Introduction
Neurosurgeons continuously strive to improve the safety and effectiveness of their interventions. Stereotactic neurosurgery uses imaging modalities such as computed tomography (CT) and magnetic resonance imaging (MRI) to serve as spatial guides for the surgeon. These spatial guides provide patient-specific anatomical roadmaps which, although not a replacement for knowledge of neuroanatomy, aid the surgeon in successfully treating the patient.
The theory behind stereotactic neurosurgery is relatively simple. Three points define a volume in geometric space. If the same three points can be defined on a patient and on an image of that patient, then the three-dimensional space of that patient and image are known and can be defined relative to each other. Registration is the process whereby the location of any point on or within the patient is defined on the image, and vice versa. Stereotactic systems differ in the manner in which the spatial points are defined, the geometric coordinate spaces used, and the method to register the patient and image coordinate spaces.
Frame-based systems involve fixing a structure to the patient’s head. The patient and frame are imaged together and the two spaces (patient and image) are registered. Points in the patient’s space were defined based on how the frame hardware was designed, such as an x, y, z coordinate system, a radius and angle system, or some variation of the two. Today’s frame-based systems are highly accurate and have become the gold standard for localization techniques. Frame-based systems are bulky and uncomfortable for the patient, however, and frames can obstruct the operative field. Most systems limit the surgeon to targets along a straight-line trajectory, and none offer real-time feedback.
Roberts et al. [1] and Friets et al. [2] introduced frameless stereotactic systems in the 1980s. Many different frameless stereotactic systems have been developed since that time. Some aspects are universal to all systems, frameless and frame-based alike. All systems must define points in the image and the patient and be able to map the two groups to each other.
Point Definition and Registration
Two common methods for defining points are surface-based and point-based registration. Each offers a distinct set of advantages and disadvantages.
Surfaced-based registration fits a set of points from the contours of one image to a surface model from contours of the patient’s head or from other images. An advantage of this system is that it allows retrospective registration from prior images. Its disadvantage, however, is that it is less accurate. Point-based registration involves selecting corresponding points in different images and on the patient. The coordinates of each set of points are defined, and then a geometric transformation is calculated between them; these points may be anatomical landmarks or may be applied artificial markers. Anatomical landmarks allow retrospective registration with existing images. Artificial extrinsic markers do not allow this, but have other advantages that have made them common in stereotactic guidance.
Extrinsic point registration allows greater accuracy than other registration methods. In addition, a calculation of the accuracy for each registration attempt is possible, giving the user a quantitated degree of accuracy. As long as a detectable marker can be manufactured, registration can theoretically be performed with any imaging modality.
Mobile markers and rigid markers are two types of extrinsic markers that are currently available. Mobile markers, which are taped, glued, or otherwise affixed to the patient’s skin, have the advantage of ease and speed of application. They are, however, prone to error because the skin may move relative to the skull and intracranial contents during registration. Rigid markers eliminate this potential error, but are more difficult to apply and may be more uncomfortable for the patient because they are anchored to the patient’s skull or other bony structure.
Three noncolinear points define a volume in space. Computer studies have analyzed the addition of more points to a registration system and its effect on registration error. It was found that four points increased registration accuracy over three, but that adding further points did not significantly improve accuracy [3]. However, if external mobile fiducials are to be placed, additional markers offer the benefit of redundancy should one or more markers become displaced. Fiducials should encompass the intracranial volume of interest and be noncolinear to ensure the most accurate registration possible. Small areas of hair shaving may be necessary.
Imaging
Once the markers are positioned, the patient is imaged with one or more of several possible modalities. The imaging need not be either on the day of marker positioning or the day of surgery, although markers are more apt to be dislodged with time. Fiducial markers are commonly filled with a material that is visible on T1-weighted, T2-weighted, and proton-density MR images. MR angiography and venography, functional MR, and MR spectroscopy may be performed, as necessary. Fiducial markers for CT scans are easily available, and thin slice CT images (such as 3 mm) are usually obtained. Positron emission tomography (PET) and single photon emission computed tomography (SPECT) scans may be performed with fiducials that contain special materials.
