血管介入机器人多模态影像配准机理研究

Vascular interventional robot is a promising trend in the treatment of cardiovascular and cerebrovascular diseases, and multi-modal image registration is of great significance for three-dimensional navigation of vascular interventional robot system. According to the clinical needs of image navigation for vascular interventional robot, "precise, fast and safe", we intends to carry out the research as follows: 1) Unveiling the relation of the deformation and displacement of vessels with reference to respiration and heartbeat, building the model of vessel deformation and motion, to correct the 3D vessel model and operation path map; 2) In order to achieve fast and accurate registration, investigating the CNN registration of 3D vascular preoperative three-dimensional model to the intra-operation X-ray image, and finding the mechanism of CNN to predict the real X-ray image where the CNN is trained with simulated X-ray images; 3) Introducing the ultrasound image to guide the robotic vascular intervention surgery so as to reduce the usage of X-ray, studying the mutual properties and rapid transformation method for the registration and integration of ultrasound image and 3D medical images. This project is a basic research on the application of vascular interventional robots to the problems which are found in the transition to clinical experimental research. Through the research, we will improve the accuracy of image navigation information of vascular interventional robot, and break through the bottleneck limitation of clinical application of vascular interventional robot.

血管介入机器人是心脑血管疾病治疗非常有前景的发展方向，而多模态影像配准对于血管介入机器人系统的三维图像导航具有重要的意义。针对血管介入机器人图像导航“精准、快速和安全”的临床需求，拟开展如下研究：1）揭示血管随呼吸和心跳的变形与位移规律，建立血管变形位移模型，对血管三维模型与手术路图信息进行校正；2）研究血管术前三维模型与术中X光二维影像的CNN配准问题，探索利用仿真X光图像数据训练CNN以预测真实X光图像位姿的机理，实现快速而精确的配准；3）在血管介入机器人手术中引入超声探测技术以减少术中X光的使用，研究血管B超影像与三维影像共有的特征属性和快速变换方法，实现B超与三维数据的融合配准。本项目是针对血管介入机器人在向临床实验研究转化过程中发现的问题而展开的应用基础研究。通过本项目的研究，将提高血管介入机器人图像导航信息的准确性，突破血管介入手术机器人进行临床应用在导航方面的瓶颈限制。

针对血管介入机器人三维图像导航的临床需求，围绕图像配准开展了术前三维血管分割、术中DSA血管分割、超声探头标定、超声图像分割、血管的3D/2D配准等关键技术开展研究，主要研究工作为：.1）针对术前CTA血管分割，改进了mmUnet分割网络，提高了三维建模效果；针对术中DSA分割，提出了MDCNN分割网络，提高了DSA血管分割的准确率。为获得血管变形位移，提出了自训练模型，能较准确获得血管变形的光流。血管分割为进一步研究配准算法提供了数据信息。.2）针对术前3D血管模型与术中2D血管影像的配准，首先针对术中DSA血管不全，提出了基于局部加权归一化互信息算子WLNMI的3D/2D注册方法，相比其他互信息算子有效提高了配准精度；其次针对互信息方法配准迭代优化计算量大耗时长的缺点，以及术中DSA血管直径因心跳而周期性变化现象，提出了基于多通道卷积网络（MCNN）的深度学习配准方法，将配准时间降低到0.4s，进一步提升了配准精度。.3）针对超声技术在血管介入机器人手术的应用，提出了一种基于圆形检测的B超探头标定方法，减小B超噪声大独立点检测不精确的缺陷；对B超影像中组织分割方法开展研究，实现了超声影像中血管的分割；开发了血管介入手术导航软件，引入了IVUS影像，实现了超声信息的引入。.在本项目的支持下，发表期刊论文14篇，会议论文6篇，申请发明专利3项（授权1项），开发血管介入导航软件1套，获得软件著作权1项，为进一步开展相关研究奠定良好基础。