多模态神经导航辅助下的术中磁共振

在过去的十余年间，术中成像已成为神经外科热点之一。在术中就能客观地评估肿瘤的切除程度是非常有利的。如果术中发现有残留，术者可继续进行肿瘤切除。相对传统方式术者主观判断来说，术中影像能够提供更为客观的术中评估信息，是一种很好的术中质量控制手段［1－6］。 除此之外，借助于计算机辅助外科，能够同时进行神经导航［5，7，8］。导航系统使得术者对手术野术前和术中影像进行必要的联系，从而实时反馈信息。而更重要的是，不仅术中影像能够发现遗漏的残留肿瘤，并使得术者进一步扩大切除范围，还能够借助术中导航来避免损伤脑重要结构，防止增加新的神经功能障碍。

在传统导航系统（又称为无框架立体定向导航）中，手术区域是与三维空间内的解剖影像相关联的。显微镜下导航在实际使用中更具优势。借助于现代显微镜的抬头显示器功能，导航影像能被叠加并投射在显微镜下的手术视野内。传统的解剖结构导航在很多中心已常规使用，近年来，更逐渐演化至整合多种影像信息的多模态神经导航。

进行多模态神经导航的首要步骤就是开发功能神经导航，譬如将脑磁图（MEG）［9－11］或功能磁共振（fMRI）［5，12］显示的脑重要功能区（如语言区和运动区皮层）整合进导航系统，并生成个体化的导航计划。功能神经导航的使用，使得在重要功能区更彻底地切除肿瘤，而不增加术后致残率成为可能。弥散张量成像纤维束示踪技术（DTI）整合进导航系统后，将功能神经导航的范围扩展至皮层下区域［13－16］。 而整合 PET 或磁共振波谱（MRS）则使导航系统能够根据肿瘤的代谢情况进行多模态神经导航［17－22］。

1 将导航整合进外科常规流程

神经外科手术室应该使手术者方便的获取患者的资料和信息。多模态神经导航能够借助于手术区域周边的显示器和手术显微镜内的投射影像显示病人的多模态脑功能影像，是整合过程中的主要部分。

与传统导航系统使用的观察棒相比，显微镜下导航可以直接在手术野中观察各重要结构。因为观察棒只能在导航图像上显示其尖端，因此在术中进行导航时，术者常常需要将视线离开显微镜并观察导航屏幕，中断了正常的手术流程。而显微镜下导航具备的抬头显示功能可以将脑功能结构的信息以不同的颜色投射在手术视野内，与此同时，显微镜焦点或是手术器械尖端可以显示在导航系统屏幕上。

尽管如此，现有的导航系统仍有很大的改进潜力。在大多数情况下，手术视野内显示的是二维图像，对手术视野内深度的立体结构显示不佳，未来将有可能改进为视野内三维图像。而目前常用的导航系统的图像质量尚需提高。另外的一个重要问题是，导航系统应该避免因同时投射过多的影像资料而影响手术者的判断。下一代的高级导航系统需要开发相应功能以使系统在不同的手术阶段智能显示相关的重要脑功能结构，以免误导手术者或中断正常的外科手术流程。

那么，在术中磁共振环境下的导航应该是什么样的概念？一种方式是磁共振机本身就作为一个导航仪，如使用0.5T的GE磁共振机的系统，患者在磁体内接受手术，手术同时进行磁共振扫描。用此方式，手术时可以在磁共振图像上实时观察到手术器械的位置［1，23，24］。 这种导航方式基于实时磁共振成像，也被称为实时立体定向，常被用来进行穿刺置管或是立体定向活检［25－27］。

另一种术中磁共振和导航的整合概念是术中磁共振和导航不是同时进行。大多数系统将导航设备安放在磁体5高斯（5G）线外，以便能使用非磁共振兼容的标准手术器械进行手术。如果仅仅将普通导航系统放置于磁体附近有可能引起相关的安全事故，因此，吊顶式导航系统是一个比较好的选择。整合术中影像和导航对于优化术中导航流程至关重要。术前、术中的影像学数据高效地传输至导航系统必不可少。优化排列术中成像序列的顺序，以便充分利用时间，在进行磁共振扫描的同时进行导航计划的更新。

在进行神经导航辅助肿瘤切除的过程中，导航信息的准确性是非常重要的。在所有可能导致导航误差的因素当中，患者手术开始时的首次注册最易导致误差。目前最常用的导航注册方法是在皮肤表面粘贴导航标记物，并以影像学方法显示标记物，然后通过计算机注册计算将标记物的实际空间位置与导航系统内虚拟位置相对应，完成注册和配准的过程。自动注册法由于无需操作者的操作，因此，可以减少操作者相关的注册误差［28］。

2 自动注册

导航准确性受一系列因素影响，如操作误差、注册系统的移位、术中脑组织的变形（脑组织移位）等。操作误差常见影响因素包括：影像系统成像质量，导航系统本身设备精度和患者注册过程的质量［29］。磁共振成像过程中的影像变形常由成像过程中的化学位移和扫描物体的非匀质性产生［30］。对使用者来说，由此原因产生的导航误差和导航系统技术缺陷产生的误差很难被克服。因此，通过改进使用者的注册操作和积累经验是更为现实的减少导航误差的方法。

有多种方法可以完成标准导航注册。一方面，可以利用导航标记物（粘贴于皮肤表面或是钉入颅骨）或是头部解剖部位进行注册。另一方面，还可利用面部或头部解剖轮廓进行注册。多个研究［31－34］证实了上述注册方法的可靠性，而借助于影像自动或半自动探测标记物技术，也可以比较快速地完成此类导航注册［35］。尽管如此，上述注册过程仍然需要人工进行操作，因此不仅费时，而且容易产生误差。

虽然自动注册的基本原理仍是基于空间位置的点对点注册，但是它避免了使用点对点注册时常规注册方式容易产生误差的环节。自动注册的主要任务是将导航所用影像学数据与导航参考架关联起来。参考架带有反光标记物，并被刚性固定在手术头架上，其作用是给导航系统追踪手术器械提供一个参照物。从数学角度看，参考架和患者影像数据构成了两个独立的坐标系统。一个系统以参考架为基准，定位导航器械与参考架的空间相对位置。另一个系统则将导航器械的三维图像投射在患者的影像图像上，从而在屏幕上显示在患者影像资料上的器械位置，或是将图像投射在手术显微镜内，进行镜下导航。

