Use of Cortical Surface Vessel Registration for Image-Guided Neurosurgery
Shin Nakajima, M.D.1, Hideki Atsumi, M.D.2, Ron Kikinis, M.D.1, Thomas M. Moriarty, M.D., Ph.D.2, David C. Metcalf, M.S.1, Ferenc A. Jolesz, M.D.1, Peter McL. Black, M.D., Ph.D.2
1Department of Radiology and
2Division of Neurosurgery, Brigham and Women's Hospital; Brain Tumor Center, Brigham and Women's Hospital, Children's Hospital, Dana-Farber Cancer Institute, and Joint Center for Radiation Therapy; Departments of Radiology and Surgery, Harvard Medical School Boston, Massachusetts
Running title:Cortical vessel registration
Address correspondence to:
Shin Nakajima, M.D.:
Surgical Planning Laboratory, Department of Radiology, Brigham and Women's Hospital
75 Francis Street, Boston, MA 02115, USA
Phone: (+1)(617) 732-7692
Fax: (+1)(617) 732-7963
We have treated patients with brain surface tumors using video registration of a 3D image with the surgical field to identify an eloquent cortex, localize the lesion and define the tumor margin. "Skin to skin" registration using the skin surface to produce alignment was performed in earlier patients, but was difficult in areas with few prominent registration landmarks. For this reason, "vessel to vessel" registration using the cortical vessels as fiducials was applied to 17 cases to improve the accuracy. This paper presents the advantages and limitations of "vessel to vessel" registration from the data of these cases. The accuracy is also estimated.
A 3D model was reconstructed from MR data and a 2D projection was superimposed onto the video image of the actual surgical field. The tumor was resected, guided by the registered video image. The 2D projection accuracy of "vessel to vessel" was compared to that of "skin to skin" registration using a phantom study.
All 17 tumors had gross total resection, and the patients had no major permanent neurological deficit. In the phantom study, the 2D projected target registration error of a tumor in "skin to skin" registration was estimated as 8.9 ± 5.3 mm and that of "vessel to vessel" registration was 1.3 + 1.4 mm (99th percentile confidence intervals are 24.8 mm and 5.5 mm, respectively).
Video registration using cortical surface vessels is practical and improves 2D projection accuracy significantly, over skin registration.
Three dimensional reconstructed CT and MR images are currently available for diagnostic or therapeutic purposes. For several years, we have been making 3D brain computer models using preoperative MR data for surgical planning [6, 9]. Interactive manipulation of this 3D model provides comprehension of anatomical relationship and a simulated surgical field. To localize the lesion during surgery, we used video registration, superimposing a projection of the 3D model onto the video image of the actual surgical field . The registration was guided by skin surface landmarks, such as the external auditory meatus, nasion or lateral canthus, and worked quite well for many cases. This "skin to skin" registration, however, was still difficult in areas with few prominent registration landmarks, such as the top of the head.
We attempted to use the cortical vessels as fiducials for registration to improve the accuracy, and as a "map" to identify each anatomical structure as well. For these purposes, the cortical vessels, which tend to be neglected in many other 3D reconstructions, were carefully reconstructed and "vessel to vessel" registration was applied to clinical cases.
In this paper, the cases using the cortical vessel registration are reviewed, and the 2D projection accuracy of "vessel to vessel" registration, compared to "skin to skin" registration, is estimated. We found that "vessel to vessel" registration significantly improved accuracy.
PATIENTS AND METHODS
From January 1995 to January 1996, we performed both "skin to skin" and "vessel to vessel" video registration for 17 patients (ages 15-72 years) with brain tumors (15 gliomas, 2 metastatic tumors). Each patient's profile is detailed in Table1.
Data acquisition and transfer
The MR data were obtained by a 1.5T MR unit (Signa¨, GE Medical Systems, Milwaukee, WI). A series of 124 images of postcontrast SPGR (3D Fourier transformation, 1.5 mm thickness, 256 x 256 matrix of 220-240 mm FOV) was acquired in 14 min 24 sec and a series of 120 images of postcontrast phase contrast angiogram (3D Fourier transformation, 1.0-1.5 mm thickness, 256 x 256 matrix of 200-240 mm FOV) was acquired in 18 min 35 sec. The data were digitally transferred to a Sun computer workstation (SPARC-20¨, Sun Microsystems, Inc., Mountain View, CA) via a computer network.
