Computer Aided Surgery 4:45? 49 (1999) Clinical Paper Clinical Evaluation of Multimodality Registration in Frameless Stereotaxy Hunaldo Villalobos, m.d., and Isabelle M. Germano, m.d. Department of Neurosurgery, Mount Sinai Medical Center, 1 Gustave Levy Place, New York, New York ABSTRACT Computer-assisted frameless neurosurgery bases its accuracy and reliability on registration. The aim of this prospective study was to compare the clinical accuracy of different registration techniques used for computer-assisted frameless neurosurgery. Ninety-eight registrations in 44 patients were used to compare the clinical accuracy of self-adhesive marker (MR) and facial landmark (FR) registrations used alone or in conjunction with surface-fit registration (MR/SR and FR/SR, respectively) for cranial neurosurgery. The computer estimated error (CEE) of each registration was compared to the real error (RE). This was obtained by holding the frameless pointer at the center of three different markers and measuring the distance from the real-time representation on the computer three-planar images to the center of the marker on the screen. The most accurate registration was obtained using MR; the RE of MR was 1.6 6 0.1 mm compared to 3.4 6 0.4 mm for FR. Although the smallest CEE error was obtained using MR/SR, this was not sustained by the RE. Furthermore, the RE of FR/SR was significantly larger than the CEE (Student t test, p < .001). This study corroborates previous results showing that, in the clinical setting, self-adhesive marker registration is more accurate than facial landmark registration. Furthermore, although surface-fit registration can be used in conjunction with self-adhesive marker registration, this does not improve the degree of real accuracy for cranial registration. Comp Aid Surg 4:45?49 (1999). �99 Wiley-Liss, Inc. Key words: frameless stereotaxy, computer-assisted, cranial, neuronavigation, accuracy, surface-fit, point registration, facial landmarks registration INTRODUCTION During the past decade, several centers have developed computer-assisted frameless stereotactic systems to be applied in neurological surgery.9,18,22?24 These systems are integrated by a three-dimensional (3D) digitizer connected to a computer workstation that displays the position of the instruments in a set of reformatted images. Image-guided computer-assisted surgery has become a tool of inestimable value for neurosurgeons.4 ? 8,23 Registration is the cornerstone of frameless stereotaxy. The majority of systems use different types of external reference points or fiducials to register the patient?s head in the surgical field to the data set of images in the computer. This process involves preoperative acquisition of images with fiducial markers in place. Intraoperatively, these are overlapped to the patient?s own anatomy by entering into the computer the same point on the Received September 11, 1998; accepted March 11, 1999. Address correspondence/reprint requests to: I. M. Germano. E-mail: firstname.lastname@example.org. �99 Wiley-Liss, Inc. 46 Villalobos and Germano: Registration in Frameless Stereotaxy images and on the patient anatomy. The fiducials can consist of self-adhesive markers, implantable soft-tissue/bone screws, or facial landmarks.15,16 Previous studies have suggested that self-adhesive markers provide a more accurate registration than facial landmarks. The former, however, require trained personnel to place them appropriately and a new set of preoperative images that is often redundant. The purpose of this prospective study was to compare the clinical accuracy of frameless registration using self-adhesive markers (MR) with facial landmarks (FR) used alone or in conjunction with surface-fit registration (MR/SR and FR/SR, respectively). This study was performed with the StealthStation frameless stereotaxy system (Sofamor-Danek, Broomfield, CO).4 MATERIALS AND METHODS During the period September 1997 through February 1998, 44 consecutive patients undergoing a cranial frameless neurosurgical procedure performed by the same surgeon (IMG) were entered in this prospective study. Preoperatively, 8 ?10 selfadhesive fiducial markers (Medical Products, Baltimore, MD) were applied to