通常我们认为上述两个坐标系统通过一个刚性配准系统关联在一起，从而使这两个系统通过旋转和移位吻合在一起，或是将一个坐标系内的导航器械位置，转化为另一坐标系内影像资料上的相应位置。为了达成上述目的，必须分别在实际空间内及影像资料上至少定义三个相应标记点。然后，点对应算法将选择最优旋转和移位参数，减少注册误差，并进行最优化注册配准。这种标记点注册法的一大改进是自动注册法的引入。此法需使用自动注册架，该器材由一系列位置相对固定的标记点和导航反光球组成。注册架附加在磁共振上片射频线圈上，并使各标记点尽量靠近患者头部，以便进行磁共振成像时能显示各标记点。扫描完成后，图像资料被传送至导航工作站，操作者仅需简单调整导航红外摄像头，追踪导航参考架和自动注册架的反光球部分，然后导航系统即可计算出参考架和自动注册架之间的空间关系。当自动注册架的反光球空间位置确定后，自动标记物探测算法可以在影像图像上定位出各个标记点。因为系统已内置了关于各个标记点确切位置和排列的空间位置信息，所以上述信息可直接被用来进行标记物配准注册。综合自动注册架和导航参考架的相对位置，自动注册架内各标记物的空间位置，以及各标记物在影像上和患者头部的相对位置，通过计算，就可以定义患者头部在空间的具体位置并开始导航。通常可在患者头皮上放置一个额外标记物，并不将其列入注册标记物，以便可以在注册后，借助该标记物判断注册误差（通常在0.3～2.5 mm之间）。而针对自动注册系统进行的磁共振水膜测试证实该方法误差为0.88～2.13 mm，与传统方法的误差相当。在大多数的测试中，自动注册甚至比使用4或7个标记物的注册方法精度更佳［28］。

3 多模态导航

降低术后并发症的发生率至关重要，尤其是对于高级别胶质瘤来说。因为此类患者，扩大切除范围往往只能延长数周生存时间，而术后长期的并发症将严重影响患者的生存质量。因此，必须同时兼顾最大化的病灶切除和最大化的神经功能保护。术中成像不仅有助于术中最大化的切除病变，还能够通过整合术中多模态神经功能导航来最小化神经功能损伤［13，36］。随着外科技术和围手术期技术的发展，现在已能够对毗邻重要功能区的恶性胶质肿瘤进行最大化切除而并不明显增加术后致残率。有关研究表明，大于98%的病变切除率将明显延长恶性胶质肿瘤患者术后的生存时间，尤其是对Karnofsky评分较好的年轻患者而言［37］。

为达到上述目标，整合脑解剖和功能信息的功能神经导航是术中磁共振的良好补充。借助功能神经导航，可以避免过于激进的切除，并降低术后新发神经功能障碍的发生率。同时，将MEG和fMRI影像整合进功能神经导航系统有助于定位重要脑功能区，如运动区和语言区等［11-12］。

我们曾进行了一项回顾性研究，目的是评估脑磁图对于胶质瘤切除决策的影响［9］。研究包括了5年内所有患有波及功能区的胶质瘤患者，共有191例，其中119例幕上胶质瘤。根据脑磁图检查的结果，约有26.8%的病例在术后有很大可能性发生神经系统严重并发症，因此被认为不适合手术。该研究结果与另一项已发表的研究结果吻合。在该研究中，约30%（12／40例）的肿瘤或动静脉畸形患者由于脑磁图的结果而选择了非手术治疗［38］。当脑功能成像结果融合进无框架导航系统用于手术后，术后致残率低至2.3%，而总体的致残率为6.8%。与其他使用常规手段的研究（致残率6%～31.7%）相比，上述数据说明了功能神经导航能够降低术后的致残率［39-43］。当然，我们也能将该数据解读为，由于术前进行了全面细致的脑功能成像，从而使病例的选择更为仔细和谨慎，使得术后致残率降低。术前的脑功能成像对胶质瘤术前风险评估有帮助，而功能神经导航能降低术后致残率。同时，运动区和语言区皮层的术前定位，也对临床有很大指导意义［44-45］。

脑磁图或功能磁共振只能定位脑皮层功能区。然而，在胶质瘤切除过程中，对脑深部结构，如重要白质纤维束的损伤，同样能够导致神经功能障碍。弥散张量成像技术，不仅能够描绘肿瘤边界，也能被用来显示锥体束等白质纤维束。了解纤维束的走行及与肿瘤的解剖关系，对预防术后神经功能障碍很有帮助［46-47］。将DTI生成的纤维束图像融合进导航系统指导手术可以实现术中对这些重要结构的保护［16，48－49］。但有一个重要前提，就是必须对术中脑解剖结构的变化 （亦称为脑移位）加以考虑。和功能磁共振、脑磁图相比，脑移位对DTI成像的影响更大，因为术中皮层结构的移位，能被直接观察到，而手术创腔深部接近白质纤维束的地方，几乎是不可能在直视下观察到移位的。术中的DTI纤维束成像，显示肿瘤切除过程中，纤维束有显著的移位［15，50］。 因为纤维束的移位，术前的相关功能成像信息将不再准确，所以，只要脑移位没有获得补偿，则术中导航将不再可靠。因此，在术中不仅有必要更新解剖影像，同时也需要更新脑功能影像。术中的DTI纤维束示踪能够显示残余肿瘤与重要白质纤维束的相对关系［15，50］。术中电刺激正中神经来进行被动刺激的功能磁共振则能显示感觉区皮层［51］。

除了脑解剖和功能成像之外，还有一些其他影像数据可以用来进行多模态影像导航。如PET、磁共振波谱（MRS）和磁共振弥散成像能提供肿瘤边界的信息。将代谢影像融合进导航系统能将代谢信息和病理信息进行关联［52－53］。上述信息能否用于手术中，目前正在研究中。而新兴技术，如磁共振分子成像，也许会在未来成为术中成像的手段之一。

4 导航更新

肿瘤切除、脑肿胀、脑压板牵拉以及释放脑脊液，都会引起术中的脑组织变形，又称为“脑移位”［24，54］。 因此，如果单纯依赖术前的影像学数据，导航系统的准确率在术中将下降。术中影像数据提供了术中补偿脑移位的手段。借助于术中成像，能够显示术中的脑移位情况，同时也能显示肿瘤切除程度［24，55－57］。 通过术中重新注册来进行脑移位的补偿，并更新导航计划操作繁琐［56－57］。 在我们使用低场强术中磁共振系统的临床实践中，必须在开颅骨窗周边固定标记物，然后进行术中再注册。然而，低场强术中MRI系统的图像质量较差，从而使得在术中图像上分辨这些标记物变得十分复杂。除此之外，更新导航计划本身也很费时，因为上述标记物首先要被固定在骨窗周边，然后再在术中MR图像上辨认并将其标记出来，随后再使用导航观察棒逐个定位实际标记物，并将标记物的空间位置和影像上的位置一一对应起来 （亦称为术中患者再注册）。正因为上述繁琐的操作，使得在我们进行的330例低场强术中磁共振手术患者中，尽管有26%的病例在进行术中成像时，发现有肿瘤残留（如胶质瘤）且能被继续切除，最终只有 16 例（4.8%）进行了真正的术中导航更新［56，58］。