Each image was preprocessed automatically using anisotropic diffusion filtering to reduce noise , and an expectation maximizing algorithm to correct the spatial non-uniformity of the signal intensity . The images of the phase contrast angiogram were registered to the SPGR images using a maximization of mutual information algorithm , visually inspected and reformatted to 124 sagittal images of 1.5 mm thickness with the same FOV as the SPGR images. After preprocessing, the segmentation of both SPGR and reformatted phase contrast series was performed by a thresholding-based technique. A connectivity algorithm [1,2] and some manual editing were used afterwards to complete the skin, brain, tumor, vessels, and other related structures. This segmentation process needed 1-5 hours of human intervention. Using the marching cubes algorithm and surface rendering [1, 2, 3]we made 3D objects of each anatomical structure from the series where the structure was best articulated. These objects were integrated into a 3D model. It took a total of 5-20 hours to complete the 3D model after MR data transfer. The model was colored and displayed on a Sun computer workstation (SPARC-20¨ with video board, Sun Microsystems, Inc., Mountain View, CA) with 3D display software (Leotool¨, Sun Microsystems, Inc., Mountain View, CA). In this software, each object can be individually colored, made translucent, removed, rotated, translated and scaled as the viewer wishes.
Before surgery, we identified each relevant important anatomical structure, such as the central sulcus, precentral sulcus, precentral gyrus, or superior temporal gyrus, by precise observation of this 3D model according to the morphologic features [4, 13]. Then, the anatomical relationship between the tumor and these structures was examined. The main interests were whether an eloquent cortex was affected by the tumor, and through which sulcus or cistern the tumor was accessible . These issues were explored on a 3D model of the brain and the tumor, with cortical vessels not shown. The appearance of the brain surface in the actual surgical field, however, is totally different from that of the brain-tumor 3D model, because the brain surface is covered by the vascular network. Therefore, the next concern was determining which cortical vessel on the full 3D model covered each sulcus, and which vessel indicated the tumor margin. For this purpose, the surgical scar on the brain surface was also used as a landmark when possible. After this surgical planning was completed, the anatomical relationship among the eloquent cortices, the tumor, and the cortical vessels was well understood.
Display and video registration
We registered this 3D model to the patient in a two-step method, "skin to skin" registration and "vessel to vessel" registration. In the operating room, the 3D model was displayed on the Sun workstation screen. The high resolution NTSC format image was then converted to a standard resolution video format using a scan converter (CVS-980 NTSC scan converter¨, YEM, Okada, Japan). After the patient's head was positioned and fixed, a CCD video camera (TK-1070U¨ with HZ-C611AF¨ lens, JVC, Tokyo, Japan) was placed on the floor near the patient so that it could have a similar viewpoint to that of a surgeon. The image from the video camera was merged with the image of the 3D model using a video mixer (WJ-AVE5¨, Panasonic, Tokyo, Japan), and displayed on the TV monitor. Using the video signal mixer, the signal intensity weighting for each image could be varied from 0% to 100%. Then the 3D model was translated, scaled, and rotated, and a 2D projection was superimposed onto the video image of patient's head in the TV monitor (Figure 1). We used the external auditory meatus, nasion, lateral canthus, and, if available, the previous surgical scar as landmarks for registration ("skin to skin" registration). It took approximately 3 minutes to align the 3D model to the patient's head. Then the skin incision was performed in accordance with this video registration. After opening the dura mater, we identified the visible cortical vessels, and refined the video registration using the cortical vessels as fiducials ("vessel to vessel" registration). It took approximately 1 minute to fit the 3D model to the surgical field according to the shape of the cortical vessels. The tumor margin was frequently ambiguous from direct observation of the brain surface, but was easily defined using the 3D model displayed on the monitor. The video mixer was adjusted back and forth to assist visualization. Combined video was recorded for later review.
Fourteen patients were operated on under local anesthesia, and 3 patients under general anesthesia. Each patient's head was fixed with a head clamp preventing movement. After opening the dura mater, each cortical vessel, each sulcus and gyrus, and the abnormal area was identified, and the surgeon outlined the tumor's margin on the brain according to the "vessel to vessel" video registration. Electrical cortex stimulation with a bipolar isolated forceps was performed to confirm the location of each eloquent structure, such as the primary motor area, primary sensory area or language area. Then the tumor was excised, while the patient's neurological status was monitored.