the patient?s scalp in the standard noncollinear fashion minutes before a magnetic resonance image (MRI) of the brain was obtained. Prior to applying the markers, the hair was minimally shaved and Benzoin tincture was applied to the skin to improve adhesion. Each fiducial marker was then marked in the center with an indelible marker pen and a circumferential head wrap was placed around the patient?s head. The MRI parameters were as follows: TR 5 600; TE 5 8; thickness 5 2 mm; interspace 5 0 mm; NEX 5 2; FOV 5 34; matrix 5 256 3 256. In all cases, the patient was brought to the operating room immediately after the MRI. After adequate anesthesia had been achieved, the patient?s head was fixed in the Mayfield head holder. Care was taken to place the Mayfield pins at least 2 cm away from the skin markers. If one of the pins was closer than 2 cm to a skin marker, this was not used for registration. All registrations were performed by the same person (HV). Clinical correlation accuracy was determined by comparing four registration techniques as described below. Clinical Experience: Registration Techniques Marker Registration (MR) The self-adhesive markers were numbered on the computer screen in counterclockwise fashion be- ginning with the right frontal one. The same marker chosen on the computer screen was then identified on the patient?s head using the blunt-tip probe. When all markers had been entered, the registration transformation and error were calculated by the computer. For point registration, this error is known as computer estimated error (CEE), which is a mean fiducial error. A CEE , 4 mm was accepted. To measure the true accuracy of the registration between the image and physical space, the frameless pointer was then placed in the center of three different markers. The real-time projections of the pointer appeared on the computer screen as crosshairs on the three-planar images. The distance from the real-time projection of the pointer on the images to the center of the marker was then digitized in all three planes. The longest distance for each marker in all three planes was recorded in each patient. The average of these measurements was called real error (RE). Surface Registration after Marker Registration (MR/SR) In all cases where MR was completed with a CEE of ,4 mm and a predicted accuracy (PA)3 at 10 cm of ,6.0 mm, surface registration (SR) was performed. For surface registration, approximately 40 points along the surface of a patient?s head were digitized and stored in the computer until a measure of registration uniqueness known as geometric constraint was satisfied. The location of these points was randomly chosen and had to be within the volume that was scanned. The points were entered randomly, touching the skin very lightly. They were primarily selected from the fronto-facial area, as well as the calvarium and auriculo-temporal area. The computer reconstructed a geometric model of the surface of the patient?s head based on the acquired images. The geometric model was then aligned to the $40 points by the computer, by finding a transformation which minimized the distance between them. In surface registration, an error measure based on the distances between each of the digitized points and the corresponding geometric surface model can be used to estimate the quality of the registration. After registration was completed, RE was calculated as described above. Facial Landmark Registration (FR) For this registration method, anatomical landmarks were used instead of self-adhesive markers. Any distinct anatomical feature can be used for this registration process. Typically, we used the following points in a counterclockwise fashion: the right Villalobos and Germano: Registration in Frameless Stereotaxy 47 pinna, the right external cantus, the nasion, a midline frontal fiducial marker, the left external cantus, and the left tragus (Fig. 1). Other facial landmarks were selected when the above could not be used or were outside the scanned field. The registration was performed on the patient using the same sequence. A CEE of ,4 mm was accepted, and the RE between the image and physical space was calculated as described above. Table 1. Comparison of Mean