高场强术中磁共振和显微镜下导航使得术中导航更新成为现实［5］。除了传统的颅骨标记物方法之外，还有另外两种方法可以进行术中患者再注册，一个经过校准的注册参考架可以固定在术中磁共振的上片线圈上，并能被导航系统追踪［28］。在导航注册前，先需要进行一次三维磁共振扫描，扫描范围需包括整个参考架，以便参考架上的所有标记物能显示出来并被用来进行注册。同样的过程也能用于术中再次注册。此时，使用一个消毒的注册参考架，将其与消毒的术中上片磁共振线圈连接，该方法也可作为其他注册手段失败时的后备方法。

如果不需要患者术中再注册，术中的导航更新则更为简单。首先将术中影像与术前影像进行刚性配准，然后描绘残余肿瘤轮廓，最后恢复患者的初始注册数据。这样可以将原来的注册空间坐标用于术中影像上来完成快速地术中影像注册。术中更新导航数据能够可靠地定位残留肿瘤或修正穿刺导管位置。而镜下导航则能够在手术视野内快速准确地定位残留肿瘤，有非常重要的作用。术后病理结果显示，扩大切除的部分均为肿瘤组织，没有假阳性的结果。这当然是由于残留肿瘤被可靠地描绘出来并用来更新导航所致。而将术前和术中影像并列显示也对影像判读大有帮助，并且有效地减少了创腔边缘手术痕迹对影像判读的干扰。

通过将术中和术前的影像并列显示，可以清楚显示残留肿瘤，也可以同时显示脑移位程度。重复的解剖标志确认可以证实通过刚性配准，进行可靠的术中导航更新，而无需再次进行导航注册。刚性配准注册法不太容易受脑移位或是肿瘤切除的影响。这可以通过对比刚性配准后的比较固定的解剖位置来证实，比如对比眼眶和头钉等。一项关于导航自动注册的研究显示，通过术中和术前影像的刚性配准，注册误差在2 mm以内［59］。当然，为避免发生严重注册误差，在进行刚性配准后，必须进行人工确认。

利用术中影像学资料更新导航是目前校正脑移位的最可靠方法。同样的工作流程也可以用来更新多模态的脑功能数据。脑功能成像信息，如fMRI或DTI等可以在术中获取并直接用于术中导航更新。当然，在实际临床操作中，有时上述工作比较费时，如在显像进行连接语言区的纤维束时。

另外，非线性配准及模式识别等更为复杂的技术有可能使包含脑功能信息的术前影像与术中磁共振影像融合配准［60，61］。该技术在无法进行高场强术中磁共振，但能进行其他形式的脑解剖成像时可能会很有帮助。因为此时，可以将术前的多模态脑功能数据与术中获得的较低质量的脑解剖数据进行非线性注册配准，从而预测脑功能结构在术中的位置。一个比较好的替代手段是术中超声，尤其是术中 3D 超声［62－64］。 当然，使用术中超声等其他替代手段能否清晰显示胶质瘤切除范围目前尚存在争议。但是，作为一种真正的术中实时成像方法，术中超声可以提供术中解剖影像，从而能将术前脑功能影像通过非线性配准变形后，来显示术中脑功能结构位置，抵消脑移位的影响。此时，术前的磁共振影像根据非线性配准生成的脑移位数学模型进行相应的形变，从而反应术中脑功能结构的移位［65－69］。

关于使用数学模型来模拟脑移位，目前尚无法准确预测术中的实际情况。术中的多种因素，如脑室系统开放，患者体位变动或脑水肿等，都可能影响脑移位的程度和方向。然而，数学模型加上术中的部分三维影像数据，则有可能调整术前影像来反映术中移位［67，69－71］。术中高场强磁共振，结合术中解剖和脑功能影像是验证和改进这些数学模型的最佳工具。

5 结论

融合了多模态脑功能成像的镜下导航技术和术中高场强磁共振，是目前最有希望做到最大程度切除肿瘤且最大程度保留神经功能的方法之一。多模态神经功能导航融合了脑常规解剖，功能及代谢信息。定位残留肿瘤，分辨肿瘤周边的重要功能结构，以及将影像信息与病理结果关联是多模态功能神经导航的主要目的。利用术中影像资料更新导航后，可以修正脑移位的影响，并准确地定位残留肿瘤。

同英文版）

（中国人民解放军总医院 陈晓雷译）

doi：10.3969 ／j.issn.1002－0152.2012.04.002

Department of Neurosurgery，University Marburg，Marburg，Germany

Intraoperative imaging has attracted increasing interest in the last decade.The ability to objectively determine the extent of tumour removal during surgery is highly advantageous.If the resection is incomplete，one can attempt to remove the tumour residues that were initially missed during the same operation.In contrast to a subjective estimation by the neurosurgeon，intraoperative imaging allows an objective evaluation of the intraoperative situation，thus acting as quality control during surgery［1－6］.In addition to intraoperative imaging，an integral part of our concept of computer aided surgery is the possibility to apply navigation simultaneously［5，7，8］.Navigation allows essentially visualizing the results of pre-and intraoperative imaging in the surgical field，so that the image data provide an immediate feedback.The most important aspect is to prevent increased neurological deficits despite increased resections that might result from the attempt to remove initially overlooked tumor remnants that are detected by intraoperative imaging.

In standard navigation，also known as frameless stereotaxy，the real space of the surgical field is registered to the 3-D image space，which is based on anatomical data only.We prefer the application of microscope-based navigation，where the extent and localization of a tumour is superimposed on the microscope field of view through contours using the headsup display technology of the modern operating microscopes.Standard navigation based on anatomical information only，which has become a routine tool in many neurosurgical departments，was developed further by the integration of further information from other modalities resulting in the so-called multimodal navigation.

A first step in the direction of multimodal navigation was the development of functional navigation，where preoperative data from magnetoencephalography（MEG）［9－11］and functional magnetic resonance imaging （fMRI）［5，12］defining localizations of cortical eloquent brain areas，such as the motor and speech areas，in individual patients，were integrated in the navigation setup.This method of functional neuronavigation allowed more a thorough resection of tumours in risk zones with low morbidity.Integration of diffusion tensor imaging （DTI） data delineating the course of major white matter tracts extended this concept also to subcortical areas［13－16］，while the co-registration of PET data and information from MR spectroscopy （MRS）added metabolic information leading to true multimodal navigation［17－22］.

1 Implementation of navigation in the surgical workflow

The neurosurgical operating room should be the integrative place where all patient data can be visualized in a surgeon-friendly fashion.Multimodal navigation with the visualization of multimodal data on several screens close to the surgical site，as well as，the parallel superimposition of the relevant structures visualized by contours in different colours in the microscope field of view，is the main tool to achieve this integration.