Intraoperative Localization
It is necessary to determine the location of a point in space to register the patient and image spaces. A variety of devices have been developed that communicate the location of some device (wand, microscope, surgical instrument, etc.) relative to the patient to a computer, which then translates that position into image space. The Zeiss MKM microscope (Zeiss Corporation, Jena, Germany) combines a robotic arm mechanism and an optical microscope. The Neuronavigator (ISG Technologies, Toronto, Ontario, Canada, and FARO Medical Technologies, Miami, Florida) uses a passive localization arm, as does the OAS System (Radionics, Burlington, Massachusetts). The PUMA Industrial Robot (Westinghouse Electric, Pittsburgh, Pennsylvania) is an active robotic arm system. Stealth System (Bucholz;
Surgical Navigation Technologies, Boulder, Colorado) and the ACUSTAR I (Codman/JJPI, Raynham, Massachusetts) localize with infrared light-emitting diodes. The Utah Machine Vision System (Heilbrun) and the VISLAN System (Thomas; London, United Kingdom) use passive stereoscopic video [4-15].
All localizing systems have their disadvantages. Mechanical arms usually use either potentiometers, which are prone to drift, or digitizers, which are expensive, to determine the position of the tip relative to a fixed base. The arms, although simple, can be bulky. Sonic systems use the time delay between a pulse emission and its detection to calculate position. Although these systems are inexpensive and durable, they are prone to ultrasonic noise and require an unobstructed ”line-of-site” between emitters) and detector. Magnetic systems detect position using magnetic waves and fields and, like mechanical arms, do not require an unobstructed line-of-site. A major disadvantage is the distortion that can be created by metal objects in or near the operative field. Optical systems, whether infrared emitters, fluorescent markers, or video cameras, all suffer from line-of-site difficulties [16].
The patient usually requires repositioning intraoperatively, and a system must be able to continuously monitor head positioning and adjust the registered coordinates rapidly, accurately, and in a minimally invasive manner. Detectors mounted to the table move with the head and fiducial markers and provide a simple and inexpensive solution. These detectors are often bulky and limit operative exposure and are not an ideal solution. Reference markers that communicate with a remote sensor, thus allowing tracking of the registered points, are another solution. If a marker is mounted on the head or head-holding assembly directly (such as the Mayfield head fixation device), head movement problems are minimized. A direct line-of-site must be maintained between the reference marker and remote sensor [16].
INTRAOPERATIVE VIDEO DISPLAY
Once the images have been registered with respect to the patient, they must be displayed in a meaningful manner. A high-resolution monitor with 512 X 512 pixel windows displays images in a variety of orientations and configurations. Color graphic overlays that represent the localizer position and trajectory are usually displayed relative to the on-screen images. The local-izer position is usually updated on the screen at 20 frames per second. Realtime displays have become more common with increases in image processing speed and power and decreases in cost.
ACCURACY AND ERROR
Certain terms should be defined in discussions of any stereotactic system. Unbiasedness is a lack of skew. Precision is a tendency to approach a certain value. Observations may be precise (minimal spread from a mean value) but biased (that mean value is not the true value). Accuracy requires both un-biasedness and precision. In an article about three-dimensional digitizers, Wohlers describes the confusion about accuracy in his sidebar "How Accurate is Accurate?”
Comparing the accuracy of 3-D digitizers can be confusing. The terms "accuracy," "precision," "resolution," and "repeatability" are often misused, misunderstood, misleading or just plain omitted. Buyer confusion also occurs when digitizer suppliers publish numbers related to the specific parts . . . that make up the system. This can be misleading. The precision of the system’s mechanics does not reflect the accuracy of the process. In fact, the precision of the mechanics is . . . always better than the accuracy of the process [17].
The accuracy of any stereotactic system is influenced in an additive manner by the error introduced at each step. A system must correctly identify the location of each fiducial during the localization process and must then accurately map patient fiducials onto image fiducials (or image fiducials to image fiducials) during registration. The sum of the root mean square error of each fiducial location can be calculated and is related to the fiducial localization error. The surgeon needs to know that the structure he is pointing to in the patient is the same structure displayed on the image. Any error in this, the target registration error, cannot be calculated by the system and is not directly related to the registration error. The registration error number may be reduced by eliminating fiducial makers, but if the remaining markers are distant from the operative site, the accuracy of target localization may decrease. Anatomical landmarks, such as the lateral canthus, are useful guides for checking the system and avoiding errors.
