Available online at www.sciencedirect.com

Surgical Neurology 71 (2009) 640–648www.surgicalneurology-online.com

Technology

Image-guided robotic neurosurgery—an in vitro and in vivo pointaccuracy evaluation experimental study☆

Frank Chan, MEng,a,⁎ Irwan Kassim, MEng,a Charles Lo, MSc,a Chi Long Ho, MD, MChir,b

David Low, MRCS,b Beng Ti Ang, FRCS (SN),b Ivan Ng, FRCS (SN)baAdvanced Integrated Medical Systems, and bDepartment of Neurosurgery, National Neuroscience Institute, Singapore 308433, Singapore

Received 29 August 2007; accepted 12 June 2008

Abstract Background:We describe the development of a prototype neurosurgical robotic system called NISS.

Abbreviations: 3Dmachine; CT, computeNISS, Neuroscience In

☆ This work wasgrant (BMRC 04-1-33

⁎ Correspondingroscience Institute, Sfax: +65 6357 7137.

E-mail address: ch

0090-3019/$ – see frodoi:10.1016/j.surneu.2

The aim is to implement a robotic system capable of achieving accurate registration of roboticcoordinate systems based on CT images, so that it can be used in clinical application. This systemhas been refined with a better level of predictability, reliability, and robustness sufficient for animaltrial evaluation in stereotactic biopsy of brain lesions.Methods: Point accuracy evaluation of NISS began with an in vitro study. The in vitro roboticapplication accuracy result was 0.1 ± 0.05 mm and absolute needle-to-target deviation was 0.3 ±0.2 mm. An in vivo experiment approach of using steel balls of 1.56-mm-diameter as targets insidethe brain of an anaesthetized dog was used to evaluate the performance accuracy of NISSstereotactic probe placement. Five dogs underwent surgical insertion of steel balls into the brain,and the steel balls were served as targets to be reached by a core needle (1.56-mm-diameter). Theexperiment was carried out by precise manipulation of the needle to reach the steel ball usingframeless stereotactic localization principles.Results: A total of 9 needle results were collected from procedures involving 5 dogs. In the first5 procedures on 3 dogs, the results were less than 1.9 mm, with an average of 1.3 ± 0.5 mm. Theremaining 4 procedures on 2 dogs yielded results of less than 0.7 mm, with an average of 0.3 ±0.2 mm.Conclusion: The in vitro and in vivo studies represent the first approach toward evaluating targetingaccuracy of a robotic surgery system by using stereotactics biopsy application in a living subject.© 2009 Elsevier Inc. All rights reserved.

Keywords: Stereotactic brain biopsy; Robotic surgery; Image Guided Surgery; Stereotactic neurosurgery

1. Introduction

Advances in microsurgical techniques and minimallyinvasive surgery have resulted in a scale of surgery so small

, 3-dimensional; CMM, Coordinate measurementd tomography; FDA, Food and Drug Administration;stitute Surgical System; TRE, Target registration error.funded by the Biomedical Research Council research-19-349).author. Department of Neurosurgery, National Neu-ingapore 308433, Singapore. Tel.: +65 6357 7637;

ee_fatt_chan@nni.com.sg (F. Chan).

nt matter © 2009 Elsevier Inc. All rights reserved.008.06.008

that even the best human surgeons are limited by their naturaldexterity [9]. This is especially so given the delicateconsistency of the brain and the extremely narrow marginfor error in the field of neurosurgery. Recent advances insurgical navigation technology, digitized imaging modal-ities, increased computational power, and advancement inengineering have resulted in the incorporation of roboticsinto the field of surgery [11]. From the neurosurgical aspect,advances in image guidance technology when applied to aframeless navigation field can provide a robotic system withprecise spatial orientation in 3D space. With such technol-ogy, craniotomies and stereotactic brain biopsies are idealprocedures that can be performed by a robot.

641F. Chan et al. / Surgical Neurology 71 (2009) 640–648

The use of robots in surgery has several advantages.They have the potential to be more precise and accurate,thereby producing more reproducible outcomes withsmaller margins of error. In addition, they also have thesuperior 3D spatial accuracy and provide significantimprovement in manual dexterity compared to humans.The use of a robotic system could conceivably reduce timespent on a procedure, have greater accuracy, minimizeunwanted tremors or deviations from original surgical pathand minimize surgeon fatigue. The possibility of develop-ment of such systems could potentially allow neurosur-geons to transcend the restraints posed by current surgicaldemands and enter a new era in which robots maysupplement the surgical input.

