A Comparative Statistical Error Analysis of Neuronavigation Systems in a Clinical Setting Abbasi H. MD PhD, Hariri S. CandMed, Martin D. MD, Kim D. MD, Shahidi R. PhD Department of Neurosurgery – Stanford University, 300 Pasteur Dr, Stanford CA, 94305 USA Abstract The use of neuronavigation (NN) in neurosurgery has become ubiquitous. A growing number of neurosurgeons are utilizing NN for a wide variety of purposes, including optimizing the surgical approach (macrosurgery) and locating small areas of interest (microsurgery). The goal of our team is to apply rapid advances in hardware and software technology to the field of NN, challenging and ultimately updating current NN assumptions. To identify possible areas in which new technology may improve the surgical applications of NN, we have assessed the accuracy of neuronavigational measurements in the Radionics™ and BrainLab™ systems. Using a phantom skull, we measured the accuracy of the navigational systems, taking a total of 2616 measurements. We found that, despite current NN tenets, the six marker count does not yield optimal accuracy in either system, and the spreaded marker setting yields best accuracy in both systems. Placing less markers around the region of interest (ROI) minimizes registration error, and active tracking does not necessarily increase accuracy. Comparing the two systems, we also found that the accuracy of NN machines differs both overall and in different axes. Introduction ▪ Experimental and methodological assumptions that have been retained in the use of NN over its development: 1. Using more markers increases accuracy. 2. Putting more markers around the area of interest increases accuracy. 3. Using 6 markers yields the most efficient accuracy. 4. Using active tracking [e.g. an active DRF (dynamic reference frame) and probe which emit infrared beams rather than passively reflecting them] increases accuracy. 5. Minimizing the distance from the marker or marker group to the area of interest increases accuracy. ▪ While rapid advances in hardware and software have emerged in the last few years, there have been only few attempts at challenging the old NN tenets and applying new technology to update these systems. ▪ To identify possible areas in which new technology may improve the surgical applications of NN, we conducted accuracy tests of NN measurements in two currently used systems: Radionics™ and BrainLab™. ▪ Questions we sought to address: 1. How many markers efficiently maximize accuracy? 2. What pattern of marker localization maximizes accuracy? 3. 4. 5. Are there significant accuracy differences between various marker arrangements in different systems? Are systems significantly different in their level of accuracy? Is accuracy different in the x-, y- and z-axes? Materials and Methods ▪ We obtained a standard plastic skull, removed the calvaria, and installed three Plexiglas square rods of different heights in each of the three anatomical fossae (anterior, middle and posterior). We used the edges of these rods as our targets. We installed a Plexiglas ball of known diameter on the phantom's sella turcica. (Fig. 1 and 2) ▪ Replacing the calvaria, we placed a total of 12 markers bilaterally on the exterior of the skull in the following regions: 6 frontal, 2 mastoid, 2 occipital and 2 high parietal. (Fig. 3) Fig. 3: Placement of the markers Marker count Local Setting 4 F1, F3, F4, F6 6 F1, F2, F3, F4, F5, F6, 8 (Radionics only) F1, F2, F3, F4, F5, F6, O1, O6 F1: Frontolateral right below calvaria line F3: Frontolateral left below calvaria line F5: Frontomedian above calvaria line O1: Above right mastoid process O3: High parietal right O5: Left occipital Spreaded Setting F1, F3, O1, O6 F1, F3, O1,O3, O4, O6 F1, F3, O1, O2, O3, O4, O5, O6 F2: Frontomedian below calvaria line F4: Frontolateral right above calvaria line F6: Frontolateral left above calvaria line O2: Right occipital