ExL和ExR)、口角点(ChL和ChR)的垂直线作为垂直边界线，将通过鼻下点(Sn)、上唇点(Ls)、下唇点(Li)的水平线作为水平边界线，以此将面部分为19个区域，如图 3所示，每个区域中测试样本相对于参考样本的RMS组成特征集4。
图 1.
测试样本(蓝色)与参考样本(黄色)的迭代最近点算法配准
Iterative nearest point registration of test sample (blue) to reference sample (yellow)
A, test sample and reference sample; B, coarse registration; C, fine registration using iterative nearest point.
图 2.
三维颅骨的上、中、下三个区域
Upper, middle and lower regions of 3D skull
图 3.
面部软组织19个分区示意图
Schematic diagram of the 19 regions of facial soft tissue
1.3. 相似度评分
以参考样本的面容及颅骨形态作为标准，采用德尔菲法(Delphi method)评估每一测试样本与参考样本的相似性。3位有经验的正颌外科专家经统一培训并完成一致性检验后参与评分，相似度评分范围为1 (完全不同) ~ 10 (完全相同)，评分值在7.5分及以上设置为临床可接受的相似度评分。
1.3.1. 一致性检验
自数据库所有患者中随机选取3组患者，每组16例，其中1例作为参考样本，其余15例作为测试样本，由3位专家进行评分，评分进行3次，每次间隔一周，采用Kendall’ s W检验评估专家主观评分的稳定性及不同专家之间的相似度评分一致性。
1.3.2. 相似度评分
将52例患者根据性别分为两组，每组26例，随机抽取一例作为参考样本，评估25例测试样本分别相对于参考样本的相似度。在评估过程中，提供病例的三维面部图像和重建的颅骨图像；每轮评分结束后，对专家意见进行分析和总结。每组患者均进行3轮评分，每次评分间隔1周。
1.4. 相似性度量模型的建立与检验
1.4.1. 相似性度量模型建立
构建流程如图 4所示，采用基于相关性的特征选择方法(correlation feature selection, CFS)选择合适的特征子集[14]。CFS假设良好的特征子集包含高度相关的分类特征，而特征之间的相关性最小，并启发式地评估特征子集是否最优，该方法首先采用等宽离散化方法对数值特征进行离散化，然后利用对称不确定性计算离散特征之间的相关性。使用CFS在不同类型的特征集中创建特征子集，选择最优特征进行相似度评价拟合。分别在特征集1和2(基于标志点的特征)、特征集3和4(基于点云表面的特征)以及特征集1~4(基于标志点的特征及基于表面的特征)中选择特征子集。使用线性回归模型建立可理解和可解释的相似性度量模型。模型训练采用交叉验证方法，将数据集分为训练集和测试集后进行反复验证，以减少随机性和偏差。
图 4.
相似性度量模型的构建流程
Workflow of similarity model construction
CFS, correlation feature selection.
1.4.2. 相似性度量模型的有效性验证
随机选取14例新病例并按照第1.2小节所述方式采集及处理术前颅面图像。将处理后的图像输入相似性度量模型，由模型对数据库内的病例进行相似度排序；对于每个新病例，从数据库中选择排名前3位的相似病例。采用第1.3小节的评分方法，由3位专家为相似性度量模型选取的案例进行相似度评分，以3位专家的平均得分作为最终得分。大于7.5分的病例被认为非常相似，如相似性度量模型选取的案例评分大于7.5分，则说明该模型可以准确地筛选类似病例。
1.5. 统计学分析
采用SPSS 19.0进行统计学分析。采用一致性系数(Kendall’ s W test)分别检验同一专家对同一患者在不同时间点的相似度评分稳定性，以及不同专家对同一患者相似度评分的一致性。一致性系数范围从0(完全不一致)到1(完全一致)。采用线性回归方法分析自变量与因变量之间的关系，回归系数(β值)作为方程参数，表示自变量对因变量的影响程度。方差膨胀因子(variance inflation factor，VIF)用于评估自变量之间的多重共线性问题，自变量的VIF值越大，说明与其他自变量之间的相关性越强，可能导致回归模型的不稳定。采用单样本t检验评价相似性度量模型选取案例的相似度评分结果与7.5之间差异是否有统计学意义，P < 0.05为差异有统计学意义.
