CHS: Medium: Collaborative Research: Inverse Anatomical Modeling of the Face for Orthognathic Surgery

脸是个人认同感和自尊的中心，在人际关系中起着至关重要的作用。目前的技术状态是从两个不同的角度对人脸进行计算建模。计算机图形模型具有很高的视觉真实感，就像在电影中看到的那样。而生物力学侧重于物理真实性，将面部建模为遵循物理定律的复杂机械系统。这个项目将弥合这两种观点之间的差距，并构建面部模型，提供视觉和物理现实主义。这在手术预测等应用中至关重要。近5%的美国人有牙面异常，可能需要进行下颌手术，这可能对面部外观产生深远的影响。几乎每个病人都会问：“治疗后我该怎么照顾自己？”尽管这是一个重要且研究充分的问题，但目前还没有能够预测手术后面部表情变化的方法。通过结合视觉和物理现实，这项研究将创建第一个系统，通过在模拟手术过程后显示逼真的面部动画，为患者的问题提供自然的3D视觉答案。项目成果将产生更广泛的影响，因为用于生物材料高效模拟的新数值技术将为各种具有明显非均质性和各向异性的工程材料的计算建模提供可重复使用的基础。在这项工作中开发的解剖建模框架也将作为未来对医学科学（软组织手术建模，面部表情力学中的衰老或病理探索等）感兴趣的探究的启动平台。为了达到这些目的，该项目将创建算法，用于自动开发准确的患者面部解剖模型，这些模型能够代表软组织的真实行为，包括面部表情的形成。这项研究的目标是解决需要跨学科协调努力的挑战，包括计算机图形学、计算机视觉、生物力学和颅面外科。新的计算机视觉方法将利用来自3D成像（MRI/CT）的信息来捕捉活体人脸变形的细节。获得的数据将作为反有限元解算器的输入，该解算器将计算个人特定软组织的未知力学参数，包括预应变和肌肉激活单元。这种以数据为中心的方法与现有的模型构建方法不同，并有可能对解剖建模领域产生革命性的影响。此外，尽管正颌手术的临床应用被用作这项工作的动机和关键基准，但在这项活动中产生的算法创新超越了这项任务的特定范围，并在视觉计算和计算动力学领域提供了更广泛的效用。弹性物体的形状和变形的物理模型将作为结构先验纳入视觉采集系统，增强数据收集的稳健性和准确性。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The face is the center of an individual's sense of identity and self-esteem, and plays a crucial role in interpersonal relationships. The current state of the art approaches computational modeling of human faces from two distinct angles. Computer graphics models feature high visual realism, as seen in the movies. Whereas biomechanics focuses on physical realism, modeling the face as a sophisticated mechanical system that obeys the laws of physics. This project will bridge the gap between these two viewpoints and construct models of the face that offer both visual and physical realism. This is of the utmost importance in applications such as surgical prediction. Close to five percent of the population of the United States has a dentofacial anomaly that may require jaw surgery, which can have a profound effect on the appearance of the face. Virtually every patient asks, "How will I look after the treatment?" Even though this is an important and well-studied problem, there are currently no methods capable of predicting post-operative changes in facial expressions. By combining both visual and physical realism, this research will create the first system that can provide a natural, 3D visual answer to the patient's question by displaying a photorealistic facial animation after a simulated surgical procedure. Additional broad impact will derive from project outcomes because the new numerical techniques for the efficient simulation of biomaterials will provide a reusable foundation that can be leveraged for computational modeling of a variety of engineering materials that exhibit pronounced heterogeneity and anisotropy. The anatomical modeling framework developed in this work will also serve as a launchpad for future inquiry of interest to medical science (modeling of soft-tissue surgery, exploration of aging or pathology in the mechanics of facial expression, etc.).To these ends, the project will create algorithms for the automated development of accurate patient-specific models of facial anatomy capable of representing realistic behavior of soft tissues, including the formation of facial expressions. The research aims at challenges which require coordinated efforts across various disciplines, including computer graphics, computer vision, biomechanics and craniofacial surgery. Novel computer vision methods will leverage information from 3D imaging (MRI/CT) to capture details of in-vivo human face deformations. The acquired data will serve as input to inverse finite element solvers, which will compute the unknown mechanical parameters of person-specific soft tissues, accounting for pre-strain and muscle activation units. This data-centric approach is a departure from established model-building methodologies, and has the potential to make a transformative impact on the anatomical modeling field. Furthermore, although the clinical application of orthognathic surgery is used as the motivation and key benchmark for the work, the algorithmic innovations produced in this activity transcend the specific scope of this task and deliver broader utility in the fields of visual computing and computational dynamics. Physics-based models of shape and deformation of elastic objects will be incorporated into visual acquisition systems as structural priors, enhancing the robustness and accuracy of the data collection.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.