The process of registration consists of the establishment of a rigid relationship between two coordinate systems: that by which the images were obtained
and that superimposed on the surgical field. Once images have become registered to the patient, the surgeon can translate from any point on the patient to the same point on the images, and vice versa, using a transformation matrix dictated by the registration process. This process, easily allowed by computers, permits the surgeon to plan a procedure on the preoperative images and superimpose the plan onto the patient's anatomy. In this fashion, registration can be viewed as a Rosetta Stone that allows translation between the 3D space of the images (a virtual space) and the anatomical structure of the patient (a real space).
By its very nature, image guidance requires registration in order to render the images of the patient's anatomy useful to the surgeon. Registration can be performed by a number of different methods, and, as with all factors relating to surgery, there is a tradeoff between accuracy and ease of use. Easier methods are, as a rule, less accurate; highly accurate methods are more difficult to employ and are generally more invasive for the patient. Furthermore, registration methods can require significant computational ability and therefore were not feasible prior to the introduction of computers into the operating room. As even the most computationally complex registration algorithms are easily performed by standard computers available today, computational demands will not be taken into consideration in the following discussion. This requirement for computer power for registration did, however, limit image guidance in its early history to devices that required little in the way of computer power. These simpler and more traditional methods are familiar to many neurosurgeons, who often regard them as being inherently more accurate. However, it is impor tant to realize that traditional methods are not necessarily more accurate because they are less complex. Computationally demanding registration techniques, even those not immediately comprehensible by the surgeon, may actually afford greater accuracy compared with traditional techniques. The intense use of computation power does not imply that a process is inherently less accurate, and accuracy can only be determined by a rigorous study of accuracy in side-by-side comparisons of two registration techniques.
The most direct registration method involves the use of intraoperative imaging. By taking an image in the operating room, with a surgical instrument or implant in place, one can immediately check suitability of position without resorting to any computational method or registration technique; the image literally speaks for itself. Employed in this way, intraoperative imaging has been used to reduce the invasiveness of many procedures, including the use of fluoroscopy during transsphenoidal pituitary surgery (3), placement of intracranial electrodes (4), breast needle biopsies (5), and minimally invasive spine surgery (6). The vast majority of experience with intraoperative imaging involves fluoroscopy, given its ready availability and ease of use. A major limitation of intraoperative fluoroscopy is its inherently 2D nature (the same limitation also being applicable to ultrasound) and the tissue contrast of the images obtained. Even though a rough appreciation of location in three dimensions can be obtained by taking a matching pair of images obtained approximately orthogonal to one other, precise 3D localization is not possible with standard fluoroscopy. Furthermore, fluoroscopy is in general useless for determining location relative to soft tissue targets, as the soft tissue contrast of this imaging modality is very limited. As a result, there has been considerable interest in using inherently 3D imaging techniques intraoperatively.
A few centers have used intraoperative CT but have found this technique to be useful only for localization of radiologically detectable structures such as bone or implants, as the ability of CT to detect soft tissue is almost as impaired as that associated with fluoroscopy (7). This has motivated a significant research effort into the use of intraoperative MRI, using modified scanners that allow access to the patient, scanners that allow the patient to be moved into the scanner, or miniature scanners that can be placed about the head of the patient. This considerable variation in approach, with some centers using high-field magnets (8) and other centers using magnets with far weaker field strengths, specifically designed for use in the operative situation (9), indicates that no consensus has been reached as to the most useful intraoperative MRI device (see Chapter 6).
Regardless of the technology used, it should be noted that imaging does not occur simultaneously with surgery. Instead, a preliminary set of scans is obtained for guidance, and then surgery is performed, followed by a second course of imaging, which is analyzed to decide what further surgery should be performed. The serial nature of this process interrupts the normal flow of surgery to accommodate imaging, which increases the overall duration of surgery and thus has the unintended effect of increasing the invasiveness (in terms of anesthesia time) of the procedure. In addition, as the position of the instrumentation needs to be related to the images taken in the operating room, some form of tracking mechanism is mandatory. The technique of tracking is usually identical to the means employed by image guidance systems used for navigation with preoperative images. As long as the use of these navigational systems is required, there is no reason that preoperative images could not be employed as well, especially if they convey information not easily obtained in the operating room, such as functional localization. Furthermore, the quality of intraoperative imaging, particularly with the lower strength magnets, is almost invariably inferior to images obtained with purely diagnostic machines. Therefore, most systems provide some means of relating the intraoperative images to the pre-operative diagnostic studies to allow enhanced interpretation. As has been experienced with fluoroscopy, the quality of imaging used for therapy does not have to match the quality of images used for diagnosis.