Since we are registering a 2D projection of our 3D model to a 2D video image of the surgical field, only 2D projection accuracy can actually be measured. This 2D projection accuracy should be distinguished from a true 3D error which is used for accuracy estimation of stereotactic systems . It could be argued that, since the projection is generally close to perpendicular to the surface, the 2D projection error, represented by misregistration along the surface, represents a larger component in the true 3D error than does the error in the direction of projection, represented by misregistration normal to the surface. However, for the remainder of this section it should be assumed that all accuracy and error measurements refer to errors in the projected plane only. In accordance with the definitions used in Maurer et al. , we define "fiducial registration error" as the distance between corresponding fiducials after registration, and "target registration error" as the distance between corresponding non-fiducial points. Thus, in "skin to skin" registration misalignment, both the vessel branch points and the lesions represent target registration errors. In "vessel to vessel" registration, the distance between corresponding vessel branch points represents fiducial registration error, and the misalignment in the lesions correspond to target registration error. In the clinical cases, it is not possible to measure directly the target registration error of the tumor (the discrepancy between the tumor on the video and in the 3D model), because the tumor is unclear in the actual surgical field. However, the discrepancy between the vessel branch points on the video and in the 3D model can be measured. The target registration error of a vessel branch point in "skin to skin" registration was measured as follows: After opening the dura mater, the actual surgical field and the registered display of the 3D model were recorded to a video tape. Both images were digitized to TIFF format images. An appropriate vessel branch point in the surgical field was compared to the same structure in the 3D model (Figure 2). For each landmark in the two TIFF format images, the discrepancy in pixels was measured. The image pixel dimensions were calibrated by comparing the size of prominent structures in the video images and in the MR images, and these dimensions were used to calculate the errors in millimeters. Then, fiducial registration error of "vessel to vessel" registration was measured. This occurs mainly from misalignment of rotation and scaling of the 3D model. The spatial errors of four vessel branch points in each case were estimated in the TIFF images digitized from the recorded video tapes. In order to get an estimate of 2D target registration error of the lesions, and to confirm the vessel error measurements, a phantom study was performed. The phantom was made using a styrene foam mannequin head (#800W¨, John J. Cahill Display Inc., Boston, MA)(Figure 3A). The cortical vessel model was made with modeling compound (Play-Doh¨, Playskool Inc., Pawtucket, RI) and was fixed on it. Seventeen artificial landmarks (Multi-Modality Radiographic Markers 3004¨, IZI Medical Products, Baltimore, MD) were attached on the mannequin head in the frontal region (6 points), parietal region (3 points), temporal region (5 points) and occipital region (3 points). The surface of the phantom was coated with agar (Rakuraku-Kanten¨, Yamashin Co. Ltd. Nagoya, Japan) so that it could be visualized in the MR images (Figure 3B). After the MR scanning of this phantom, a 3D model, including artificial landmarks, the skin and the cortical vessels, was constructed using the same image processing method as the clinical cases. The phantom was placed on the table, and either "skin to skin" registration or "vessel to vessel" registration was performed (Figure 2). The 2D projected errors of 17 targets were measured 3 times in both registration methods on the phantom surface (target registration error for an artificial landmark). In "skin to skin" registration, the error of a cortical vessel branch point close to the target was also measured in each trial (target registration error for a vessel branch point) in the same way as was done for the clinical cases. In "vessel to vessel" registration, errors of four vessel branch points were measured in each trial using the phantom measurements (fiducial registration error).