2. 结果
2.1. 专家评价的一致性检测
为保证相似度评估的一致性，采用3位专家对同一患者的评分进行检验，结果显示，3位专家在3次重复性测试中的评分一致较好(Kendall’ s W = 0.65, 0.71, 0.81, P < 0.05)。此外，3位专家之间的相似度评分也较好(Kendall’ s W =0.52, P < 0.05)，因此，可使用3位专家的平均评分作为三维颅面相似度的基础。
2.2. 模型构建
在特征集上采用CFS寻找最佳相关的特征子集。CFS基于标志点选择的特征子集包含10个特征：前鼻棘点-颏下点(anterior nasal spine-menton，ANS-Me)距离、右上尖牙点至眶耳平面(right canine-FHP，U3RH)距离、左髁顶点-左下颌角点(left superior point of condyle-left gonion, CoL-GoL)距离、颏前点至正中矢状面垂直(pogonion-MSP, Pog-MSP)距离、左髁顶点-颏下点(left condyle-menton, CoL-Me)距离、口裂点-软组织颏下点(stomion-soft tissue menton, Stm-Me’)距离、右鼻翼点-左鼻翼点(right alar base-left alar base, AlR-AlL)距离、鼻尖点-软组织颏前点-下唇点(pronasale-soft tissue pogonion-labrale inferius, Pn-Pog’-Li)交角、发际点-右侧耳屏点(trichion-right tragus, Tri-TraR)距离和左外眦-左鼻翼点(left exocanthion-left alar base, ExL-AlL)距离，命名为特征子集1。基于表面特征选择的特征子集包括7个特征：骨性面下1/3、面中下2/3及全颅颌面，软组织右颊部(图 3的分区5)、左鼻旁(图 3的分区8)、软组织上唇区域(图 3的分区12)和软组织下唇-颏区域(图 3的分区19)，命名为特征子集2。基于标志点特征及表面特征的最佳相关的特征包括：ANS-Me距离、U3RH距离、CoL-GoL距离、CoL-Me距离、Pog-MSP距离、AlR-AlL距离、Pn-Pog’-Li交角、Tri-TraR距离、ExL-AlL距离、骨性面下1/3、骨性面中下2/3及软组织上唇区域(图 3的分区12)，命名为特征子集3。基于所选特征子集进行线性回归，构建相似性度量模型。采用交叉验证方法进行模型验证，最优的相关系数来自特征子集3，达到0.759，而基于标志点的特征与基于表面的特征的相关系数分别为0.695和0.551，因此，利用特征子集3建立的模型可作为最终的相似性度量模型。
2.3. 相似性度量模型关键因素分析
采用线性回归分析评估与相似度评分相关的自变量，发现CoL-Me距离、Pog-MSP距离、Pn-Pog’-Li交角、ExL-AlL距离、软组织上唇区域与相似度评分具有显著相关性(P < 0.05)，详见表 3。
表 3.