Alternatively, if only preoperative images are to be used, some form of registration is needed. When stereotactic surgery was first employed, computational capabilities were limited, and registration was obtained by making the coordinate systems of the images and the operating room identical (10). This was accomplished by attaching a frame to the patient's anatomy, which established the plane of origin of the preoperative images. A localizer was attached to this frame that produced marks in the images obtained and established the coordinate system of the frame within the images. The technique became more extensively employed when 3D imaging techniques, such as CT, became available. As early tomographic devices had inaccuracies in their stepping procedures between subsequent scans, the localizers employed were modified with the addition of redundant image markers, called fiducials, that allowed the precise orientation of each scan in relation to each of the other images and the frame itself (11). The redundancy of this marking system, added to the rigid nature of the attachment of the frame to the head, led to a reasonably accurate registration technique that had the additional benefit of holding the patient stable during both imaging and surgery. This rigidity makes these systems preferable even today when one is dealing with uncooperative or confused patients. However, the need to have the frame and localizer in position during imaging made the routine use of stereotaxis very difficult, and few surgeons availed themselves of the accuracy of these systems.
In the late 1980s two developments allowed the use of less invasive registration methods for image guidance. First, CT scanners were improved in their ability to step accurately in the z-dimension between scans. Second, computational capabilities sufficiently inexpensive to be used routinely in the operating room became available to translate instantaneously between two coordinate systems arranged at random angles and displacements with respect to each other. By introducing at least three points in the images, which have to be clearly seen on the imaging modality employed, point fiducials can be employed to replace the redundant fiducials of framed systems. If these points are then touched in the operating room with a device that could precisely determine position in 3D space (a device called a 3D digitizer), the second set of coordinates can be paired with the position of the same fiducials on the images to produce a translational matrix. Increasing the number and spacing of point fiducials generally results in improved accuracy. Important assumptions inherent in such a process are that the fiducials would approximate as close as possible a point in 3D space and that the relative position of the fiducials with respect to the underlying anatomy does not vary between the process of imaging and the end of the surgical procedure.
This technique allows the easy dissemination of stereotactic technique to almost all intracranial procedures. A major concern remains the technique of attaching the fiducials to the patient. As many procedures do not require high accuracy, a common technique is to use self-adhesive fiducials applied to the scalp. Given the mobility of the scalp, the need for a rigid relationship between the fiducials to each other and the underlying anatomy during imaging and registration is only partially satisfied. This error can be reduced by using a multiplicity of fiducials and a process that selects those fiducials that seem to have maintained a rigid relationship between one another. However, for maximal accuracy of registration, the need for a fixed relationship between registration and imaging can only be achieved with fiducials attached to the skull. It has been demonstrated that use of such fiducials matches, if not exceeds, the accuracy of frame-based registration (12). Given this demonstrated accuracy, it is surprising that frames are still used for applications, such as functional neurosurgery, that are perceived as requiring the utmost in accuracy. It is to be anticipated that all applications will eventually use point-based registration methods, given the ease with which these techniques can be used and the decrease in discomfort experienced by the patient. Clearly, the act of incising the scalp and screwing a fiducial into the skull simply to achieve registration cannot be perceived as adding to the concept of minimal invasiveness; when using point-based fiducial registration, an important tradeoff between accuracy and minimal invasiveness must be considered by the surgeon.
A final method exists that is highly accurate and completely noninvasive. If two X-ray sources, oriented at approximately right angles to each other, are employed to take an image of an anatomical structure at the junction of their beams using either radiographic film or radiosensitive digital devices, the two resultant 2D images can be employed to determine the exact position and orientation of the anatomical structure in space. To perform registration, a highresolution CT scan of the anatomical part is required. By using a projection algorithm, multiple "virtual radiographs" of the CT dataset can be produced, emulating various angles of source and detector location around the anatomic part. By comparing this library of "virtual radiographs" with the actual radiographs taken in the procedure suite, the exact location of the head relative to the source and detectors can be determined. This registration technique is used for the CyberKnife, discussed in the Therapeutic Intrevention section; as it is totally noninvasive, it can be repeated as necessary with no risk to the patient. This capability allows for fractionated stereotactic radiosurgery and is highly accurate (13), but has not as yet been employed in the operating room.
Registration has therefore gone through a cyclical process. It was initially avoided entirely by using intraoperative 2D imaging and then enhanced using 3D preoperative images that were employed through the use of an invasive, but accurate, stereotactic frame. Use of scalp-based markers eliminated the use of the frame in all procedures except those needing the highest accuracy; skull-based markers will eventually eliminate the use of the frame in all procedures owing to their high accuracy, but at the expense of some invasiveness in terms of pain in placing these markers. Finally, a method now exists to use 2D intraoperative radiographs to register high-resolution 3D preoperative images in a completely noninvasive technique. Technology has therefore addressed the issue of maximizing registration accuracy while essentially eliminating registration invasiveness.
Was this article helpful?
Do You Suffer From High Blood Pressure? Do You Feel Like This Silent Killer Might Be Stalking You? Have you been diagnosed or pre-hypertension and hypertension? Then JOIN THE CROWD Nearly 1 in 3 adults in the United States suffer from High Blood Pressure and only 1 in 3 adults are actually aware that they have it.