Computer Estimated Error (CEE) and Real Error (RE) in Clinical Setting Using Different Registration Techniques* Surface Registration for Facial Landmarks Registration (MR/SR) * Point registration with self-adhesive markers (MR), point registration using self-adhesive markers followed by surface-fit algorithm (MR/SR), point registration using facial landmarks (FR), and point registration using facial landmarks followed by surface-fit algorithm (FR/SR). In all cases where FR was completed with a CEE of ,5 mm and a PA at 10 cm of #6.0 mm, SR was performed as described above. Laboratory Experience: Registration Techniques Registration technique CEE (mm) MR MR/SR FR FR/SR 2.7 1.1 3.2 1.2 6 6 6 6 0.2 0.1 0.2 0.1 RE (mm) 1.6 2.9 3.4 4.7 6 6 6 6 0.1 0.3 0.4 0.7 The object was then scanned using the identical MRI protocol used for the clinical study. Cranial Phantom Statistical Analysis Five phantoms were used for evaluation of the registration techniques under laboratory conditions. To mimic facial landmarks, we used three hardskin fruits to simulate facial registration, and these were sculpted on one side for that purpose. In addition, 10 self-adhesive markers were used to simulate point registration (MR). We used similar MRI imaging and registration techniques and protocols as described above in Clinical Experience. Computer estimated error and RE within registration modality were compared using Student?s unpaired t test. Analysis of variance (ANOVA) with post hoc correction was used to compare CEE and REE among registrations. A probability (p) , .05 was considered significant. Fig. 1. StealthStation computer screen showing the localization of the anatomical landmarks used for facial registration. RESULTS Frameless stereotactic cranial procedures were performed in 44 patients aged 51 6 2.7 years (range 5? 86). Three children aged 8 6 3 years were included in the study. The procedures were performed on the following locations: temporal (12), parietal (10), frontal (9), basal ganglia (7), and posterior fossa (6). Forty-four registrations and relative errors using MR were compared with 26 registrations using point registration followed by MR/SR, 28 using FR, and 16 using FR/SR. Table 1 summarizes the CEE and RE for each registration type. The most accurate registration, i.e., the smallest RE, was obtained using MR (RE: 1.6 6 0.1 mm; ANOVA, p , .0001). Although the absolute smallest error was the CEE found using MR/SR (1.1 6 0.1 mm), the RE showed a higher value (2.9 6 0.3 mm). The mean number of registrations to reach a CEE , 4 mm (MR) or , 5 mm (FR) and a PA at 10 cm , 6.0 mm was 2.8 6 0.3 using self-adhesive markers and 5.7 6 0.1 mm using facial landmarks. The approximate time for registration was 5 min with self-adhesive markers, 10 min with facial landmarks, and 30 min when surface registration was used. Data pertinent to the children included in this study are reported in Table 2. In this subgroup, no differences were found comparing the four modal- 48 Villalobos and Germano: Registration in Frameless Stereotaxy Table 2. Comparison of Mean Computer Estimated Error (CEE) and Real Error (RE) in Clinical Setting in the Pediatric Population Registration technique CEE (mm) RE (mm) MR/SR FR/SR 1.9 6 0.1 0.9 6 0.1 2.2 6 0.1 1.4 6 0.2 Abbreviations as in Table 1. ities. Table 3 summarizes our laboratory experience using the surface-fit registration on a rigid phantom. Although no statistical differences in accuracy were found, the smallest RE was found with MR. DISCUSSION The accuracy of the StealthStation in the laboratory with optimal parameters has been reported to have a maximum error of 0.55 6 0.29 mm.13 In the clinical setting, however, the registration accuracy may depend on several factors including the registration techniques used. The two main registration techniques used in neuronavigation are point and surface-fit registration.14 For point registration, self-adhesive markers, facial landmarks, or bone-implanted fiducials have been used. In our practice, we do not use implanted fiducials, as these are more invasive than the two former types. Previous studies using a different system showed that, compared with anatomical landmarks followed by surface registration, fiducial