The effect of tissue displacement between imaging and surgery and during surgery must also be considered. Movement of tissues is inevitable during surgery, and studies have shown as much as 1 cm of displacement of superficial brain tissues [18]. Deeper structures, such as those at the skull base or along the falx shift less than superficial structures, improving the accuracy of stereotactic systems at those regions. Steps can be taken to minimize the error caused by tissue displacement. Displacement is often greatest in the direction of gravity. Structures may shift downward along a given trajectory. Minimizing diuretic use does not seem to have an appreciable effect on displacement. Cerebrospinal fluid shunting can cause shifts of tissues in unpredictable directions, so its use should be avoided until after the primary approach has been made, if possible. Debulking of tumors in the ”inside to outside” approach also yields significant tissue shifts; defining the outer margins of a tumor before proceeding with debulking may help to minimize this error [19-22].
Frameless stereotactic systems incorporate many of the advantages of frame-based systems without the discomfort and bulkiness of the frame. The systems differ in ease of use, cost, and availability, and at this time no one system is clearly superior to the others. The continued improvement in processor speed and cost and the advances in engineering promise exciting future developments in frameless systems.
The neurosurgeon considering the frameless stereotaxy for his practice has several options. Most commercially available systems (including those made by Medtronic/SNT, BrainLab, Radionics, and Leibinger) use optical infrared-based technology. By tracking a probe in a magnetic field, Cygnus offers a frameless stereotaxy device at a lower cost than the IR-based systems, but concerns regarding interference with ferromagnetic materials and resulting loss of accuracy have limited its appeal.
To make frameless stereotaxy a routine part of most neurosurgeons’ operating room, certain criteria should be met:
1. The system should accept standard CT and MRI scans from any standard imager. Scan data should be transferable to the operating room computer via portable media, such as optical disk, as well as over an ethernet connection (which would be the preferred method). Ability to run the system software on an office or home personal computer is useful to allow for convenient preoperative planning. Image fusion of CT, MRI, and any other appropriate digital studies (e.g., functional MRI or PET) should be possible.
Versatility is important as well. Most neurosurgical groups will want the ability to use frameless stereotaxy in the placement of spinal instrumentation; packages designed for ENT approaches may be useful for trans-sphenoidal approaches.
2. The system should be as easy to use as possible. In practice, this means that the neurosurgeon should not have to supervise the imaging and data transfer himself; rather, a physician’s assistant, nurse, or other ”extender” should be able to manage the downloading of the scan and at least begin the registration process in the operating room. Ideally, the neurosur-geon will confirm the accuracy of the registration and use the information to plan the surgery itself.
During the operation, the system should be easily controllable from the sterile field or be easily manipulated by other operating room personnel, even those without significant experience in frameless stereotaxy.
3. Cost must be kept in mind. A ”fully-loaded” system typically will run between $300,000 and $400,000. Few, if any, physician groups will want to bear this cost themselves and will ask their affiliated hospital(s) to purchase a frameless stereotaxy device. Therefore, caveat emptor, as this purchase will have to service you for years to come. Note that the preoperative scans almost always will be done before admission to the hospital, allowing the facility to amortize the purchase costs of the unit by charging for these studies.
A NOTE ON INTRAOPERATIVE IMAGING
The desire to compensate for brain shift and to obtain tumor resection control has led to a strong interest in intraoperative imaging, especially with MRI (discussed elsewhere in this volume). Some of the intraoperative MRI (iMRI) units that are commercially available or under development incorporate frameless stereotaxic technology. In time, they may prove to be preferable to ”standalone” units for certain operations, such as low-grade glioma resections or transsphenoidal removal of pituitary macroadenomas. However, it is far too early to suggest that, for the bulk of stereotactic approaches, the added cost and cumbersomeness of iMRI will make frameless stereotaxy obsolete. In fact, for navigation to fixed structures where updated imaging is relatively unimportant, such as parasaggital meningiomas and pituitary microadenomas, frameless stereotaxy will remain the ideal technology for the foreseeable future.