The 3D models of these 17 cases well described the brain surface features, and were useful for surgical planning. The tumors and eloquent cortices were clearly identified on these 3D models, and the spatial relationship between these important structures and the cortical vessels were confirmed before surgeries. That is, if there were a large cortical vein in the central sulcus on the 3D model, the precentral gyrus would be localized anterior to the vein after identifying it in the surgical field. The video registration using the "skin to skin" registration method did not work well in every case. The cortical vessels on the brain surface were sometimes found to be misaligned after opening the dura mater. Each important structure, however, was easily identified after "vessel to vessel" registration in all cases. In the recurrent tumor case, it was difficult to identify the cortical vessels, because adhesion between the brain surface and the dura mater was severe (case 7). However, once a cortical vessel was found, the "vessel to vessel" registration became quite easy. There are a few cases in which the size of the 3D model vessels was different from that of the actual vessel. The large vessel in the surgical field sometimes looked small or invisible in the 3D model. We needed to be careful when identifying the cortical vessels in these cases. All patients had gross total resection of their tumors confirmed by postoperative contrast-enhanced MR images, and excellent outcome was achieved. In these 17 cases, the functional motor and sensory cortices confirmed with electrical cortex stimulation were consistent with pre- and postcentral gyri, respectively. In the phantom study, the 3D model well described the models of cortical vessels and fiducials. Using the video registration method, the projection of the 3D model was overlaid onto the video image of the phantom. Each 2D projection error was measured in both "skin to skin" registration method and "vessel to vessel" registration method for 17 regions. In the phantom study, in addition to measuring the registration error for a vessel branch point (target registration error for "skin to skin" registration, but fiducial registration error for "vessel to vessel" registration) as in the clinical cases, target projected registration error measurements for an artificial landmark were also made for each registration method (Table 2). The target projected registration errors for an artificial landmark in "skin to skin" registration were from 0 mm to 20 mm (n=51), and the average was 8.9 ± 5.3 mm (99th percentile confidence interval of this value is 24.8 mm). This value was similar to target projected registration errors for a vessel branch point in "skin to skin" registration for clinical cases (from 0 to 17 mm, the average was 9.4 ± 5.1 mm, 99th percentile confidence interval of this value is 24.7 mm) and significantly larger than the target projected registration error for an artificial landmark in "vessel to vessel" registration (from 0 to 6 mm, the average was 1.3 ± 1.4 mm, 99th percentile confidence interval of this value is 5.5 mm)(Table 2). From this estimation, we conclude "vessel to vessel" registration is more accurate than "skin to skin" registration.
AN ILLUSTRATIVE CASE
An illustrative case of this technique is presented below.
Case 17 This 23-year old female patient presented with increasing left arm discord and numbness. The MRI showed a 4 x 4 x 3.5 cm mass in the right frontoparietal cortex, which proved to be an oligo-astrocytoma by a previous surgery. The 3D model of this patient was created from the SPGR MR images and phase contrast MR angiogram (Figure 4). According to this model, the tumor seemed to be in the central sulcus. A large cortical vessel was observed posterior to the tumor. We thought that this tumor was resectable, but there was a possibility of damaging the precentral and/or postcentral gyrus. In surgery, the tumor looked a little more swollen than the brain, and most of the boundary between the tumor and normal tissue was clear in appearance. The large cortical vessel was easily found, and the precentral and postcentral gyrus were identified, because this vessel was a useful landmark. The primary motor and sensory cortices were confirmed with cortical mapping, and were consistent with the pre- and postcentral gyrus, respectively (Figure 4D). With "vessel to vessel" technique, the video registration was performed according to the curve of the large cortical vessel (Figure 4C). Using video registration as a guide, gross total resection of the tumor was performed under local anesthesia . During surgery, the patient's response to commands was monitored. This patient showed slight sensory disturbance of her left arm postoperatively.
Before an operation, a neurosurgeon integrates the patient's individual radiological data and makes a surgical strategy. He uses MR, CT, conventional angiography, SPECT, and other modalities for this purpose and reconstructs the simulated surgical field inside his mind. All necessary information is included in this mental surgical field, while useless information is neglected. Likewise, the computer-generated 3D model for surgical planning should be an integration of necessary radiological information. Therefore, the reconstructed 3D model is an abstraction of the relevant information from a set of data.
