沉浸式虚拟显微青光眼手术训练系统关键技术研究

Glaucoma is the number one blinding eye disease. There is a large number of patients in our country, thus leading to a great demand for doctors with related surgical skills. However, the surgical training takes a long time and a lot of resources due to the requirement for high surgical precision of eye surgery. Training system based on computer simulation provides a good alternative, which not only can greatly reduce surgical training cost, but also can improve training efficiency by making reasonable training plan from feedbacks. This project focuses on the simulation of a common glaucoma surgery, trabeculectomy. The procedure is complex and each step requires high accuracy. We focus on the research of designing the interaction model between multiple surgical tools and multilayer eyeball models in trabeculectomy. A real force feedback model will be introduced for guiding doctors in surgical training to improve the perception of eye tissue through surgical tools and surgical skills. Objective evaluation metric will be designed to provide special training suggestions from the results of sub-item evaluation. Besides, the system models the textures for eyeball and surrounding tissues for constructing a realistic training scene in microscopic perspective. By integrating the above technologies, the project designs and manufactures a virtual reality based human-computer interactive operation training platform.

青光眼是头号致盲眼疾，在我国具有庞大的患者基数，对手术医生有非常大的需求。由于眼部手术对精准度有着很高的需求，手术训练需要很长的时间以及大量的资源。基于计算机仿真的训练系统提供了很好的解决方案，不但大大降低了手术训练的代价，还可以及时反馈训练的进度，提供合理的训练计划，提升训练效率。本项目对现阶段最普遍的青光眼手术治疗方法，小梁切除术，进行交互式模拟系统关键技术研究。该手术步骤复杂，并且每一环节都有非常高的精确度要求。项目重点研究设计小梁切除术中多种手术工具与多层次眼球模型的交互碰撞模型，通过构建真实的力反馈模型以引导医生进行模拟手术，提高通过手术工具对眼部组织的感知能力以及手术技能；同时通过制定客观评价标准，通过分步分项评价结果提供专项训练建议；另外通过对眼球及周边组织的纹理进行合理建模，提供显微视角下真实的训练场景。综合以上技术，项目设计并试制基于虚拟现实的人机交互手术训练平台。

随着社会经济水平的不断发展，人们对于优质医疗服务的需求与日俱增，与医疗事故数量居高不下的现状呈鲜明对比。近些年虚拟现实和人机交互技术的飞速发展给医护技能培训带来了新的途径。本文对沉浸式虚拟显微眼科手术训练系统的关键技术进行探索，围绕医学影像分割，虚拟手术仿真和沉浸式虚拟手术训练系统开展了研究工作，为个性化高真实感虚拟手术训练系统提供理论和方法依据。主要研究包括：.1） 提出了一种新的多径密集连接卷积神经网络（MDCNN）来自动分割大小、形状和位置未知的人体目标组织，该网络能够有效地提取出更精确的分割目标位置和轮廓，并提高了稳定性，能够为构建个性化的手术训练场景奠定基础；.2） 提出了基于能量的切削断裂演化模型以实现高度逼真的虚拟剪切仿真，能真实地模拟可变形体的三维切割，并能有效地生成精确的切割面，为临床手术中最为常见的剪切仿真操作提供有效途径；.3）提出一种基于图神经网络的碰撞感知交互物理模拟框架实现工具-软组织的高效高真实感交互仿真。相比传统基于物理的仿真算法，该框架能够大幅提升交互仿真效率，能够为复杂人体解剖下的手术仿真提供实时运算框架；.4） 设计并搭建一套具有高度沉浸感的虚拟显微眼科手术训练系统的软硬件环境，可以实现自然的显微眼科手术人机交互环境，使年轻医师获得高沉浸感的真实手术训练体验，从而达到一定的培训效果。