Neurosurgical stereotactic applications require spatialaccuracy and precise targeting to minimize collateraldamage. Errors in placement and advancement of a surgicaltool can result in haemorrhage and severe neurologicalcomplications [5,17]. Stereotactic frame based brain biopsyhas been used for a number of decades. The accuracy of thesedevices have been reported to be below 1 mm [2,6].However, with the increasing use of frameless stereotacticbiopsies, controversy exists with regard to which system ismost accurate and cost-effective for biopsy of brain lesions[1,3,12,16]. For the past few years, 2 FDA clearedcommercial robotic systems like Pathfinder™ (producedby Armstrong Healthcare) and NeuroMate™ (produced byIntegrated Surgical Systems) that have emerged in themarket for stereotactic neurosurgery. Both systems are ableto automatically position themselves with respect to a targetdefined based on preoperative images. The accuracy for bothrobotic systems are, however, not superior to that ofconventional stereotactic biopsies which are below 1 mm.Their in vitro application accuracy result for PathFinder™ is2.7 mm [10], and that for NeuroMate™ is 1.95 ± 0.44 mm[7] for frameless registration. Their widespread clinical usehas been limited by the lack of strong evidence from well-designed clinical trials that is essential to determine theireffectiveness such as accuracy relative to current practices.The strongest evidence possible for assessing the accuracyof robotic precision manipulation is through in vitroexperiments using precise targets in phantom models aswell as in vivo experiments using precise targets implantedinto the brain of a living subject. In the literature, suchstudies are lacking.

2. Materials and methods

We report the development of a new prototype roboticsystem called NISS that is suitable for validating theapplication accuracy of the system. Similar to the firstprototype [14], it consisted of a parallel manipulator(Hexapod), an optical navigation system, a force torquesensor, a surgical drill, a pointer, a fixation unit, and aworkstation. The complete working procedure necessary to

perform robot assisted surgery was evaluated starting withdata acquisition, medical image data processing, surgicalplanning, and eventual conduction of a frameless stereo-tactic biopsy by integrating an intraoperative navigationsystem with a surgical robot.

2.1. Neuroscience Institute Surgical System

The robot is made up of 4 distinct modules as shown inFig. 1: NeuroPod (A), NeuroBase (B), NeuroVision (C), andNeuroPlan (D). NeuroPlan is a planning system that allowsthe surgeon to define targets, trajectories and critical pointsto avoid. NeuroVision, an intraoperative tracking system(Certus, Northern Digital [highest accurate tracking device])to track the patient and robot. NeuroPod is a parallel robot(Hexapod, Physical Instruments) that holds the surgical toolsand performs the surgery. Lastly, NeuroBase is a base robotthat positions the parallel robot into the optimum task-workspace. It has 5 degrees of freedom [http://en.wikipedia.org/wiki/Degrees_of_freedom_(engineering)] (2 motorizedand 3 manual) to position the hexapod within its optimumworkspace for robotic intervention. The current size of thebase robot is suitable for operating theatre environment. Ithas a base dimension of 350 × 600 mm and an adjustableheight of between 1180 and 1630 mm.

When planning the surgical procedure, the surgeon firstdefines the areas that are to be avoided in the planningsystem, which presents information obtained from CTimages in 3D format. The system subsequently andautomatically generates the required path using a proprietarypath-generation algorithm. The surgeon verifies the pathbefore the instructions are sent to the base robot (NeuroBase)which carries the hexapod (NeuroPod). The base robotpositions the hexapod to an optimal position with respect tothe patient before it is locked rigid. The drilling sequence isexecuted according to the preplanned path, and the entiresurgery is performed with the surgeon having both a virtualand a real-world view of the procedure.

2.2. In vitro study

Many targeting methods exist in assessing the accuracy ofstereotaxy devices in an in vitro study [13]. There are,however, not many methods that are suitable to assess theaccuracy of robots relevant for medical application. Hence,during the design of this project, a new measurementtechnique was developed to find the application accuracy ofNISS. The following 2 methods described were designed tomeasure the accuracy of robotic manipulation that would besimilar to a surgical procedure such as needle biopsy.

2.3. Experiment 1, robotic application accuracy

The experiment setup concept was similar to thePathfinder™ application accuracy study [10]. The overallapplication accuracy was mainly determined by the errorsin the calibration of the mounted tool of NISS and the TREin the image registration protocol [8]. The main focus was

Fig. 1. Overview of NISS. A: NeuroBase. B: NeuroPod. C: NeuroVision. D: NeuroPlan.