In contrast to pointer-based navigation，microscope-based navigation provides a more intuitive data visualization directly in the surgical field.Pointer-based systems only delineate the position of an instrument，e.g.typically the tip of a pointer，in the image space，so that during surgery when navigation information is needed the surgical workflow is interrupted by necessitating that the surgeon is looking away from the surgical field to a navigation screen.Microscopebased navigation has the advantage of heads-up displays superimposing additional information on the surgical field by colour contours or semi-transparent 3-D objects，while in parallel still the position of an instrument，e.g. the autofocus position of the microscope is still additionally displayed on the navigation screen.

However，there is still much room for further enhancements of these display technologies.In most setups there is only a 2-D visualization，lacking real depth information，which would be possible in a real 3-D image injection setup.Also the realtime rendering of the displayed objects in the current standard commercial systems does not represent what is nowadays possible with near photo-realistic visualizations in other fields of computer graphics.One of the most important aspects in such systems is to avoid a confusion of the neurosurgeon by a potential information overload.Sophisticated systems have to be developed that present the most relevant information at a certain stage of surgery without impeding the surgical workflow and without distracting the neurosurgeon from his main task.

What are the navigation concepts in an intraoperative MR environment？ Either the MR scanner serves as a navigational device per se，as it was the basic principle in the 0.5T double-doughnut GE scanner concept，where the patient was operated directly in the scanner，so that surgical space and imaging space were identical，so that an instrument in the surgical space could be tracked in image space without much additional effort［1，23，24］.Direct navigation in the MR scanner is often based on realtime imaging，like in the so-called prospective stereotaxy，a method for trajectory alignment for placements of catheters or sampling biopsies［25－27］.

Other attempts to integrate intraoperative MRI and navigation result in a classical navigation setup because image space and surgical space are not identical，so that some kind of patient registration has to be applied.Most setups implement navigation at the 5 G line，so that standard non-MR-compatible instruments can be used.Ceiling mounted solutions of the navigation camera and screens［5］are an optimal solution in the intraoperative scenario，since just placing a standard navigation system close to an MR scanner increases the risk of potential magnetic accidents.To optimize the navigation workflow a close integration of imaging and navigation is necessary.For pre-and also intraoperative registration an efficient data transfer between scanner and navigation computers is mandatory，as well as an optimized order of the different imaging sequences allows navigation planning and preparation，while scanning is still going on.

High navigation accuracy is a prerequisite if the navigation information is to be used at critical steps during the resection of a tumor.Among all errors contributing to the overall navigation accuracy，the initial patient registration process is mostly prone to errors.The most common strategy for patient registration relies on placement of skin-adhesive fiducials，which can be detected in the images，so that their position in virtual and real physical space can be correlated to define the registration coordinate system.Automatic registration setups，allowing an user-independent registration of patient space and image space，try to reduce the user dependent errors［28］.

2 Automatic registration

Navigation accuracy is influenced by a variety of factors，among them are the so-called application accuracy，factors relating to a unwanted movement of the registration coordinate system （positional shift），and intraoperative events like brain deformation，which is known as brain shift.The application accuracy is influenced by the quality of imaging，by the technical accuracy of the system itself，and by the quality of patient registration，which defines the process of registering image space and real／surgical space［29］.While the spatial distortion of MR images，which is due to gradient field non-linearities and resonance offsets（chemical shifts and magnetic field inhomogenities）［30］，and the technical accuracy of the navigation system are not easily influenced by the user，the patient registration process is much user-dependent and is influenced by the individual registration strategy and user experience.

Standard patient registration can be achieved in different ways.On the one hand there are “marker-fit”-techniques using either extrinsic markers（so-called fiducials either self-adhesive or implanted in the skull bone） or anatomical landmarks for registration.On the other hand there are “surface-fit”-techniques using the outer contour of the face and the skull for the referencing process.Several studies［31－34］showed the reliability of these registration methods and rapid progress was made towards automatic marker detection in image space and a semiautomatic registration［35］.Nevertheless，the registration process itself still has to be performed manually and therefore remains time-consuming and prone to error.

The concept of an automatic registration is based on paired point registration but sparing the user all the error prone steps associated with it.The main task of the automatic registration is to create an unambiguous relationship between the reference array and the acquired images used for navigation.The reference array is an important reflective marker structure rigidly attached to the patient head via the head clamp.It is used as a relative reference for the system to be able to track instruments，even when the stereoscopic camera is moved into a different position.Looking from the mathematical standpoint the reference array and the volumetric images span two independent coordinate systems.One coordinate system is used to describe the position of tracked instruments in the surgical field or relative to the reference array，the other to render a 3-D model of the tracked instrument in the volumetric images，which in return then is projected on a 2-D view on a screen，or superimposed in the operating microscope，allowing the user to navigate on the images.

It is generally assumed that these two coordinate systems relate to each other via a rigid transformation matrix，describing the rotation and translation between these coordinate systems，or how to transform one position of the instrument，in the surgical space into a corresponding position in the virtual／image space.In order to be able to calculate the rotation and translational parameters，at least 3 points in the patient space and corresponding points in the virtual space have to be defined.A paired point matching algorithm then optimizes the rotation and translational parameters to minimize the root-mean-square-error between these point pairs or even to permute the pairs to achieve an optimal result.An additional indirection for fiducial registration is of importance to implement the automatic registration method.A so-called registration matrix is introduced，which already contains a fixed constellation of fiducials relative to a reflective marker structure，making it possible to use the registration matrix as a tracked instrument.The registration matrix is attached to the upper part of the head coil，so that the fiducials from the registration matrix are automatically imaged，when placed close enough to the patient’s head.Once the scan is completed and the acquired images are transferred to the navigation system，the user only has to adjust the navigation camera so that it can identify the reflective marker structure of the registration matrix and the reference array for a brief moment，so that the navigation system can determine the spatial relation between registration matrix and reference array.Once the acquisition of the reflective marker structures has been completed，an automatic marker detection algorithm detects the fiducials from the registration matrix in the image data set.Since the system knows the exact arrangement and position of the fiducials integrated in the registration ma-trix this information can be used as input for the paired point matching algorithm.Combining the information where the registration matrix was in relation to the reference array and how the detected fiducials in the images relate to the fiducials of the registration matrix and also knowing how the fiducials from the registration matrix relate to the spatial position of the matrix itself by the defined construction，a transforma-tion matrix can be calculated directly relating the refer-ence array with the acquired images，so that then the relation between image space and physical／surgical space is defined and navigation can be used.An additional skin fiducial that is not used for the registration process is localized after patient registration to docu-ment a target registration error，which is typically in the range between 0.3 and 2.5 mm.Phantom studies resulted in median localization errors between 0.88 and 2.13 mm for the automatic registration approach，which was at least not worse，in most test series even signifi-cantly better，than that of the standard registration no matter whether 4 or 7 fiducial markers were used［28］.

