毕业生知识库

随着肿瘤学领域的快速发展和创新，肺癌治疗已从低效的、普适的通用治疗方法转变为精准高效的、对患者具有特异性的专用治疗方法。在肺癌筛查与诊断过程中，使用支气管镜机器人的微创手术存在着逐渐替代传统的开胸手术的趋势。在支气管镜机器人集成系统中，对支气管镜进行实时力学仿真是一项重要的技术。对支气管镜进行快速、准确的实时力学建模与模拟，可以在虚拟支气管镜导航、器械与人体组织间的交互安全评估、支气管镜手术仿真训练、术前/术中的虚拟现实可视化表达等领域中得到应用。然而，目前的支气管镜机器人仿真存在着诸多挑战，如：实时有限元法参数整定、支气管镜插管过程可视化模拟等。因此，本研究课题在中国科学院人工智能创新研究院2035创新任务“医疗机器人集群——面向肺气管疾病的精准无创手术柔性机器人”子课题支持下，围绕支气管镜建模问题的实时有限元法参数整定、插管全过程虚拟模拟、非线性大变形物理问题求解加速等问题展开研究。本文的主要研究和贡献如下：

1.本文提出一种基于实时有限元的柔性体力学建模方法。针对在对柔性体实物使用实时有限元方法进行力学建模时存在的如何确定简化结构、如何确定等效材料参数以及如何确定网格密度三个主要问题，本文以支气管镜为建模对象，提出一种基于实时有限元的柔性体力学建模方法。该方法通过由实物测量实验、结构比较实验、杨氏模量标定实验和网格密度比较实验组成的方法流程，逐一确定力学模型的简化结构、网格密度以及等效材料参数。从而得到在力学特性上贴合实物、在计算速度达到实时性要求的柔性体力学模型。

2.本文提出一种支气管镜插管过程虚拟模拟场景搭建方法。本文使用基于实时有限元的柔性体力学建模方法，在开源仿真框架SOFA下搭建了支气管镜插管过程虚拟模拟场景。该场景由碰撞检测与响应、物理变形、人机交互和实时渲染四个模块组成，用户可以使用Touch X触控设备控制模拟场景中支气管镜模型的插入、回撤与弯曲动作，并通过该设备实时感受到支气管镜模型与环境的接触与碰撞。最后，通过实物与模拟场景的对比实验验证了该场景内支气管镜模型的变形真实性，通过操作者的使用反馈优化了力反馈功能，并通过操作虚拟支气管镜模型动作使得虚拟支气管镜末端成功抵达设定的六个靶支气管。

3.本文提出一种基于实时有限元的预计算有限元建模方法。针对有限元法计算速度慢的问题，本文以支气管镜可控末端为建模对象，提出一种预计算有限元建模方法。首先，本文使用基于实时有限元的柔性体力学建模方法得到支气管镜可控末端的力学模型；接着，使用该力学模型进行大量预计算工作，得到基于实时有限元的大变形预计算数据；然后，使用本文提出的贝塞尔曲线拟合算法对预计算数据中的线型形状处理，仅使用3个三维贝塞尔控制点即可表达整个线型形状；最后，使用神经网络方法对经过处理的预计算数据进行拟合学习，在训练集和测试集上分别取得了5%形状匹配误差判定下95.04%和92.60%的准确率，将非线性大变形问题的解算时间由实时有限元的1500~3000ms压缩到2ms以内。

With the rapid development and innovation in the field of oncology, the treatment of lung cancer has changed from an inefficient and universal treatment method to a precise, efficient and patient-specific treatment method. In the process of lung cancer screening and diagnosis, minimally invasive surgery using bronchoscopic robots is gradually replacing traditional thoracotomy, and real-time mechanical simulation of bronchoscope is an important technology in the integrated system of bronchoscope robot. Fast and accurate real-time mechanical modeling and simulation of bronchoscopes can be used in virtual bronchoscope navigation, interactive safety assessment between instruments and human tissues, bronchoscope surgery simulation training, preoperative/intraoperative virtual reality visualization and other fields be applied in. However, there are many challenges in the bronchoscopic robot simulation currently, such as: real-time finite element method parameter setting, virtual simulation scheme of bronchoscopic intubation process, etc. Therefore, supported by the sub-project of the 2035 innovation mission "Precise Non-invasive Surgery Flexible Robot for Pulmonary and Tracheal Diseases" of the Artificial Intelligence Innovation Institute of the Chinese Academy of Sciences, this research topic focuses on the real-time finite element method parameter tuning of bronchoscope modeling, virtual simulation of the whole process of intubation, nonlinear large deformation physics problem solving acceleration and other issues. The main research and contributions of this paper are as follows:

1. This paper proposes a mechanical modeling method for flexible objects based on real-time finite element. Aiming at the three main problems of how to determine the simplified structure, how to determine the equivalent material parameters and how to determine the grid density in the mechanical modeling of flexible objects using real-time finite element method, this paper takes the bronchoscope as the modeling object and proposes a method for modeling flexible objects mechanics based on real-time finite element method. In this method, the simplified structure, grid density and equivalent material parameters of the mechanical model are determined one by one through the method flow consisting of physical measurement experiment, structure comparison experiment, Young's modulus calibration experiment and grid density comparison experiment. In this way, we obtain a flexible object mechanical model that fits the real object in terms of mechanical properties and meets real-time requirements in calculation speed.

2. This paper proposes a virtual simulation scene construction method for the bronchoscopic intubation process. In this paper, a flexible object mechanics modeling method based on real-time finite elements was used to build a virtual simulation scene of the bronchoscopic intubation process under the open source simulation framework SOFA. The scene is composed of four modules: collision detection and collision response, physical deformation, human-computer interaction and real-time rendering. Feel the contact and collision between the bronchoscope model and the environment in real time. Finally, a comparative experiment between a real object and a simulated scene is performed to verify the authenticity of the deformation of the bronchoscope model in this scene, the effect of force feedback is optimized through the evaluation by some operators, and the virtual bronchoscope end successfully reach the predetermined six target bronchi by the operator manipulating the action of the virtual bronchoscope model.

3. This paper proposes a precomputed finite element modeling method based on real-time finite element. Aiming at the slow calculation speed of the finite element method, this paper proposes a precomputed finite element modeling method with the controllable end portion of the bronchoscope as the modeling object. First, this paper uses the flexible object mechanics modeling method based on real-time finite elements to obtain the mechanical model of the controllable end portion of the bronchoscope. Then, a large amount of precomputation work is performed using the mechanical model to obtain large-deformation precomputed data based on real-time finite element method. Then, use the Bezier curve fitting algorithm proposed in this paper to process the line shape in the precomputed data, therefore only three 3D Bezier control points are used to express the entire line shape. Finally, the neural network method was used to perform fitting learning on the processed precomputed data, and the accuracy rates of 95.04% and 92.60% under 5% shape matching error judgment were obtained respectively on the training set and the test set. The solution time for nonlinear large deformation problems was compressed from 1500 to 3000 ms of real-time finite element analysis to within 2 ms.