三个模型相似度评分相关自变量的线性回归分析
Linear regression analysis of independent variables related to similarity score for 3 models
| Variable | β(95%CI) | P value | Variance inflation factor |
| * P＜0.05. ANS-Me, anterior nasal spine-menton; U3RH, right canine-Frankfurt horizontal plane; CoL-GoL, left superior point of condyle-left gonion; CoL-Me, left condyle-menton; Pog-MSP, pogonion-midsagittal plane; Stm-Me’, stomion-soft tissue menton; AlR-AlL, right alar base-left alar base; Pn-Pog’-Li, pronasale-soft tissue pogonion-labrale inferius; Tri-TraR, trichion-right tragus; ExL-AlL, left exocanthion-left alar base. | |||
| Landmark-based features | |||
| ANS-Me | 0.171 (-0.545 to 0.887) | 0.631 | 2.745 |
| U3RH | -0.350 (-0.942 to 0.241) | 0.238 | 2.000 |
| CoL-GoL | 0.608 (-0.253 to 1.469) | 0.161 | 1.075 |
| CoL-Me | -1.085(-1.984 to -0.187) | 0.019* | 1.962 |
| Pog-MSP | -1.082 (-1.693 to -0.471) | 0.001* | 1.153 |
| Stm-Me’ | -0.766 (-1.421 to -0.111) | 0.023* | 1.793 |
| AlR-AlL | -0.344 (-0.987 to 0.298) | 0.285 | 1.498 |
| Pn-Pog’-Li | -0.656 (-1.260 to -0.052) | 0.034* | 1.232 |
| Tri-TraR | -0.277 (-0.764 to 0.211) | 0.258 | 1.597 |
| ExL-AlL | -0.404 (-1.060 to 0.251) | 0.220 | 1.578 |
| Surface-based features | |||
| Lower 1/3 cranium | 0.129 (-0.769 to 1.027) | 0.773 | 2.065 |
| Lower 2/3 cranium | -1.975 (-3.497 to -0.453) | 0.012* | 4.173 |
| Entire cranium | 0.516 (-0.668 to 1.700) | 0.384 | 2.855 |
| Right cheek region | 0.506 (-0.163 to 1.175) | 0.134 | 1.081 |
| Left paranasal region | 0.543 (-0.166 to 1.251) | 0.130 | 1.109 |
| Upper vermilion region | 1.308 (0.358 to 2.258) | 0.008* | 1.406 |
| Lower facial region | -1.140 (-1.838 to -0.443) | 0.002* | 1.271 |
| All features | |||
| ANS-Me | 0.031 (-0.634 to 0.696) | 0.925 | 2.592 |
| U3RH | -0.226 (-0.807 to 0.355) | 0.435 | 2.112 |
| CoL-GoL | 0.673 (-0.191 to 1.537) | 0.123 | 1.188 |
| CoL-Me | -1.329 (-2.227 to -0.432) | 0.005* | 2.146 |
| Pog-MSP | -0.659 (-1.272 to -0.046) | 0.036* | 1.272 |
| AlR-AlL | -0.456 (-1.089 to 0.177) | 0.153 | 1.592 |
| Pn-Pog’-Li | -0.678 (-1.259 to -0.096) | 0.024* | 1.253 |
| Tri-TraR | -0.082 (-0.545 to 0.381) | 0.722 | 1.579 |
| ExL-AlL | -0.641 (-1.270 to -0.012) | 0.046* | 1.594 |
| Lower 1/3 cranium | -0.188 (-0.847 to 0.471) | 0.568 | 1.929 |
| Lower 2/3 cranium | -0.626 (-1.492 to 0.239) | 0.151 | 2.341 |
| Upper vermilion region | 0.814 (0.122 to 1.505) | 0.022* | 1.293 |
2.4. 模型准确性验证
纳入14例新病例作为模型测试病例。使用相似性度量模型检索相似案例并进行专家评分，相似度评分均值为(7.627±0.711)分。将相似度评分与7.5分进行单样本t检验，差异无统计学意义(t=1.158，P＞0.05)。
3. 讨论