"Skin to skin" registration and its accuracy
We have been making 3D models and applying them to surgical planning . For the cortical or subcortical brain tumor cases, we have attempted to localize the tumor and eloquent cortices using a video registration system . Since this system is based on a simple concept, it can be constructed with commercially available devices and from any type of 3D image. An additional advantage of this system is that it does not require any fiducials on the patients. It was, however, sometimes very difficult to get an accurate localization of the tumor and important anatomical structures with "skin to skin" registration in areas with few prominent registration landmarks. The target registration error in "skin to skin" registration was measured in the phantom study as 8.9 ± 5.3 mm. This error was estimated in a 2D plane of the recorded video tapes, and is not a true 3D error, because this system registered the 2D projection (display of the 3D model) to 2D projection (video image of the surgical field). Other frameless stereotaxy systems without fiducials have achieved a true 3D mean error of 6.2 mm using an anatomical landmark-surface fit algorithm method , and 3-8 mm with retrospective patient-image registration  in three-dimensional space.
"Vessel to vessel" registration and its usefulness
Early efforts to refine the registration based on the appearance of the sulcal and gyral patterns after opening the dura mater proved challenging. Sulcal and gyral patterns are often hard to recognize visually, even in healthy anatomy, because the brain surface is covered with the vascular network. In addition, the presence of a tumor often cause the sulcal and gyral patterns to become flattened against the skull; thus, frequently the surface of the brain modeled from MRI is lacking in recognizable features. The vessels visible on the surface of the brain, however, provide a good map for navigation, because they parallel the sulcal and gyral patterns, and these vessels are well imaged by phase contrast MR angiography even in tumor patients. Using cortical vessels as fiducials, the 2D projection accuracy of the video registration system was significantly improved to 1.3 ± 1.4 mm in the phantom study. The remaining error was caused by misalignment of rotation and scaling of the 3D model, perspective effects, and the inherent distortion of the original MR images. The cortical vessels used as fiducials are not lines but have some width, and this also causes error. A serious error could occur if there were no visible vessels close to the tumor on the 3D model or the craniotomy were too small to find prominent vessels. The misalignment could occur in registering MRI and MR angiogram. This potential error was estimated as 1.00 - 1.70 mm from 25 mm random offset initial registration . To avoid this error, we always visually inspected the alignment of MRI-MR angiogram registration. It is sometimes very difficult to visually define the tumor margin of glial tumors, even if it is well demarcated on the MR images. Stereotactic surgical resection is one conventional and successful solution. Our video registration system also enables the surgeon to define the border of the tumor. In addition, since the tumor contour is projected on the video image of the surgical field, video registration provides contextual information to the surgeon as well as the geometric coordinates. "Vessel to vessel" registration is also powerful for the intraoperative identification of cortical anatomy. Since most of the sulci are covered and hidden with the vascular net, it is difficult to identify a specific sulcus or gyrus without any vascular information. Before surgery, we identified important sulci and found appropriate cortical vessels as landmarks on the 3D model. During the surgery, the identification of the eloquent cortex was quite easy, once "vessel to vessel" video registration was achieved. Although the cortical vessels provide very important information for a surface lesion, their importance has not been well admitted.
Our method, however, still needs some improvement. It is necessary to improve the quality of the 3D model; particularly to be able to model a fine 3D vascular network. We are now using a simple thresholding technique to segment the phase contrast MR angiogram, but this does not provide a sufficient segmentation of all cortical vessels. It is also difficult to get a 3D object with the true width of the vessels, because the magnitude of the phase contrast angiogram is in proportion to the flow velocity of the blood vessel. That means, a small vessel with high flow could appear wider than a large vessel with low flow. As a result, it is difficult to identify the true size of each vessel. To overcome these problems, several experiments are in progress including automatic segmentation , line filtering , using three-directional flow information, and improvement of the original MR angiogram. If we had an accurate 3D object of the fine cortical vessels, the identification and registration of the cortical vessels would become much easier and more accurate even through a small craniotomy. Another basic problem is that the video image of the surgical field is 2D and does not have depth information while the 3D model has depth information. Therefore, our method is currently applicable only for surface tumors. To solve this problem, we integrated this video registration system with an optical 3D digitizer and started applying it to clinical cases.
The concept of "vessel to vessel" registration is an important addition to image-guided neurosurgery. Especially for the brain surface cases, the cortical vessel information should be included in 3D models.
The authors gratefully acknowledge the support of the NIH (Grant No. P01 CA67165-02). We also acknowledge many of our colleagues for their help and support in this work. In particular we thank Adam Shostack, Mark Anderson, Marianna Jakab, William Wells, Gary Zientara, Hiroshi Shinmoto, Nobuhiko Hata, Laverne Gugino, Kazue Nakajima, Maureen Ainslie and Mary Knapman.
