三维复杂颜面畸形对称参考平面智能提取算法研究

Three-dimensional facial symmetry analysis became an important means for clinical diagnosis and evaluation of oral and maxillofacial surgery, orthodontics and prosthodontics. The key of facial symmetry analysis is to determine the symmetry reference plane (SRP). With the continuous development of digital dentistry, the dominant method for determining facial SRP is “Original-Mirror Alignment” method, which is mainly based on the registration algorithm of iterative closest point (ICP) to overlap the original face model and its mirror matching together, and then calculate symmetry plane in mathematics. Because the ICP algorithm is lacking experience and intervention of doctors, the suitability of ICP algorithm for complex facial deformity in oral clinic do not perform well. In early studies of the applicants, Procrustes analysis (PA) algorithm was used to establish a new Original-Mirror Alignment method for extracting SRP, which one could properly introduce doctor’s experience and intervention. But this algorithm still lack of the weights allocation mechanism for clinic use. So, this project intends to carry out a further research focused on PA mathematical model and weight allocation mechanism for determining facial SRP. Firstly, we will establish a PA algorithm with weight allocation mechanism, explore the mathematical expression of the weighting function and its parameter optimization. Then, we will in-depth study and develop an intelligence algorithm for weights allocation strategy based on patients’ personalized facial anatomical feature in three-dimensional. Finally, the clinic research will also carry out to analysis the suitability of the “Weight PA algorithm” for facial deformities’ SRP determining by clinical expert evaluation. The result of this research could provide theoretical reasons of SRP algorithm for oral clinical application.

三维颜面对称性分析是口腔正颌、正畸和修复诊断设计及疗效评价的重要手段，其关键环节是确定三维颜面的“对称参考平面”。随着数字技术的发展，颜面对称参考平面的提取已由传统标志点法发展为基于迭代最近点(ICP)算法的“本体-镜像关联法”，因ICP算法缺少医生临床经验的体现，其对复杂颜面畸形对称参考平面的提取效果欠佳。申请者团队前期基于Procrustes分析(PA)算法建立了一种适当引入医生参与的对称参考平面提取算法，弥补了ICP法的不足，但该算法仍欠缺对复杂畸形情况的权重分配机制。本研究拟对颜面对称参考平面提取算法的赋权PA数学模型和权重分配策略开展研究：建立一种具有权重分配机制的赋权PA算法，探索其权重函数的数学表达及参数优化；深入研究基于个性化颜面三维特征的智能权重分配策略的算法实现；通过临床评价研究，分析赋权PA算法针对复杂颜面畸形对称参考平面提取的适合性，以期为临床应用提供理论依据。

随着口腔数字化诊疗技术的发展和普及，三维颜面部对称性分析成为口腔正颌、正畸和修复等学科数字化诊断设计及疗效评价的基础和前提，三维颜面对称参考平面的确定决定了对称性分析的准确性与可靠性。口腔临床传统诊疗方式多基于主观判定获得颜面对称参考平面，受人为因素影响较大，对临床经验的依赖性较强，尤其对于复杂颜面畸形的情况，主观判定的对称参考平面稳定性、重复性不高，且不同医师间的差异性也较大。口腔临床亟需一种准确、稳定、高效、智能的三维颜面对称参考平面确定方法，满足复杂颜面畸形患者的诊疗需求。.本项目在经典普氏分析（PA）算法研究基础上，提出一种基于三维颜面个性化解剖特征赋权（Weight）的对称参考平面构建算法——赋权普氏分析（WPA）算法。研究针对口腔临床三维颜面不对称畸形特点，创新提出了2种标志点不对称性量化指标，在数学函数层面探索了三维颜面解剖特征的权重分配策略，建立了6种WPA算法函数，并编程实现了三维颜面对称参考平面WPA算法程序。通过项目开展的临床患者三维颜面畸形数据应用评价研究，项目建立的6种WPA算法构建三维颜面对称参考平面与临床专家定义平面的平均角度误差分别为1.21±0.65°、1.18±0.62°、1.39±0.82°、1.39±0.85°、1.39±0.78°和1.34±0.74°，上述6种WPA算法角度误差均小于1.5°，标准差均小于1°，体现了WPA算法对构建三维颜面对称参考平面具有较好的可行性和有效性。.本研究创新建立的6种赋权PA算法已初步实现了“模拟”口腔临床专家诊断策略的三维颜面对称参考平面构建效果，算法程序全部基于开源python软件平台自主研发，成果获得专利授权并实现了专利技术转化，用于国产软件产品的研发。项目研究成果一定程度上实现了国产口腔数字诊断设计软件关键算法技术的突破，有利于数字化诊疗技术的基层普及和推广，具有较好应用前景和社会价值。