3 Multimodal navigation

Low postoperative neurological deficits are mandatory，especially in surgery of high grade tumors it is of no benefit for the patient to maximize the extent of a resection to potentially increase the survival time by only some weeks，when risking permanent neurological deficits right after surgery.It is absolutely mandatory to combine the goal of maximum resection with the goal of preservation of function.Intraoperative imaging helps not only to maximize the extent of resection but in combination with functional multimodal navigation also minimization of postoperative neurological deficits is possible［13，36］.With the advances in surgical techniques and perioperative technology，it is now possible to maximally resect malignant intrinsic glial neoplasms，even close to functionally critical areas，without increased morbidity. Studies have demonstrated a survival advantage of these lesions with a resection extent of 98%or greater，particularly in younger patients with good Karnofsky scores［37］.

To achieve this，functional navigation，i.e.integrating functional data into anatomical navigational datasets，is an important add-on to intraoperative MRI since it prevents too extensive resections，which would otherwise result in new neurological deficits.Meanwhile，data from MEG and fMRI are routinely integrated in functional navigation allowing identification of eloquent brain areas such as the motor area and speech related areas［11，12］.

In a retrospective study we focused on how the decision to resect a glioma was influenced by MEG［9］.In a time period of 5 consecutive years we have investigated every patient proposed for surgery，who harboured a lesion adjacent to an eloquent brain area.Altogether 191 patients were examined，119 of them harboured supratentorial gliomas.About every forth patient （26.8%） yielded a severe possible danger of postoperative neurological morbidity according to MEG and thus was not considered being a good candidate for surgery.This is a corresponding result to published data where 12 out of 40 investigated patients （30%）with tumours and vascular malformations underwent non-surgical therapy according to the MEG results［38］.When functional data were used in combination with frameless stereotactic devices the postoperative morbiditywasaslowas2.3%.Overallmorbidityhoweverwas 6.8%.These data reflect the beneficial effects of functional navigation in comparison to data of other studies with morbidity rates varying from 6 to 31.7%［39－43］.These figures can also be interpreted as a result of a more careful patient selection through the help of preoperative brain mapping.Preoperative identification of eloquent brain areas has an impact in the risk evaluation in glioma surgery，as well as functional navigation reduces the risk for postoperative neurological deficits.Besides identification of the motor strip，the localisation of language areas is of great clinical impact［44，45］.

Functional data from MEG and fMRI only localize function at the brain surface.However，neurologi-cal deficits can also occur during tumor resection due to damaging of deeper structures，such as major white matter tracts.DTI can be used not only to delineate tumor borders，but also to display the course of white matter tracts，such as the pyramidal tract.The knowledge of the course of major white matter tracts in relationship to a tumor helps to prevent new postoperative neurological deficits［46，47］.Registration of diffusion data with the navigational dataset［16，48，49］facilitates the intraoperative preservation of these eloquent structures.A prerequisite is that intraoperative changes of the brain anatomy，known as brain shift，are taken into account.In contrast to the use of fMRI and MEG，brain shift clinically effects much more the DTI data，because the intraoperative shifting of cortical areas during surgery can be well detected by the naked eye，however changes in the depth of a resection cavity，close to major white matter tracts，is nearly undetectable for the neurosurgeon during tumor resection.Intraoperative DTI，revealed a marked shifting of the pyramidal tract due to tumor resection［15，50］.As a consequence of this shifting the preoperative functional data are no longer valid，so navigation can no longer relied on，if this shifting is not compensated for.Therefore，it is necessary that not only intraoperative anatomical data are used to compensate for the effects of brain shift but also functional data have to be updated.Intraoperative acquisition of DTI data enables intraoperative fiber tracking to visualize how a tumor remnant is localized in relation to major white matter tracts［15，50］.Even intraoperative fMRI applying electrical stimulation of median and tibial nerves as a passive stimulation paradigm is possible and enables identification of the somatosensory cortex［51］.

Besides functional and structural data further information is available for a multimodal navigation setup.PET，MRS，and diffusion weighted imaging may provide information on the diffuse tumor border.Integration of metabolic maps into the navigation datasets enables a spatial correlation of metabolic data and histopathological findings［52，53］.Whether these techniques can also be used intraoperatively，so that these data can also be updated，is under investigation.Furthermore，upcoming techniques such as MR-based molecular imaging may find its role in the intraoperative imaging armamentarium.

4 Navigation updating

Tumor removal，brain swelling，the use of brain retractors，and cerebrospinal-fluid drainage all result in an intraoperative brain deformation，which is known as brain shift［24，54］.Thus，navigation systems relying on preoperative image data only have a decreasing accuracy during the surgical procedure.Intraoperative imaging offers a possibility to compensate for the effects of brain shift，because it provides a virtual reproduction of the actual intraoperative physical reality，on how the brain is deformed and on the achieved extent of tumor removal［24，55－57］.Compensation for brain shift by intraoperative registration of the intraoperative MRI data to update the navigation setup has been a cumbersome process［56，57］.In our low-field MRI setting bone fiducial markers had to be placed around the craniotomy opening，which were then used for intraoperative patient re-registration.With the restricted image quality of the low-field MRI system it was quite complicated to identify these markers in the intraoperative images.Furthermore，updating was a time consuming process，because these bone fiducials had to be placed around the craniotomy opening； then they had to be identified and marked in the intraoperative images and then they had to be individually identified in physical space using the tip of a pointer and correlated to the position in the intraoperative images to define the navigation coordinate system，i.e.intraoperative patient re-registration.This is the main reason why only in 16 out of 330 patients investigated with low-field MRI an actual navigation update was performed，despite intraoperative imaging had detected tumor remnants e.g.in the gliomas that could undergo further resection in about 26%［56，58］.

The setup integrating high-field MRI and micro-scope-based navigation offered to facilitate this intraoperative update procedure［5］.Besides the bone-fiducial placement for intraoperative re-registration two alternative patient registration methods and thus update strategies are available.A calibrated registration matrix can be attached to the upper part of the head coil and tracked by the navigation system.28 This automatic registration matrix can be used for an initial patient registration.Before the patient registration it is required that the anesthetized patient gets a preoperative 3-D MR scan after head fixation，which fully covers the registration matrix，so that all markers integrated in the registration matrix are visible and can be detected and used for registration.This approach is also a possibility to perform a registration of intraoperatively acquired images.A sterile registration matrix has to be connected to the sterile upper part of the head coil before it is attached to the head holder.This method serves as a backup approach if other registration strategies fail.