1. Cline HE, Dumoulin CL, Lorensen WE, Hart HR, Ludke S: 3D reconstruction of the brain from magnetic resonance images using a connectivity algorithm. Magn Reson Imaging 5:345-352, 1989.
2. Cline HE, Lorensen WE, Kikinis R, Jolesz FA: Three-dimensional segmentation of MR images of the head using probability and connectivity. J Comput Assist Tomogr 14:1037-1045, 1990.
3. Cline HE, Lorensen WE, Souza SP, Jolesz FA, Kikinis R, Gerig G, Kennedy TE: 3D surface rendered MR images of the brain and its vasculature. Technical Note J Comput Assist Tomogr 15:344-351, 1991.
4. Ebeling U, Steinmetz H, Huang Y, Kahn T: Topography and identification of the inferior precentral sulcus in MR imaging. AJNR 10:937-942, 1989.
5. Gerig G, Kubler O, Kikinis R, Jolesz FA: Nonlinear anisotropic filtering of MRI data. IEEE Trans Med Imaging 11:221-232, 1992.
6. Gleason PL, Kikinis R, Altobelli D, Wells W, Alexander E III, Black P McL, Jolesz F: Video Registration Virtual Reality for Nonlinkage Stereotactic Surgery Stereotact Funct Neurosurg 63:139-43, 1994.
7. Golfinos JG, Fitzpatrick BC, Smith LR, Spetzler RF: Clinical use of a frameless stereotactic arm: results of 325 cases. J Neurosurg 83:197-205, 1995.
8. Harkey HL, Al-Mefty O, Haines DE, Smith RR: The surgical anatomy of the cerebral sulci. Neurosurgery 24:651-654, 1989.
9. Kikinis R, Gleason PL, Moriarty TM, Moore MR, Alexander E III, Stieg PE, Matsumae M, Lorensen WE, Cline HE, Black PM, Jolesz FA: Computer-assisted Interactive Three-dimensional Planning for Neurosurgical Procedures. Neurosurgery 38:640-651, 1996.
10. Koller T, Gerig G, SzŽkely G, Dettwiler D: Multiscale detection of curvilinear structures in 2-D and 3-D image data. Proceedings Fifth International Conference on Computer Vision (ICCV'95), Cambridge, Massachusetts, 1995, pp 864-869.
11. Maciunas RL, Galloway RL, Lattimer JW: The application accuracy of stereotactic frames. Neurosurgery 35:682-695, 1994
12. Maurer CR, Fitzpatrick JM: A review of medical image registration. in Maciunas RJ (ed): Interactive Image-Guided Neurosurgery. Park Ridge, American Association of Neurological Surgeons, 1993, pp 17-44.
13. Steinmetz H, FŸrst G, Freud HJ: Variation of perisylvian and calcarine anatomic landmarks within stereotaxic proportional coordinates. AJNR 11:1123-1130, 1990.
14. Tan KK, Grzeszczuk R, Levin DN, Pelizzari CA, Chen GTY, Erickson RK, Johnson D, Dohrmann GJ: A frameless stereotactic approach to neurosurgical planning based on retrospective patient-image registration. Technical note J Neurosurg 79:296-303, 1993.
15. Wells WM III, Viola P, Atsumi H, Nakajima S, Kikinis R: Multi-modal volume registration by maximization of mutual information. Medical Image Analysis 1:35-51, 1996.
16. Wells W, Kikinis R, Grimson W, Jolesz F: Adaptive segmentation of MRI data. IEEE Trans Med Imaging 15:429-442, 1996.
Diagram of the operative arrangement. The high resolution 3D model display is converted to a video signal and superimposed onto the video image of the surgical field by a video mixer. This merged image is displayed on the TV monitor to provide the information to the surgeon in the operating room.
The concepts of "skin to skin" registration and "vessel to vessel" registration, and their accuracy estimation. The 2D projection of the 3D model was superimposed onto the 2D video image of the surgical field using the skin surface. After "skin to skin" registration was performed, the artificial landmark was localized by the superimposed 3D model ("skin to skin" registration). When the brain surface was exposed during surgery, the cortical vessels were used as fiducials to register the 3D model to the surgical field ("vessel to vessel" registration). The target registration error for a vessel branch point (the error of a vessel branch point close to the target) of "skin to skin" registration and the fiducial registration error (the error of vessel branch points surrounding the target) of "vessel to vessel" registration were measured in the clinical cases. In the phantom study, target registration error for an artificial landmark was measured in both "skin to skin" registration and "vessel to vessel" registration.