registration had greater accuracy.7,21 The higher error found using facial landmarks can be partially attributed to the fact that the points used for the registration are fairly coplanar.25 In our study, we found that the smallest RE was obtained using point registration. This corroborated the previous studies mentioned above. Self-adhesive markers are less appealing than facial landmarks because they have to be applied prior to surgery by trained personnel and in most cases require repetition of recent radiographic images.19 Originally developed for registration of multiple 3D image data sets,17 surface-fit registration is based on an algorithm of the least-squares fit between head and hat. Surface registration techniques have been compared with the landmark surface method and found to have a similar accuracy within 2 mm.17 Surface-fit registration was developed for different settings of imaging studies to improve registration error.12,18,19 This method consists of digitizing the surface of the skin, which is then correlated with a distance map created by the computer. This technique is used for multimodality imaging coregistration with acceptable accuracy.10,12,16,20 The use of surface-fit registration to optimize a registration done using facial landmarks is appealing. If proven to be accurate, this modality would obviate the need for applying adhesive fiducials and duplicating preoperative imaging studies. Our clinical experience showed, however, that facial registration followed by surface-fit registration is not as accurate as marker registration alone: RE 5 1.6 6 0.1 and 4.7 6 0.7 mm, respectively (p , .001). Furthermore, our study showed that, although the CEE using facial landmark registration followed by surface-fit is small (1.2 6 0.1 mm), this does not correspond to the RE (4.7 6 0.7 mm). Thus, this software must be used with great care for cranial registration. It should be noted, however, that using a rigid model in the laboratory, the CEE and RE for surface-fit registration were small and not significantly different. We believe that this discrepancy is due to the skin flexibility and elasticity that may interfere with the surface algorithm. Furthermore, pressure deformity occurring during the scan and lack of turgor with presence of wrinkles in the adult all may contribute to the decreased accuracy using the surface-fit registration for cranial procedures. In our study, in the children in whom there was good turgor and lack of wrinkles, the accuracy of the surface-fit program was significantly better than the results from the adult population. Furthermore, no difference was found between CEE and RE using the surface-fit algorithm. Similar findings were reported using a potentiometer arm system.1 To overcome the error secondary to skin depression (known as ?shrinking sphere?2) during acquisition of points with the manual digitizer, Henderson and Bucholz proposed an ergonomic depth-sensing laser instrument that digitized the forehead contour in 1 min.11 This and other modalities should be further investigated to ensure greater clinical accuracy. In conclusion, this study showed that, in the clinical setting with an adult population, the most accurate registration is achieved using self-adhe- Table 3. Comparison of Computer Estimated Error (CEE) and Real Error (RE) in Laboratory Using Rigid Phantoms Registration technique CEE (mm) RE (mm) MR MR/SR 1.9 6 0.1 1.1 6 0.2 0.9 6 0.2 1.8 6 0.4 Abbreviations as in Table 1. Villalobos and Germano: Registration in Frameless Stereotaxy sive markers. Further prospective studies are needed to determine the relationship between registration accuracy and accuracy at the brain target. 14. REFERENCES 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. Drake JM, Prudencio J, Stephanie H, Rutka JT, Hoffman HJ, Humphreys RP. Frameless stereotaxy in children. Pediatr Neurosurg 1994;20:152?159. Evans AC, Peters TM, Collins DL, Neelin P, Gabe C. Image registration based on discrete anatomic structures. In: Maciunas RJ, editor. Interactive image guided neurosurgery. Park Ridge, IL: AANS Publications Committee; 1993. p 47? 