642 F. Chan et al. / Surgical Neurology 71 (2009) 640–648

to investigate the TRE in the overall application accuracyfor NISS.

A depth gauge (100 mm from tip to tool holder) wasmounted on NISS to represent the tool unit. A perspexphantom was fabricated with 12 alumina oxide spheres onstalks of varying lengths as shown in Fig. 2B, and thiswas sent for CT scanning. The depth gauge tip positionwas digitized with respect to the tool coordinate frame(digitized method is shown in Fig. 2A, and toolcoordinate frame is shown in Fig. 3A). The depthgauge tip was moved to the target positions by NISS,which resulted in a reading on the depth gauge, asillustrated in the Fig. 2B. Up to 3 spheres were used astargets and the others were selected to be fiducial points

Fig. 2. A: Digitizing the depth gauge tip as the tool tip. B

for registration purposes. The sphere coordinates wereobtained with the CMM [18].

2.4. Experiment 2, absolute needle tip to targetdistance measurement

The objective of this experiment was to measure theabsolute error of the needle tip to the target. Registrationmethod was similar to the method described in experiment 1(ie, using 3 spheres as the fiducial points for registration). Torepresent the target, a phantom with 16 aluminium-taperedpoles with flatted tip of 1.56 mm diameter was fabricated andsent for CT scanning (Fig. 3C). The locations of the flattedtips of the poles can be easily identified from the CT images(Fig. 3B). A total of 16 targets from CT images were selected

: The phantom as the targets and fiducial markers.

Fig. 3. A: Coordinate frames and transformations for NISS calibration with respect to tracked rigid bodies—base, tool, and phantom. B: CT image. C: Phantomwith the 16 aluminum pointed tip poles.

643F. Chan et al. / Surgical Neurology 71 (2009) 640–648

to be reached by the needle using NISS. The needle to beused (same for animal trials) consisted of an inner stainlesssteel core needle (1.56-mm diameter and 200 mm long) andan outer needle sheath (1.6-mm internal diameter with0.2 mm wall thickness and 100 mm long). The specificationis similar as the brain biopsy needle. Three important rigidbodies such as base, tool, and phantom to be tracked areindicated in Fig. 3A. The tool mounted onto the NISS holdsthe needle. With the tracked information, the distance of theneedle tip to the target can be determined.

2.5. In vivo study, live specimen setup and protocol

The studies were conducted with 5 dogs at the animallaboratory after the protocol was approved by the localinstitutional ethics committee. Selected dogs weights werein the range of 29 to 35 kg. At the start of the trial, eachanesthetized dog had gone through surgical insertion of1.56-mm-diameter steel balls into the brain to serve as thetargets to be reached by the biopsy needle robotically. CTscans (ie, preoperative CT images) of the dog's head was

Fig. 4. A: NISS in position with dog specimen in the operating thea

taken and loaded onto the computer. The CT images were512 × 512 pixels, with voxel size of 0.6 × 0.6 × 1 mm3.

The dogs were positioned in the lateral position. A specialportable bed termed NeuroBed was built which incorporateda self-adjustable, 3-point clamping mechanism for the dogs'heads. The primary material for the bed was made of acrylic,whereas the clamp was made of an aluminium alloy. It wasdesigned as such so that the area around the head region willnot result in significant artifacts during the CT scans (Fig. 4shows the animal set up for the in vivo studies.) The dogswere intubated throughout the duration of the procedure aswell as during the CT scans, and their vital signs were closelymonitored by a veterinarian. The respirator was adjusted toinflate the lungs at a rate of 20 breaths per minute, with apeak inspiratory volume of 450 mls. The dogs wererestrained to minimize respiratory movements. A test CTscan was carried out to demonstrate the quality of the imageswith (Fig. 5B) and without (Fig. 5A) NeuroBed.

The dog was transported to the animal operating room,and 2 20-mm-diameter burr holes were made. The

tre. B: NeuroBed with specimen being sent for CT scanning.

Fig. 5. A: Visible staggered contours of a fiducial sphere extracted from CTimages as compared to after restraining using NeuroBed (B).

644 F. Chan et al. / Surgical Neurology 71 (2009) 640–648

experiment was carried out by precise manipulation of 21.56-mm-diameter core needles to reach 2 steel ballsseparately while keeping the dog alive throughout theprocedure. In order to ensure that the needle will not movewhile the dog was transported to the CT scanning room,adhesive acrylic bone cement was used with the steel bracketto secure the needle onto the skull. After the needles weresecured, the needle was removed from the robot. A secondCT examination was carried out (ie, postoperative CTimages) to calculate the absolute distance of each of needletip to the target.