Navigation updating without an intraoperative patient re-registration is an even more straightforward approach.It is based on a rigid registration of the intraoperative image data with the preoperative image data，subsequent segmentation of the tumor remnant，and final restoring of the initial patient registration，so that the registration coordinate system of the preoperative image data is applied on the intraoperative images，allowing an immediate intraoperative image update.Updated image data allow a reliable identification of a tumor remnant or correction of a catheter.Microscope-based image injection with the direct visualization of the segmented tumor remnant in the surgical field plays a crucial role in the precise localization and orientation in the resection cavity.Histological analysis of the extended resections proved pathological tissue in all cases； there were no false positive findings.This is of course due to the fact that only areas of reliably identified tumor remnants were segmented and used for updating.The side-by-side analysis of preand intraoperative images greatly facilitates image interpretation and to exclude misinterpretations due to surgically induced changes at the resection border.

After updating also a side-by-side display of preand intraoperative images visualizing the segmented tumor remnant in both of them，is possible，facilitat ing orientation and image interpretation-therefore，also the extent of brain shift is easily visible.Repeated landmark checks have proved that the rigid registration approach is a reliable approach to update the navigation system without a repeated patient registration.The rigid registration algorithm is robust enough to accomplish a registration that is not sensitive to the effects of brain shift and the actual reduction of the tumor mass.This could be shown by analyzing the registration accuracy of structures that are not affected by brain shift，such as the position of both orbits and the position of the skull fixation pins.An extensive analysis comparing the automatic registration，which is used to register pre-and intraoperative images，with an independent reference registration showed，that the registration error is below 2 mm even in a worst case scenario［59］.Nevertheless，to prevent a mis-registration a visual control after rigid registration of pre-and intraoperative images is mandatory.

Updating the navigation system with intraoperative MR image data seems to be the most reliable method to compensate for the effects of brain shift.In contrast to previous setups this framework is also open to update multimodal information.Functional data，such as fMRI or DTI data can also be acquired intraoperatively and directly used for intraoperative updating，which in the clinical routine might be a time-consuming effort，especially when in case of e.g. visualization of speech connecting fiber tracts some sophisticated time-consuming non-standard tracking algorithms have to be applied.

Alternatively either non-linear registration techniques，as well as sophisticated techniques from pattern recognition analysis may allow a matching of preoperative MR data sets containing functional information with intraoperative MR image volumes［60，61］.This might also be a possibility in case where intraoperative MRI is not available，but other imaging modalities provide intraoperative 3-D information about the brain configuration，so that high-resolution multi-modality data can be registered non-linearly onto the ‘low-quality’ intraoperative data.An alternative to intraoperative MR imaging might be intraoperative ultrasound，especially intraoperative 3-D ultrasound［62－64］.Whether the image quality to evaluate the extent of a glioma resection is equivalent among the different imaging modalities is still discussed controversially.However，it is out of doubt that intraoperative ultrasound has the advantage of being a real-time modality.Ultrasound data may provide information on how to deform high-quality preoperative MR image data in order to represent the intraoperative real situation，thus compensating for the effects of brain shift.This approach relies on the non-linear registration of intraoperative ultrasound data with preoperative MR volume data.Consecutively the MR data are deformed using a mathematical model describing the deforma tion of the brain during surgery［65－69］.Mathematical models trying to simulate brain shift behavior will not be able to predict the actual intraoperative situation without further data.Intraoperative events such as opening of the ventricular system and CSF drainage，patient positioning，and brain swelling all influence the extent and direction of brain shift. However，mathematical models together with some sparse data describing the actual intraoperative 3-D situation may be able to adjust high-quality preoperative data to represent the intraoperative reality［67，69－71］.Intraoperative high-field MRI with anatomical and functional imaging possibilities is the ideal tool to validate and refine these models.

5 Conclusion

Microscope-based navigation serving as a common interface for the presentation of multimodal data in the surgical field in combination and close integration with intraoperative high-field MRI seems to be one of the most promising setups allowing avoiding unwanted tumor remnants while preserving neurological function.Multimodal navigation integrates standard anatomical，structural，functional，and metabolic data.Visualizing the initial extent of a lesion，identification of neighboring eloquent brain structures，as well as allowing a direct correlation of histology and multimodal data are the main tasks of navigation.After intraoperative imaging navigation data can be updated，so that brain shift is compensated for and initially missed tumor remnants can be localized reliably.

［1］Black PM，Moriarty T，3rd AE，et al.Development and implementation of intraoperative magnetic resonance imaging andits neurosurgical applications［J］.Neurosurgery，1997，41（4）：831－842；discussion：842－845.

［2］Hall WA，Kowalik K，Liu H，et al.Costs and benefits of intraoperative MR-guided brain tumor resection ［J］.Acta Neurochir Suppl，2003，85：137－142.

［3］Hall WA，Liu H，Martin AJ，et al.Safety，efficacy，and functionality of high-field strength interventionalmagnetic resonance imaging for neurosurgery［J］.Neurosurgery，2000，46（3）：632－641； discussion：641－642.

［4］Nimsky C，Ganslandt O，Fahlbusch R.Comparing 0.2 tesla with 1.5 tesla intraoperative magnetic resonance imaginganalysis of setup，workflow，and efficiency［J］.Acad Radiol，2005，12（9）：1065－1079.

［5］Nimsky C，Ganslandt O，Von Keller B，et al.Intraoperative high-field-strength MR imaging：implementation and experience in200 patients［J］.Radiology，2004，233（1）：67－78.

［6］Sutherland GR，Kaibara T，Louw D，et al.A mobile high-field magnetic resonance system for neurosurgery ［J］. J Neurosurg，1999，91（5）：804－813.

［7］Nimsky C，Ganslandt O，Kober H，et al.Intraoperative magnetic resonance imaging combined with neuronavigation：a newconcept［J］.Neurosurgery，2001，48 （5）：1082－1089； discussion 1089－1091.

［8］Steinmeier R，Fahlbusch R，Ganslandt O，et al.Intraoperative magnetic resonance imaging with the magnetom open scanner：concepts，neurosurgical indications，and procedures： a preliminary report［J］.Neurosurgery，1998，43（4）：739－747； discussion 747－748.

［9］Ganslandt O，Buchfelder M，Hastreiter P，et al.Magnetic source imaging supports clinical decision making in glioma patients［J］.Clin Neurol Neurosurg，2004，107（1）：20－26.

［10］Ganslandt O，Steinmeier R，Kober H，et al.Magnetic source imaging combined with image-guided frameless stereotaxy：a new-method in surgery around the motor strip［J］.Neurosurgery，1997，41（3）：621－627； discussion 627－628.

［11］Ganslandt O，Fahlbusch R，Nimsky C，et al.Functional neuronavigation with magnetoencephalography：outcome in 50 patientswith lesions around the motor cortex ［J］. J Neurosurg，1999，91（1）：73－79.