62. Fitzpatrick JM, West JB, Maurer CR. Predicting error in rigid-body point-based registration. IEEE Trans Medical Imaging 1998;17:694 ?702. Germano I. The NeuroStation system for imageguided, frameless stereotaxy. Neurosurgery 1995;37: 348 ?350. Germano IM, Heilbrun P. Interactive computer-assisted image-guided technology as an adjuvant to neurosurgery for movement disorders. In: Germano IM, editor. Neurosurgical treatment of movement disorders. AANS Publications Committee; 1998. p 219 ? 226. Germano IM, Queenan J. Clinical experience with intracranial brain needle biopsy using frameless surgical navigation. Comput Aid Surg 1998;3:33?39. Golfinos JG, Fitzpatrick BC, Smith LR, Speltzer RF. Clinical use of a frameless stereotactic arm: results of 325 cases. J Neurosurg 1995;83:197?205. Gomez H, Barnett G, Estes M, Palmer J, Magdinec M. Stereotactic and computer-assisted neurosurgery at the Cleveland Clinic: review of 501 consecutive cases. Cleveland Clin J Med 1993;60:399 ? 410. Heilbrun MP, McDonald P, Wiker C, Koehler S, William P. Stereotactic localization and guidance using a machine vision technique. Stereotact Funct Neurosurg 1992;58:94 ?98. Hemler PF, Sumanaweera TS, van den Elsen PA, Napel S, Adler J. A versatile system for multimodality image fusion. J Image Guid Surg 1995;1:35? 45. Henderson JM, Smith KR, Bucholz RD. An accurate and ergonomic method of registration for imageguided neurosurgery. Comp Med Imag Graphics 1994;18:273?277. Hill DLG, Hawkes DJ, Gleeson MJ, Cox TCS, Strong AJ, Wong WL, Ruff CF, Kitchen ND, Thomas DGT, Sofat A, Crossman JE, Studholme C, Gandhe AJ, Green SEM, Robinson GP. Accurate frameless registration of MR and CT Images of the head: applications in planning surgery and radiation therapy. Radiology 1994;191:447? 454. Kaus M, Steinmeier R, Sporer T, Ganslandt 0, Fahl- 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 49 busch R. Technical accuracy of neuronavigation system measured with high-precision mechanical micromanipulator. Neurosurgery 1997;41:1431?1436. Maciunas RJ, Fitzpatrick JM, Galloway RL, Allen GS. Beyond stereotaxy: extreme levels of application accuracy are provided by implantable fiducial markers for interactive image-guided neurosurgery. In: Maciunas RJ, editor. Interactive image guided neurosurgery. Park Ridge, IL: AANS Publications Committee; 1993. p 259 ?270. Maurer CR, Fitzpatrick JM. A review of medical image registration. In: Maciunas RJ, editor. Interactive image guided neurosurgery. Park Ridge, IL: AANS Publications Committee; 1993. p 17? 44. Pelizzari CA, Chen GTY, Spelbring DR. Accurate three-dimensional registration of CT, PET and/or MR images of the brain. J Comp Assist Tomogr 1989;13: 20 ?27. Pelizzari CA, Levin DN, Chen GTY, Chen CT, Cooper MD. Image registration based on anatomical surface matching. In: Maciunas RJ, editor. Interactive image guided neurosurgery. Park Ridge, IL: AANS Publications Committee; 1993. p 63? 80. Roberts DW, Strobehn JW, Hatch JF, Murray W, Kettenberger H. A frameless stereotaxic integration of computerized tomographic imaging and operating microscope. J Neurosurg 1986;65:545?549. Ryan MJ, Erickson RK, Levin DN, Pelizzari CA, Macdonald RL, Dohrmann GJ. Frameless stereotaxy with real-time tracking of patient head movement and retrospective patient-image registration. J Neurosurg 1996;85:287?292. Simon DA, Hebert M, Kanade T. Techniques for fast accurate intrasurgical registration. J Image Guid Surg 1995;1:17?29. Sipos EP, Tebo S, Zinreich SJ, Donlin ML, Brem H. In vivo accuracy testing and clinical experience with the ISG Viewing Wand. Neurosurgery 1996;39:134 ? 202. Smith K, Frank KJ, Bucholz RD. The NeuroStation: a highly accurate, minimally invasive solution to frameless stereotactic neurosurgery. Comp Med Imag Graphics 1994;18:247?56. Watanabe E, Mayangi Y, Kosugi Y. Open surgery assisted by the Neuronavigator, a stereotactic articulated, sensitive arm. Neurosurgery 1991;28:792? 800. Zamorano LJ, Nolte L, Kadi M, Jiang Z. Interactive intraoperative localization using an infrared-based system. Stereotact Funct Neurosurg 1994;63:84 ? 88. Zinreich SJ, Tebo SA, Long D, Brem H, Mattox DE, Loury ME, Vander Kolk CA, Koch WM, Kennedy DW, Bryan RN. Frameless stereotaxic integration of CT imaging data: accuracy and initial applications. Radiology 1993;188:735?742.