2.6. Acquisition of targets and results from CT images

To avoid voxel resolution that will affect the accuracy ofthe results, the steel ball sizes were selected to be just over 1voxel size. This enabled the calculation of the centre of acloud of points to represent the target position. The samevoxel size was applied to calculate the needle core tip.Ideally, there should be zero accuracy when the needle coretip surface touches the steel ball surface.

3. Results

3.1. In vitro experiments

In experiment 1, the average TRE for invasive markerswere in the range of 0.10 to 0.13 mm, and the overallaverage TRE obtained was 0.1112 mm. In Experiment 2,

Table 1Accuracy results for in vivo experiments with 5 animal trials

Trial Needleno.

Distance of needletip to target

Average and SD(mm)

Distance of needletip to target

Pre-Op CT (mm) Post-Op CT (mm)

1 1 1.207 1.3 ± 0.5 0.9 ± 0.6 1.4692 2 0.665 2.523

3 1.046 1.0783 4 1.905 2.318

5 1.640 2.1054 6 0.420 0.3 ± 0.2 1.338

7 0.202 0.9585 8 0.045 0.571

9 0.613 1.024

the average accuracy of the absolute needle-to-targetdeviation was 0.313 ± 0.181 mm. All angular errorsmeasured were below 0.5° with respect to the line fromentry point to the target point.

3.2. In-vivo experiments

The in vivo results of 5 animal trials with 9 needleinsertions are shown in Table 1 and Fig. 6 shows anillustration of fifth trial's second needle in 3D view of the 3different values (ie, preoperative, postoperative, and move-ment). The in vivo results showed that the needle tipdeviated from the targeted point on an average of 0.9 ±0.6 mm if target positions were selected from thepreoperative CT images and an average of 1.5 ± 0.7 mmif target positions were selected from postoperative CTimages. By comparing pre and postoperative CT images,the steel ball had moved an average of 1.0 ± 0.7 mm. Thegraph in Fig. 7 below shows the different results obtainedfor the in vitro and in vivo experiments.

4. Discussion

4.1. In vitro experiments (1 and 2)

The average result obtained for the distance of needle tipto target was below 0.5 mm based on the in vitroexperiments 1 and 2. The difference in values for accuracyin both in vitro experiments could be explained by thedifference between the short and rigid depth gauge, and theslenderness and long length of the biopsy needle thatincreased its flexibility. Consequently, the long needle alsoincreased the angular deviation and caused fluctuations inthe needle-to-target distance. The results highlighted thefact that robotic systems can perform with great accuracyand precision. However, this is an artificial mechanicalsetup, and the application has to be tested on a morerealistic in vivo setting.

4.2. In vivo experiment

The targets used in the in vitro experiments 1 and 2 wererigid objects that were fixed on a tracked phantom.

Average and SD(mm)

Steel ballmovement(mm)

Average and SD(mm)

Method ofsecuringbiopsy needle

1.9 ± 0.6 1.5 ± 0.7 0.427 0.9 ± 0.8 1.0 ± 0.7 Bone cement2.2310.4710.6780.469

1.0 ± 0.3 1.837 1.1 ± 0.6 Metal bracket1.2040.6870.642

Fig. 6. 3D view of the 3 values for second needle in fifth trial result.

645F. Chan et al. / Surgical Neurology 71 (2009) 640–648

However, this was a different situation when steel ballswere used as targets that were implanted into the region ofsurrounding soft tissue of brain where there is constantblood flow and pulsation of blood vessels. This phenom-enon of soft tissue deformation was addressed and analyzedin the in vivo experiments.

In our results for the in vivo experiments, the needle tiphad deviated from the targeted point on an average of 0.9 ±0.6 mm when target positions were selected from thepreoperative CT images for calculation. There was anincrease to an average of 1.5 ± 0.7 mm when targetpositions were selected from postoperative CT images, andthere was an average movement of 1.0 ± 0.7 mm of thesteel balls. The possible explanations for this movementcould be due to the brain shifts from cerebrospinal fluid

Fig. 7. The accuracy range analyzed fro

loss, the influence of gravity, vibration, or handling errors(ie, during the movement of the animal to and from theanimal operating room and CT scan room). Other possibleexplanations of the movement could be due to tissuedeformation and rupture of blood vessels during the needleinsertion. This was being noted when significant amount ofbleeding occurred during needle 2 (trial 2) and needle 6(trial 4) of the experiments.