［12］Nimsky C，Ganslandt O，Kober H，et al.Integration of functionalmagnetic resonance imaging supported bymagnetoencephalography in functional neuronavigation ［J］.Neurosurgery，1999，44（6）：1249－1255； discussion 1255－1256.

［13］Nimsky C，Ganslandt O，Fahlbusch R.1.5 T： intraoperative imaging beyond standard anatomic imaging［J］.Neurosurg Clin N Am，2005，16（1）：185－200，vii.

［14］Nimsky C，Ganslandt O，Fahlbusch R.Implementation of fiber tract navigation［J］.Neurosurgery，2006，58（4 Suppl 2）：ONS－292－303； discussion ONS－303－304.

［15］Nimsky C，Ganslandt O，Hastreiter P，et al.Preoperative and intraoperative diffusion tensor imaging-based fiber tracking in glioma surgery［J］.Neurosurgery，2005，56（1）：130－137； discussion 138.

［16］Nimsky C，Ganslandt O，Merhof D，et al.Intraoperative visualization of the pyramidal tract bydiffusion-tensor-imaging-based fiber tracking［J］.Neuroimage，2006，30（4）：1219－1229.

［17］Ganslandt O，Stadlbauer A，Fahlbusch R，et al.Proton magnetic resonance spectroscopic imaging integrated into image-guidedsurgery：correlation to standard magnetic resonance imaging and tumor celldensity［J］.Neurosurgery，2005，56（2 Suppl）：291－298； discussion 291－298.

［18］Stadlbauer A，Ganslandt O，Buslei R，et al. Gliomas：histopathologic evaluation of changes in directionality and magnitude ofwater diffusion at diffusion-tensor MR imaging［J］.Radiology，2006，240（3）：803－810.

［19］Stadlbauer A，Moser E，Gruber S，et al.Integration of biochemical images of a tumor into frameless stereotaxy achievedusing a magnetic resonance imaging／magnetic resonance spectroscopy hybrid dataset［J］.J Neurosurg，2004，101（2）：287－294.

［20］Stadlbauer A，Nimsky C，Buslei R，et al.Proton magnetic resonance spectroscopic imaging in the border zone of gliomas：correlation of metabolic and histological changes at low tumorinfiltration--initial results［J］.Invest Radiol，2007，42（4）：218－223.

［21］Stadlbauer A，Nimsky C，Buslei R，et al.Diffusion tensor imaging and optimized fibertracking in glioma patients：Histopathologic evaluation of tumor-invaded white matter structures［J］.Neuroimage，2007，34（3）：949－956.

［22］Stadlbauer A，Prante O，Nimsky C，et al.Metabolic imaging of cerebral gliomas： spatial correlation of changes inO-（2-18F-fluoroethyl）-L-tyrosine PET and proton magnetic resonance spectroscopic imaging［J］.J Nucl Med，2008，49（5）：721－729.

［23］Black PM，3rd AE，Martin C，et al.Craniotomy for tumor treatment in an intraoperative magnetic resonance imagingunit ［J］.Neurosurgery，1999，45（3）：423－431； discussion 431－433.

［24］Nabavi A，Black PM，Gering DT，et al.Serial intraoperative magnetic resonance imaging of brain shift ［J］. Neurosurgery，2001，48（4）：787－797； discussion 797－798.

［25］Truwit CL，Liu H.Prospective stereotaxy：a novel method of trajectory alignment using real-timeimage guidance ［J］.J Magn Reson Imaging，2001，13（3）：452－457.

［26］Hall WA，Liu H，Truwit CL.Navigus trajectory guide［J］.Neurosurgery，2000，46（2）：502－504.

［27］Truwit CL，Hall WA.Intraoperative magnetic resonance imagingguided neurosurgery at 3-T［J］.Neurosurgery，2006，58（4 Suppl 2）：ONS-338－445； discussion ONS-345－346.

［28］Rachinger J，von KB，Ganslandt O，et al.Application accuracy of automatic registration in frameless stereotaxy ［J］. Stereotact Funct Neurosurg，2006，84（2－3）：109－117.

［29］Steinmeier R，Rachinger J，Kaus M，et al.Factors influencing the application accuracy of neuronavigation systems［J］.Stereotact Funct Neurosurg，2000，75（4）：188－202.

［30］Sumanaweera TS，Adler JR Jr，Napel S，et al.Characterization of spatial distortion in magnetic resonance imaging and itsimplications for stereotactic surgery［J］.Neurosurgery，1994，35（4）：696－703； discussion 703－704.

［31］Raabe A，Krishnan R，Wolff R，et al.Laser surface scanning for patient registration in intracranial image-guidedsurgery ［J］.Neurosurgery，2002，50（4）：797－801； discussion 802－803.

［32］Villalobos H，Germano IM.Clinical evaluation of multimodality registration in frameless stereotaxy ［J］. Comput Aided Surg，1999，4（1）：45－49.

［33］Wolfsberger S，Rossler K，Regatschnig R，et al.Anatomical landmarks for image registration in frameless stereotacticneuronavigation［J］.Neurosurg Rev，2002，25（1－2）：68－72.

［34］Barnett GH，Miller DW，Weisenberger J.Frameless stereotaxy with scalp-applied fiducial markers for brain biopsyprocedures：experience in 218 cases［J］.J Neurosurg，1999，91 （4）：569－576.

［35］Kozak J，Nesper M，Fischer M，et al.Semiautomated registration using new markers for assessing the accuracy of anavigation system［J］.Comput Aided Surg，2002，7（1）：11－24.

［36］Nimsky C，Ganslandt O，Fahlbusch R.Functional neuronavigation and intraoperative MRI ［J］. Adv Tech Stand Neurosurg，2004，29：229－263.

［37］Hentschel SJ，Sawaya R.Optimizing outcomes with maximal surgical resection of malignant gliomas［J］.Cancer Control，2003，10（2）：109－114.

［38］Hund M，Rezai AR，Kronberg E，et al.Magnetoencephalographic mapping：basic of a new functional risk profile in theselection of patients with cortical brain lesions［J］.Neurosurgery，1997，40（5）：936－42； discussion 942－943.

［39］Ammirati M，Galicich JH，Arbit E，et al.Reoperation in the treatment of recurrent intracranial malignant gliomas［J］.Neurosurgery，1987，21（5）：607－614.

［40］Black PM.Surgery for cerebral gliomas： past，present，and future［J］.Clin Neurosurg，2000，47：21－45.

［41］Cabantog AM，Bernstein M.Complications of first craniotomy for intra-axial brain tumour ［J］.Can J Neurol Sci，1994，21（3）：213－218.