The phantom used for accuracy estimation.
A:The dimensions of the phantom.
B: The phantom consisted of cortical vessels and artificial landmarks. The surface of the phantom was coated with agar so that it was visible in MR images.
An illustrative case of "vessel to vessel" registration.
A: An axial T2-weighted image showed a tumor in right front-parietal region.
B: The 3D model display consisting of the brain, cortical vessels and tumor. The tumor is colored green, and the vessels are colored blue. The upper side of the image is the patient's right side, and the left side is the patient's anterior side. There are a large cortical vessel (double arrows) and a vessel branch (single arrow).
C:The 3D model display was superimposed onto the video image of the actual surgical field. The primary motor cortex was confirmed by a bipolar electric stimulator (white arrow). Burr holes of the previous surgery are colored red (white arrow heads) on the 3D model. The contour of the tumor in the surgical field was consistent with that of the 3D model (black arrow heads).
D: The surgical field in the same alignment as the 3D model. The large cortical vessel (double arrows) and the vessel branch (single arrow) was easily identified on the surgical field. The primary motor cortex (squares) and the primary sensory cortex (triangles) were confirmed by cortical mapping. The tumor (arrow heads) existed between these two cortices. Note the sulci are difficult to identify because they are hidden under the vascular network.
TABLE 1: Patient Profile
|Case No.||Age/Sex||Pathology||Location||Critical Structures||Postoperative Neurology|
|1||15M||Oligodendroglioma||R.frontal||Primary motor cortex||Temporaryb difficulty with position sense|
|2||72M||Metastatic Colon Cancer||L.frontal||Primary motor/sensory cortex||No Deficit|
|3||46M||Anaplastic Oligo-astrocytoma||L.frontal||Primary motor cortex||Temporaryb speech arrest|
|4||49F||Glioblastoma||R.frontal||Primary motor cortex||No Deficit|
|5||27F||Oligodendroglioma||R. parietal||Primary motor/sensory cortex||No New Deficit|
|6||37M||Astrocytoma||R.temporoparietal||Primary motor/sensory cortex, visual cortex||No Deficit|
|7||21F||Astrocytoma||L.frontal||Primary motor/sensory cortex, Broca area||No Deficit|
|8||30M||Astrocytoma||R.temporal||Primary motor/sensory cortex, MCA||No New Deficit|
|9||47M||Glioblastoma||L.parietal||Primary motor/sensory cortex,Wernicke area||No new Deficit|
|10||24F||Ependymoma||L.frontal||Primary motor cortex||No Deficit|
|11||58M||Glioblastoma||L.temporoparietal||Wernicke area, Vein of Labbe'||Temporaryb difficulty with word finding|
|12||55M||Anaplastic Oligodendroglioma||R.temporal||Primary motor/sensory cortex,Vein of Labbe'||No Deficit|
|13||54M||Metastatic Lung Cancer||Bilateral parietal||SSS, Cortical veins||Temporaryb difficulty with motor initiation|
|14||50F||Glioblastoma||R.temporal||MCA,optic radiation||No Deficit|
|15||65M||Glioblastoma||R.temporoparietal||Primary motor/sensory cortex||No Deficit|
|16||28F||Oligo-astrocytoma||R.perietal||Primary motor/sensory cortex||Temporaryb slight sensory disturbance|
|17||23F||Oligo-astrocytoma||R.frontoparietal||Primary motor/sensory cortex||Temporaryb slight sensory disturbance|
a R=right, L=left, MCA= middle cerebral artery, SSS= Superior Sagital Sinus
b temporary; recovered within 30 days
TABLE2: Two-Dimensional Projection Accuracy of Video registrationa
Phantom Study (n=51)
Clinical Case (n=10)
|Target Registration Error for Artificial Landmark|
|Target Registration Error for a Vessel Branch Point|
a mean+standard deviation (mm)
b not measured
c measured by vessel branch point