Another possible handling error identified during the invivo experiments was that the biopsy needles were not wellsecured to the skull when the dogs were moved from theoperating suite to the CT scan rooms. Adhesive acrylicbone cement was used as shown in Fig. 8 to secure thebiopsy needles to the dog skull. Unfortunately, the cementhad broken off the skull surface during the initial trialswhen the dogs were transported (Fig. 8B). The process wasrectified by introducing a bracket to secure the needle morefirmly onto the skull as shown in Fig. 9 for the last 2 dogexperiments (needle numbers 6-9).

After the rectification was introduced, it was observedthat the average accuracy results have improved. Theaccuracy results taken from fourth and fifth dog trials witha further 4 needle insertions had an average accuracy of0.3 ± 0.2 mm based on preoperative CT calculations. Thishighlighted the fact that it is possible to meet the absolutedistance results of below 0.5 mm obtained in the in vitrostudies. This essentially means that the robotic settings canpotentially be implemented in the operating theatre setupwithout losing much of the accuracy.

In our in-vivo experiments, the use of a sphere steel ballmay also not replicate the actual scenario of a brain lesionsituated inside the brain parenchyma. The canine brain mayalso not be the ideal representation of the human brain.Nevertheless, based on the in vivo results obtained fromfourth and fifth dog trials, we feel that the in vivo accuracy

m the in vitro and in vivo results.

Fig. 8. X-ray image (A), 3D models (B), and animal setup (C) of the needling procedure for the initial animal trials showing shift in bone cement used to securethe biopsy needle.

646 F. Chan et al. / Surgical Neurology 71 (2009) 640–648

can potentially be achieved below the 1.0-mm mark infurther experiments.

4.3. Potential and future clinical utility

Currently, commercial Pathfinder™ and NeuroMate™are being used for stereotactic biopsy and functionalneurosurgical applications [4,15]. Similar to their concepton accurate localizing manipulation and tracking of theinstruments, the design of NISS could also be potentiallyapplied to functional neurosurgical applications such asdeep brain stimulation and pallidotomy for Parkinson's

Fig. 9. Animal setup (A) and (B) x-ray image of the fifth animal specim

disease. However, before NISS is adopted for other clinicalapplications, it is important to note that the assessmentprotocol for in vivo and in vitro should be redesignedaccording to the actual clinical application protocol asmuch as possible. For example, the instrument as well asthe target objects should change together with the newclinical application. The protocol will then be reviewed bycomparing the in vitro and in vivo results on whether thereare any changes in their positions during robotic manipula-tion in empty space and in deformable anatomicalstructures respectively.

en that had undergone 2 needle procedures with bracket in situ.

647F. Chan et al. / Surgical Neurology 71 (2009) 640–648

5. Conclusion

This article presents the in vitro and in vivo studies in asupervisory controlled robotic system designed by ourinstitution. It represents the first approach towardsevaluating targeting accuracy of a robotic surgery systemby using stereotactics biopsy application in a livingsubject. The conducted in vitro and in vivo studies havecovered spatial resolution of the overall system from CTimages space to robot's space, to phantom/animal's spacein measuring the distance between the planned target andthe positioned target. It is shown from the in vivo studythat handling and transportation of specimen are crucial.More animal trials are being scheduled for furtherevaluation of consistency of NISS.

Acknowledgments

We thank the late Professor Teo Ming Yeong for hisguidance in this project.

References

[1] Apuzzo ML, Sabshin JK. Computed tomographic guidance stereotaxisin the management of intracranial mass lesions. Neurosurgery 1983;12:277-85.

[2] Apuzzo ML, Chandrasoma PT, Cohen D, Zee CS, Zelman V.Computed imaging stereotaxy: experience and perspective related to500 procedures applied to brain masses. Neurosurgery 1987;20:930-7.

[3] Duquesnel J, Turjman F, Hermier M, Bascoulergue Y, Jouvet A,Gervesy G, et al. CT-guided needle biopsy of intracranial tumours:results in 118 consecutive patients. Acta Neurochir Suppl 1995;63:16-9.

[4] Eljamel MS. Robotic application in epilepsy surgery. Int J Med Robot2006;2(2):233-237.

[5] Field M, Witham TF, Flickinger JC, Kondziolka D, Lunsford LD.Comprehensive assessment of hemorrhage risks and outcomes afterstereotactic brain biopsy. J Neurosurg 2001;94:545-51.

[6] Horner NB, Potts DG. A comparison of CT-stereotaxic brain biopsytechniques. Invest Radiol 1984;19:367-73.