正颌外科手术方案是决定正颌外科手术效果的关键，年轻的外科医师往往需要多年的培训才能获得足够的经验。基于正常人骨骼结构特征建立数据库可用于预测手术方案，但结果缺乏个体化，且未考虑软组织的影响[7]。多篇文献证实，手术后的软组织形态改变既有规律性，也有个体化差异[15]，尤其是口周变化，一直是正颌手术效果预测的难点[16-17]。正常人的软组织形态不受手术中颌骨移位、软组织张力及附着点变化带来的影响[15]，即使牙颌面畸形患者术后的颌骨形态与正常人颌骨形态相似，其软组织表现形态也会与正常人群相异。因此，采用治疗效果满意的同类牙颌面畸形患者作为新患者方案制定的依据，比起采用正常人颅骨作为参考更加合理。为了寻找相似的同类患者数据，本研究建立了一种基于牙颌面畸形患者颅面部骨、软组织特征的相似度评估模型。
第三方或本人对于轮廓美感、协调性、对称度的满意是评价手术效果非常重要的标准，本研究将主客观的评价相融合，相互验证，纳入颅颌面软硬组织角度、距离、表面RMS等特征，分析这些特征与专家相似度评分的相关性，构建相对全面的相似性度量模型。根据线性回归分析结果，共筛选出12个与相似度评分相关的特征，其VIF值分布于1.188~2.592，共线性结果较弱，对于建立模型具有较大优势。在这些特征中，Pog-MSP距离、CoL-Me距离、Pn-Pog’-Li交角、ExL-AlL距离和软组织上红唇区域与相似度评价结果显著相关，其中Pog-MSP距离体现了颏部偏斜程度，CoL-Me距离、ExL-AlL距离分别与颌骨外轮廓、内轮廓的长度比例相关，Pn-Pog’-Li交角、软组织上红唇区域分别与下唇突度、上唇形态相关，这些特征多数与正面部特征或手术涉及的面下区域相关，而较少涉及面中、上部的相似度。这一结果体现了专家对于骨性Ⅲ类牙颌面畸形患者的相似度评价时的选择侧重点，即更加偏重面下1/3及正面观的相似度，反而与传统参照标准SNA、SNB、ANB等数值相关性并不强，这与既往的以侧貌特征作为手术设计依据的观点存在一定差异，反映出专家主观评价与既往角度测量之间的设计思路差异。在进一步研究中，可以扩大专家样本量，或者加入患者主观评价，进一步研究专家及患者关注的颅面部热点问题，为新的手术设计提供依据。
使用建立的相似性度量模型对14例新病例进行相似案例检索，对检索的案例进行相似度评价，平均相似度得分为(7.627±0.711)分，与7.5分差异无统计学意义，表明使用该模型检索的骨性Ⅲ类牙颌面畸形相似案例在临床上具有一定的有效参考性，有助于为手术设计提供参考。
本研究中的相似性度量模型仍有一定局限性，首先，该模型仅针对骨性Ⅲ类患者，其使用的普适性尚有所欠缺，未来将进一步扩大研究，基于其他畸形类型进行模型建立；其次，由于可寻求的相似度检索模型主要使用二维照片，与本研究使用的三维数据差异过大，因此，无法设置对照研究，未来可考虑进一步寻找类似模型或软件进行对比研究；最后，研究主要受试者为中国北部人群，对于该模型在南方、西部等地区的推广是否有效尚待商榷，因此，本研究中的相似性度量模型尚需要通过纳入更全面的数据进行进一步优化。
Funding Statement
国家自然科学基金(82171012)、首都卫生发展科研专项(CFH 2022-2-4104)和北京市自然科学基金(7232222)
Supported by the National Natural Science Foundation of China (82171012), Capital's Funds for Health Improvement and Research (CFH 2022-2-4104) and Beijing Natural Science Foundation (7232222)
Footnotes
利益冲突 所有作者均声明不存在利益冲突。
作者贡献声明 吴灵：设计研究方案，收集、分析、整理数据，撰写论文；方嘉琨：设计研究方案，搭建算法；刘筱菁：提出研究思路，实施手术，总体把关和审定论文；李自力：提出研究思路，实施手术，评估病例相似度，总体把关和审定论文；李阳、王晓霞：评估病例相似度。
References
- 1.Jansma J, Schepers RH. Adjunctive aesthetic procedures in orthognathic surgery. Oral Maxillofac Surg Clin North Am. 2023;35(1):139–152. doi: 10.1016/j.coms.2022.06.007. [DOI] [PubMed] [Google Scholar]
- 2.Jie B, Han B, Yao B, et al. Automatic virtual reconstruction of maxillofacial bone defects assisted by ICP (iterative closest point) algorithm and normal people database. Clin Oral Investig. 2022;26(2):2005–2014. doi: 10.1007/s00784-021-04181-3. [DOI] [PubMed] [Google Scholar]
- 3.Zheng X, Zhao JL, Lv ZH, et al. Skull similarity comparison based on SPCA. Multimed Tools Appl. 2020;79(31/32):22423–22446. [Google Scholar]
- 4.Zhang D, Wu Z, Wang X, et al. 3D skull and face similarity measurements based on a harmonic wave kernel signature. Vis Comput. 2021;37(4):749–764. doi: 10.1007/s00371-020-01946-x. [DOI] [Google Scholar]
