【作者】 王满宁；

【导师】 宋志坚；

【作者基本信息】 复旦大学 ， 生物医学工程， 2010， 博士

【摘要】 脑肿瘤、脑血管瘤、脑血管畸形等脑部疾病是严重危害人类生命与健康的病种,神经外科手术是治疗这些疾病最直接有效的方法,同时也是难度大、风险高的手术种类。它不仅要求医生将病灶准确切除,还要避免术中对重要的神经、功能区及血管造成损伤。传统的神经外科手术方式很难实现这一目标。统计表明：约有20%的病例术后存在肿瘤残留或者由于损伤了工常神经结构而引起偏瘫、失语等并发症。神经导航系统为弥补传统手术方式的不足,提高手术质量提供了可能。神经导航技术以病人自身的CT、MRI等影像学数据为基础建立图像引导空间,在手术前通过空间配准技术计算图像空间与真实病人所在的病人空间之间的坐标变换关系,而在术中通过空间定位仪实时跟踪手术器械在病人空间中的坐标,并将其变换到图像空间中去。进而在图像空间中生成虚拟手术器械并叠加显示在病人图像上,通过虚拟手术器械相对于病人图像的位置来定位真实手术器械相对于真正病人解剖结构的位置,辅助医生计划手术入路、确定病灶位置和边界以及避免对周围正常组织器官的损伤。神经导航系统不但可以帮助医生降低手术并发症,而且可以减少手术对病人造成的创伤、加快病人康复速度。准确定位是这一切功能的基础,定位精度也是神经导航系统最重要的性能指标。如何提高定位精度一直是神经导航相关研究的重要课题。本文首先对神经导航系统中的各种误差进行了全面的分析与分类,然后提出一系列创新性的技术和方法对影响定位精度的主要因素——空间配准过程进行优化。全文主要研究内容由以下四个部分组成：第一部分,神经导航系统中各种误差的分析与分类。作者根据神经导航系统的工作原理将其中存在的各种误差分成两类：第一类是由于图像空间中反映的病人解剖结构与手术过程中真实病人的解剖结构之间的差异引起的误差；第二类是将手术器械的坐标位置从病人空间转换到图像空间的过程中产生的误差。其中第二类误差主要是由于图像空间和病人空间之间的空间配准问题引起的,本文其它几项研究内容就是通过提出新的技术和方法来降低这一类误差。在这一部分中,作者进一步将每一类误差划分成两个子类,并依次对各个子类中包含的共16种误差进行了分析。第二部分,点匹配空间配准中人工标记点的自动识别。点匹配是神经导航系统临床最常用的空间配准方法,现有导航系统需要医生手动识别人工标记点在图像空间中的位置。这个过程一方面耗时较长,另一方面识别精度受医生的经验影响很大。在这一部分中,作者提出了形状指数聚类和面片投影两种人工标记点自动识别算法,提高了人工标记点识别的精度和稳定性,从而提高了空间白配准的精度和效率。形状指数聚类法首先根据体素的形状指数和曲度特征对三维数据场进行分割,然后对分割结果中对应人工标记物的体素进行聚类,最后根据聚类结果定位人工标记点的位置。作者使用21个临床实际病例的CT数据进行了实验,结果表明：113个人工标记点全部被准确识别,平均识别误差为0.43 mm。面片投影算法通过对比人工标记物的投影图像与三维数据场皮肤表面局部面片的投影图像来识别标记点。实验表明：CT数据场包含的75个标记点中69个被准确识别,MRI数据场包含的52个标记点中47个被准确识别,其它标记点的识别误差为1个体素或者2^1/2个体素。第三部分,点匹配中标记点空间位置分布的优化。神经导航中医生最关心的是手术部位的靶点配准误差（Target Registration Error, TRE）,但是研究表明：点匹配中TRE不仅取决于标记点间的配准程度,还与标记点个数和标记点空间分布密切相关。在本部分中,作者首先提出了一种交互式标记点分布优化方法,通过优化标记点的空间分布来降低手术部位的TRE。作者使用4家医院的46个导航病例进行了实验,结果表明：通过该方法进行优化可以将由于标记点分布不当引起的TRE降低一半左右。在此基础上,作者进一步提出了针对头部不同区域和针对整个头部优化的两套标记点分布模板,临床医生可以根据标记点在模板中的位置直接在病人头上设置标记点,并得到一个优化的标记点分布。第四部分,面匹配空间配准TRE特征的分析。面匹配是神经导航中一种新型空间配准方法,它使用人体固有形态特征进行空间配准,与传统的点匹配方法相比,其优点是不需要人工标记点和专门的影像扫描。但是,目前面匹配在临床中应用并不广泛,究其原因主要是它的配准精度不稳定、人们对其TRE的特征还不够了解。在这一部分中,作者在图像空间中模拟神经导航的面匹配空间配准过程,并对其TRE的特征进行了研究,进而总结出了临床使用中为了保证靶点匹配精度而需要注意的问题,为面匹配的临床应用提供了科学依据。综上所述,空间配准是神经导航的关键技术之一,空间配准的精度在很大程度上决定了导航系统的定位精度。如何实现方便、精确的空间配准一直是神经导航相关研究的重要内容。本文通过对神经导航中各种误差的全面分析以及对空间配准过程的优化,为提高神经导航系统的定位精度提供了一系列科学有效的方法。全文相关内容以第一作者发表和录用SCI论文5篇、EI论文1篇,另向SCI检索的杂志投稿论文2篇。更多还原