[7] Li QH, Zamorano L, Pandya A, Perez R, Gong J, Diaz F. Theapplication accuracy of the NeuroMate robot—a quantitative compar-ison with frameless infrared and frame based surgical localizationsystems. Computer Aided Surg 2002;7(2):90-8.

[8] Maurer Jr CR, Fitzpatrick JM, Wang MY, Galloway Jr RL, MaciunasRJ, Allen GS. Registration of head volume images using implantablefiducial markers. IEEE Trans Med Imaging 1997;16(4):447-62.

[9] McBeth PB, Louw DF, Rizun PR, Sutherland GR. Robotics inneurosurgery. Am J Surg 2004;188:68S-75S.

[10] Morgan PS, Carter T, Davis S, Sepehri A, Punt J, Byrne P, et al. Theapplication accuracy of the PathFinder neurosurgical robot. Proceed-ings of Computer Assisted Radiology & Surgery 17th InternationalCongress; 2003. p. 561-7.

[11] Narendra N, Todd P, Gene HB. Robotics and neurosurgery. Surg ClinNorth Am 2003;83:1339-50.

[12] Ostertag CB, Mennel HD, Kiessling M. Stereotactic biopsy of braintumors. Surg Neurol 1980;14:275-83.

[13] Quinones-Hinojosa A, Ware ML, Sanai N, McDermott MW.Assessment of image guided accuracy in a skull model: comparisonof frameless stereotaxy techniques vs. frame-based localization.J Neurooncol 2006;76:65-70.

[14] Sim C, Teo MY, Ng WS, Yeo TT. Development of a robotic systemfor skull-base surgery. World Congress in High-Tech Medicine; 2000.p. 165-71.

[15] Varma TR, Eldridge PR, Forster A, Fox S, Fletcher M, Steiger P,Littlechild P, Byrne P, Sinnott A, Tyler K, Flintham S. Use of theNeuroMate robot in a frameless mode for movement disorder surgery.Stereotact Funct Neurosurg 2003;80(1–4):132-5.

[16] Voges J, Schroder R, Treuer H, Pastyr O, Schlegel W, Lorenz WJ, et al.CT-guided and computer assisted stereotactic biopsy. Technique,results, indications. Acta Neurochir 1993;125:142-9.

[17] Weprin BE, Swift DM. Complications of ventricular shunts. TechNeurosurg 2002;7(3):224-42.

[18] Wiles AD, Thompson DG, Frantz DD. Accuracy assessment andinterpretation for optical tracking systems. Proc. SPIE 5367MedImaging 2004:421-32.

Commentary

Despite dramatic technological advances over the lastdecades, neurosurgical robotics is still in its infancy. Use ofsurgical robots, such as Da Vinci Surgical System (IntuitiveSurgical, Sunnyvale, Calif), has become common practice ingeneral surgery [2,3], urology [8], gynecology [1], vascularsurgery [6], and transplantology [4], but robotic applicationsin neurosurgery are still gaining only limited recognition. Onone hand, it is surprising because neurosurgeons invented theidea of stereotaxis and routinely deal with the highest degreeof precision and accuracy. On the other hand, the very sameidea of extremely high value of the brain prevents us fromletting “the machine” operate on our behalf.

This will definitely change, and the article by Chan et alsupports the idea completely. Development of simple,precise, and reliable robots is now a reality. Some havealready been approved by the Food and Drug Administra-tion for marketing, whereas others are getting close to suchapproval. The principle behind this technology is simple,and the advantages are obvious: the accuracy of robotictargeting is higher than that of the human hand; a roboticarm does not shake; each approach may be preplanned in avirtual space prior to the actual movement; there is nofatigue; and movements may be synchronized with thepatient in cases of frameless navigation, similar to roboticradiosurgery with CyberKnife (Accuracy, Sunnyvale, Calif)[5], or coupled through rigid attachment to a stereotacticframe like NeuroMate (Integrated Surgical Systems, Davis,Calif) [7]. As one can see from the Chan et al article,accuracy of the device is around 1 mm—similar to theacquisition threshold of most modern imaging devices—and this is only the beginning!

Obviously, the results presented here still need to beextensively validated. Clinical reality is far from laboratoryexperiments. Human surgical targets—be it deep tumors orfunctional nuclei—are not steel balls, and the radiographiccoordinates may indeed differ from anatomical or physiolo-gical “sweet spots,” and that's why the robots of future willbe equipped with “smart guidance” systems that will include