［42］Ciric I，Ammirati M，Vick N，et al.Supratentorial gliomas：surgical considerations and immediate postoperativeresults.Gross total resection versus partial resection［J］.Neurosurgery，1987，21（1）：21－26.

［43］Fadul C，Wood J，Thaler H，et al.Morbidity and mortality of craniotomy for excision of supratentorial gliomas［J］.Neurology，1988，38（9）：1374－1379.

［44］Grummich P，Nimsky C，Pauli E，et al.Combining fMRI and MEG increases the reliability of presurgical languagelocalization：a clinical study on the difference between and congruence of bothmodalities［J］.Neuroimage，2006，32（4）：1793－1803.

［45］Kober H，Moller M，Nimsky C，et al.New approach to localize speech relevant brain areas and hemispheric dominanceusing spatially filtered magnetoencephalography［J］.Hum Brain Mapp，2001，14（4）：236－250.

［46］Hendler T，Pianka P，Sigal M，et al.Delineating gray and white matter involvement in brain lesions：three-dimensionalalignment of functional magnetic resonance and diffusion-tensor imaging［J］.J Neurosurg，2003，99（6）：1018－1027.

［47］Clark CA，Barrick TR，Murphy MM，et al.White matter fiber tracking in patients with space-occupying lesions of thebrain：a new technique for neurosurgical planning？ ［J］. Neuroimage，2003，20（3）：1601－1608.

［48］Coenen VA，Krings T，Mayfrank L，et al.Three-dimensional visualization of the pyramidal tract in a neuronavigationsystem during brain tumor surgery：first experiences and technical note［J］.Neurosurgery，2001，49（1）：86－92； discussion 92－93.

［49］Nimsky C，Grummich P，Sorensen AG，et al.Visualization of the pyramidal tract in glioma surgery by integrating diffusiontensor imaging in functional neuronavigation［J］.Zentralbl Neurochir，2005，66（3）：133－141.

［50］Nimsky C，Ganslandt O，Hastreiter P，et al.Intraoperative diffusion-tensor MR imaging：shifting of white matter tractsduring neurosurgical procedures--initial experience ［J］. Radiology，2005，234（1）：218－25.

［51］Gasser T，Ganslandt O，Sandalcioglu E，et al.Intraoperative functional MRI： implementation and preliminary experience［J］.Neuroimage，2005，26（3）：685－93.

［52］Stadlbauer A，Moser E，Gruber S，et al.Integration of biochemical images of a tumor into frameless stereotaxy achievedusing a magnetic resonance imaging／magnetic resonance spectroscopy hybrid dataset［J］.J Neurosurg，2004，101（2）：287－94.

［53］Stadlbauer A，Moser E，Gruber S，et al.Improved delineation of brain tumors：an automated method for segmentation based on pathologic changes of 1H-MRSI metabolites in gliomas［J］.Neuroimage，2004，23（2）：454－461.

［54］Hastreiter P，Rezk-Salama C，Soza G，et al.Strategies for brain shift evaluation［J］.Med Image Anal，2004，8（4）：447－464.

［55］Hastreiter P，Rezk-Salama C，Soza G，et al.Strategies for brain shift evaluation［J］.Med Image Anal，2004，8（4）：447－464.

［56］Nimsky C，Ganslandt O，Hastreiter P，et al.Intraoperative compensation for brain shift［J］.Surg Neurol，2001，56 （6）：357－364； discussion 364－365.

［57］Wirtz CR，Bonsanto MM，Knauth M，et al.Intraoperative magnetic resonance imaging to update interactive navigation inneurosurgery： method and preliminary experience［J］.Comput Aided Surg，1997，2（3－4）：172－179.

［58］Nimsky C，Ganslandt O，Tomandl B，et al.Low-field magnetic resonance imaging for intraoperative use in neurosurgery：a5-year experience［J］.Eur Radiol，2002，12（11）：2690－2703.

［59］Veyrat A.Automatic fusion of pre-and intraoperative patienta data. A statistical evaluation of accuracy ［J］. diploma thesis Technical University Munich，2005.

［60］Archip N，Clatz O，Whalen S，et al.Non-rigid alignment of preoperative MRI，fMRI，and DT-MRI with intra-operativeMRI for enhanced visualization and navigation in image-guided neurosurgery［J］.Neuroimage，2007，35（2）：609－624.

［61］Wolf MVT，Weierich PNH，Nimsky C.Automatic transfer of preoperative fMRI markers into intraoperative MR-images for updating functional neuronavigation［J］.IEICE T Inf Syst，2001，E84-D（12）：1698－1704.

［62］Comeau RM，Sadikot AF，Fenster A，et al.Intraoperative ultrasound for guidance and tissue shift correction inimage-guided neurosurgery［J］.Med Phys，2000，27（4）：787－800.

［63］Tirakotai W，Miller D，Heinze S，et al.A novel platform for image-guided ultrasound［J］.Neurosurgery，2006，58（4）：710－8； discussion 710－718.

［64］Letteboer MM，Willems PW，Viergever MA，et al.Brain shift estimation in image-guided neurosurgery using 3-D ultrasound［J］.IEEE Trans Biomed Eng，2005，52（2）：268－276.

［65］Rasmussen IA Jr，Lindseth F，Rygh OM，et al.Functional neuronavigation combined with intra-operative 3D ultrasound：initialexperiences during surgical resections close to eloquent brain areas and futuredirections in automatic brain shift compensation of preoperative data［J］.Acta Neurochir （Wien），2007，149（4）：365－378.

［66］Arbel T，Morandi X，Comeau RM，et al.Automatic non-linear MRI-ultrasound registration for the correction ofintra-operative brain deformations［J］.Comput Aided Surg，2004，9（4）：123－136.

［67］Lunn KE，Paulsen KD，Lynch DR，et al.Assimilating intraoperative data with brain shift modeling using the adjointequations［J］.Med Image Anal，2005，9（3）：281－293.

［68］Coenen VA，Krings T，Weidemann J，et al.Sequential visualization of brain and fiber tract deformation during intracranialsurgery with three-dimensional ultrasound：an approach to evaluate the effect of brain shift［J］.Neurosurgery，2005，56（1 Suppl）：133－141； discussion 133－141.

［69］Roberts DW，Miga MI，Hartov A，et al.Intraoperatively updated neuroimaging using brain modeling and sparse data［J］.Neurosurgery，1999，45（5）：1199－1206； discussion 1206－1207.

［70］Cao A，Thompson RC，Dumpuri P，et al.Laser range scanning for image-guided neurosurgery：investigation ofimage-to-physical space registrations［J］.Med Phys，2008，35（4）：1593－1605.

［71］Ding S，Miga MI，Thompson RC，et al.Estimation of intra-operative brain shift using a tracked laser range scanner［J］.Conf Proc IEEE Eng Med Biol Soc，2007：848－851.