【Abstract】 Neurosurgical disease such as brain tumor is a big threat to human health, and surgical operation is the most effective treatment for them. Neurosurgical operation is one of the most difficult and risky operations, since neurosurgeons need not only remove the focus but also protect the normal structures in order to minimize the risk of serious impairment, paralysis, or even death. It is difficult to achieve this goal by the traditional way of operation, and according to statistics in the US, cranial procedures have high complication rates of 10 to 15%.Image Guided Neurosurgery System （IGNS） is an effective instrument to mitigate the risk of complications and improve the quality of neurosurgery. IGNS constructs an image space from the medical images of the patient who will undergo the operation and registers it to the patient space where the real patient exists before operation. During the procedures, IGNS tracks the surgical tools in the patient space and transform the coordinates of them into the image space to generate virtual surgical tools and overlay them on the images. Therefore, the relative position between the images and the virtual tools in the image space is used to help the surgeons localize the position of the real surgical tools relative to the anatomical structures of the real patient. IGNS can help neurosurgeons plan the surgical approaches, localize the position and the extent of the focus and avoid the impairment of normal brain tissue.IGNS can not only help neurosurgeons to minimize the risk of complication but also help to reduce the invasiveness and speed up recovery. Accurate localization is the basis of all these functions, and the accuracy of localization is the most important performance index of IGNS. It has been an important research field to improve the localization accuracy. In this thesis, we first analyze and classify all the factors that may contribute to the inaccuracy of IGNS and then propose several new approaches to optimize the space registration process, which is one of the main sources of the errors. The whole thesis includes the following four parts:Part 1 is the analysis and classification of the errors in IGNS. In this part, we classify all the errors into two groups according to the working principle of IGNS. The first group contains the errors caused by the differences between the anatomical structures in the images and that of the real patient, and the second group contains the errors occurring in transforming the position of surgical tools from the patient space to the image space. The space registration process is a major source of the errors in the second group, and the following parts of this thesis are mainly about how to minimize the errors in this group. In this part, we continue to classify the errors in each group into two subgroups, and discuss 16 errors falling in different subgroups.In part 2, we proposed two approaches to automatically localize the fiducial point from artificial markers. Point-based registration is widely used in IGNS to register the patient space to the image space, and the surgeons need identify the fiducial points in the image space manually in all commercially available systems. The accuracy of the manual identification relies heavily on the experience of the surgeons, and the process of identification cost relatively long time. In this part, we proposed two automatic approaches based on clustering of shape index and projection of surface patch, respectively. In the first approach, we segmented the volume data according to the shape index and curvedness of each voxel and identified the coordinates of the fiducial points by clustering the segmentation results. We tested our methods by 21 CT volume of the patients who underwent image guided neurosurgery, and all 113 fiducial points from artificial markers were identified and the average localization error was 0.43 mm. In the second approach, we identified the fiducial points by comparing the projection image of the artificial marker with that of surface patches extracted from volume data. We tested this method with both CT and MRI volume data. Sixty nine fiducial points were accurately localized from 75 artificial markers in CT data, and 47 fiducial points were accurately localized from 52 artificial markers in MRI data. The localization error of all the other fiducial points was 1 or 2^1/2 voxels.Part 3 is about the optimization of the spatial distribution of fiducial points. What the neurosurgeons care most in the application of IGNS is Target Registration Error （TRE）, and researches indicates that TRE is not only determined by the registration accuracy of the fiducial points but also influenced by the number and the distribution of them. In this part, we first proposed an interactive approach to optimize the spatial distribution of fiducial points to improve the TRE in the region of interest, and experiments showed that this approach could reduce half of the TRE resulting from improper distribution of fiducial points. On the basis of this interactive approach, we continued to introduce two sets of distribution templates that were optimized with regard to different parts of head and the whole head, respectively. Neurosurgeons can arrange fiducial points on the surface of the patient’ s head directly with the guidance of the templates to achieve a distribution optimized with regard to the target region of operation.Part 4 is the analysis of the property of the TRE in surface based registration. Surface based registration is a relatively new approach of space registration in IGNS. Intrinsic features of patients are used in surface based registration, and it need not artificial markers and dedicated image acquisition after attaching the markers. However, surface based registration is not widely used clinically, which is mainly because that its registration accuracy is not stable and the neurosurgeons are not very clear about the properties of its TRE. In this part, we simulated the process of surface based registration in the image space to study the properties of its TRE and gave several practical guidelines for clinical application of it to guarantee the registration accuracy in the region of interest.In conclusion, space registration is one of the key techniques in IGNS, and the accuracy of the space registration to a large extent determines the accuracy of IGNS. How to achieve an accurate and convenient space registration has been an important research area in the field of IGNS. In this paper, we analyzed all the errors in IGNS and proposed several scientific and effective approaches to improve its accuracy. With the research in this thesis as the major contents, four papers were published in and one paper was accepted by SCI indexed journals, and one paper was published in EI indexed journals. Tow other papers were submitted to SCI indexed journals.更多还原