- 5.Xiao D, Wang L, Deng H, et al. Estimating reference bony shape model for personalized surgical reconstruction of posttraumatic facial defects. Med Image Comput Comput Assist Interv. 2019;11768:327–335. doi: 10.1007/978-3-030-32254-0_37. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Fang X, Deng HH, Kuang T, et al. Patient-specific reference model estimation for orthognathic surgical planning. Int J Comput Assist Radiol Surg. 2024;19(7):1439–1447. doi: 10.1007/s11548-024-03123-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Wang L, Ren Y, Gao YZ. Estimating patient-specific and anato-mically correct reference model for craniomaxillofacial deformity via sparse representation. Med Phys. 2015;42(10):5809–5816. doi: 10.1118/1.4929974. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Li Z, Teng S, Cheng Y, et al. Craniomaxillofacial deformity correction via sparse representation in coherent space. IEEE Access. 2015;8:69–76. [Google Scholar]
- 9.Cheung LK, Chan YM, Jayaratne YSN, et al. Three-dimensional cephalometric norms of Chinese adults in Hong Kong with balanced facial profile. Oral Surg Oral Med Oral Pathol Oral Radiol Endod. 2011;112(2):e56–e73. doi: 10.1016/j.tripleo.2011.02.045. [DOI] [PubMed] [Google Scholar]
- 10.Fagertun J, Harder S, Rosengren A, et al. 3D facial landmarks: Inter-operator variability of manual annotation. BMC Med Imaging. 2014;14(1):1–9. doi: 10.1186/1471-2342-14-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Paulsen RR, Juhl KA, Haspang TM, et al. Multi-view consensus CNN for 3D facial landmark placement[C]. Cham: Springer, 2019.
- 12.Rusinkiewicz S, Levoy M. Efficient variants of the ICP algorithm[C]. Quebec City: IEEE, 2001.
- 13.Lo LJ, Weng JL, Ho CT, et al. Three-dimensional region-based study on the relationship between soft and hard tissue changes after orthognathic surgery in patients with prognathism. PLoS One. 2018;13(8):e0200589. doi: 10.1371/journal.pone.0200589. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Gopika N, MKME A. Correlation based feature selection algorithm for machine learning[C]. Cairo: IEEE, 2018.
- 15.Rupperti S, Winterhalder P, Krennmair S, et al. Changes in the facial soft tissue profile after maxillary orthognathic surgery. J Orofac Orthop. 2022;83(3):215–220. doi: 10.1007/s00056-021-00294-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Hou L, He Y, Yi B, et al. Evaluation of soft tissue prediction accuracy for orthognathic surgery with skeletal class Ⅲ malocclusion using maxillofacial regional aesthetic units. Clin Oral Investig. 2023;27(1):173–182. doi: 10.1007/s00784-022-04705-5. [DOI] [PubMed] [Google Scholar]
- 17.Olivetti EC, Nicotera S, Marcolin F, et al. 3D soft-tissue prediction methodologies for orthognathic surgery: A literature review. Appl Sci. 2019;9(21):4550. doi: 10.3390/app9214550. [DOI] [Google Scholar]