O4: High parietal left O6: Above left mastoid process Fig. 4: Relation of the measurements on the monitor to the actual coordinates of the skull as illustrated in Pic. 2 Screen Coordinate Skull Coordinate Screen Coordinate Skull Coordinate Axial X +X Axial Y +Y Sagital X -Y Sagital Y +Z Coronal X +X Coronal Y +Z ▪ We performed a CT of the skull in 1.25 mm slices and sent the data over the network to the two NN machines evaluated in this study. The systems utilize different registration and tracking systems to localize the probe's tip in 3D: ▪ BrainLab VectorVision version 2.3: Passive registration & tracking: the DRF and probe reflect the IR signal shot out by emitters around the cameras. Local and spreaded marker settings with 4, 6, and 8 markers. ▪ Radionics OTS version 2.2: Active registration & tracking: the DRF and probe contain diodes emitting IR signals that are collected by 2 cameras. Local and spreaded marker settings with 4, 6, and 7 markers. ▪ When the probe tip was placed at the edge of a rod, the NN systems visualized the probe's position on their screens in the original axial plane of the CT scans and in the sagital and coronal planes reconstructed from the CT scans (Fig. 2). In each of the three cross-sections, we measured how far from the actual edge of the rod (x=0, y=0) the system was representing the probe tip (Fig. 4). 12 series of measurements taken, each series consisting of 218 separate measurements. Results ▪ Marker Count (Figs. 5 & 7) BrainLab: In the local marker setting, 4 or 7, but not 6 markers, maximize accuracy. In the spreaded marker setting, accuracy declines with additional markers. Radionics: In both the spreaded and local marker setting, as expected, error declined as the experimenter went from 4 to 6 to 8 markers. The greatest improvement in accuracy is obtained when increasing from 6 to 8 markers. However, this improvement in accuracy given more markers was not as great as expected. ▪ Proximity of Markers to Region of Interest (ROI) (Fig. 5) BrainLab: In the spreaded and local marker setting placing more markers around the ROI only somewhat maximizes accuracy. Radionics: In the local marker setting, placing more markers around the ROI maximizes accuracy. However, in the spreaded marker setting, placing fewer markers around the ROI maximizes accuracy (i.e. error was maximal in the frontal fossa and minimal in the occipital fossa). ▪ Active versus Passive Tracking. (Fig. 6) Active tracking does not necessarily increase accuracy, as shown by BrainLab's greater accuracy than the Radionics system. ▪ Spreaded versus Local Marker Setting. (Fig. 7) Except for the 7 and 8 marker counts, the spreaded marker setting yielded greater accuracy than the local marker setting. However, in our experimental setup, the 7 and 8 marker spreaded and local settings were extremely similar. ▪ Radionics versus BrainLab (Fig. 5 and 6) Accuracy of the NN machines differs both overall and in different axes. The overall accuracy of the BrainLab system is greater than the Radionics system. Discussion Overall Error = Registration Error + System Error Registration error = the error generated during the process of telling the NN system where each marker is located. System error = the mechanical, engineering, and software errors, such as machine or camera error. ▪ Having more markers can decrease system accuracy. However, at some point, the marginal increase in error associated with registering an additional marker surpasses the marginal decrease in system error due to more markers (Fig. 9). ▪ The registration error found in surgeries is probably greater than the registration error found in our phantom model because skin markers used for patients are applied to the skin while the skin markers used for the phantom were applied directly to the skull (Fig. 8). Conclusion This study scrutinized conventional assumptions of neuronavigation and found many of them to be invalid. The new NN tenets supported by our findings are: ▪ 4 or 8, but not 6, markers yields most efficient accuracy. We are aware of the counterintuitive nature of this finding, and our lab is currently investigating this result. Also, the movement of skin on the skull is not included in this study and may influence the clinical results. ▪ Placing fewer markers around the region of interest (ROI) decreases registration error at the ROI. ▪ Active tracking does not necessarily increase accuracy. ▪ The spreaded marker setting increases accuracy. ▪ Accuracy of the NN machines differs both overall and in different axes. References 1. Alberti, O; Dorward, NL; Kitchen, ND; Thomas, DG. National Hospital for Neurology and Neurosurgery, London, UK. “Neuronavigation--impact on operating time.” Stereotactic and Functional Neurosurgery. 1997, 68(1-4 Pt 1): 44-8. 2. Dorward, NL. “Neuronavigation--the surgeon's sextant [editorial]” British Journal of Neurosurgery. April 1997, 11(2): 101-3 3. Germano, IM; Villalobos, H; Silvers, A; Post, KD. Department of Neurosurgery, Mount Sinai Medical Center, New York. “Clinical use of the optical digitizer for intracranial neuronavigation.” Neurosurgery. Aug 1999, 45(2): 261-9, discussion 26970. 4. Gumprecht, HK; Widenka, DC; Lumenta, CB. Department of Neurosurgery, Academic Hospital München-Bogenhausen, Technical University of Munich, Germany. “BrainLab VectorVision Neuronavigation System: technology and clinical experiences in 131 cases.” Neurosurgery. January 1999, 44(1): 97-104, discussion 104-5. 5. Kaus, M; Steinmeier, R; Sporer, T; Ganslandt, O; Fahlbusch, R. Neurochirurgische Klinik, University of Erlangen-Nümberg, Germany. “Technical accuracy of a neuronavigation system measured with a high-precision mechanical micromanipulator.” Neurosurgery. December 1997. 41(6): 1431-6, discussion 1436-7. 6. Nabavi, A; Manthei, G; Blömer, U; Kumpf, L; Klinge, H; Mehdorn, HM. Neurochirurgische Klinik der Universität Kiel. “Neuronavigation. Computergestütztes Operieren in der Neurochirurgie. [Neuronavigation. Computer-assisted surgery in neurosurgery].” Radiologe. September 1995, 35(9): 573-7. 7. Ostertag, CB; Warnke, PC. Abteilung Stereotaktische Neurochirurgie, Neurozentrum, Klinikum, Universität Freiburg i. Br. “Neuronavigation. Computerassistierte Neurochirurgie. [Neuronavigation. Computer-assisted neurosurgery].” Nervenarzt. June 1999, 70(6): 517-21. 8. Schönherr, B; Gräwe, A; Meier, U. Klinik für Neurochirurgie, Unfallkrankenhaus Berlin. ”Qualitätssichernde Massnahmen bei neurochirurgischen Operationen. Erfahrungen mit intraoperativer Computertomographie und Neuronavigation. [Quality securing procedures in neurosurgical operations. Experiences with intraoperative computed tomography and neuronavigation]” Zeitschrift fur Arztliche Fortbildung und Qualitatssicherung. June 1999, 93(4): 273-80. 9. Schmieder, K; Hardenack, M; Harders, A. Department of Neurosurgery, Ruhr University Bochum, Germany. “Neuronavigation in daily clinical routine of a neurosurgical department.” Computer Aided Surgery. 1998, 3(4): 159-61. 10. Spetzger, U; Laborde, G; Gilsbach, JM. Department of Neurosurgery, Technical University RWTH-Aachen, Germany. “Frameless neuronavigation in modern neurosurgery.” Minimally Invasive Neurosurgery. December 1995, 38(4): 163-6. 11. Wirtz, CR; Knauth, M; Hassfeld, S; Tronnier, VM; Albert, FK; Bonsanto, MM; Kunze, S. Neurochirurgische Universitätsklinik, Ruprecht-Karls-Universität Heidelberg. “Neuronavigation--first experiences with three different commercially available systems.” Zentralblatt fur Neurochirurgie. 1998, 59(1): 14-22. 12. Wirtz, CR; Tronnier, VM; Bonsanto, MM; Hassfeld, S; Knauth, M; Kunze, S. Neurochirurgische Klinik und Poliklinik, Ruprecht-Karls-Universität, Heidelberg. “Neuronavigation. Methoden und Ausblick. [Neuronavigation. Methods and prospects].” Nervenarzt. December 1998, 69(12): 1029-36.