Biomedical Paper

System for Robotically Assisted PercutaneousProcedures with Computed Tomography Guidance

Ken Masamune, Ph.D., Gabor Fichtinger, Ph.D., Alexandru Patriciu, M.Sc., Robert C. Susil, B.Sc.,Russell H. Taylor, Ph.D., Louis R. Kavoussi, M.D., James H. Anderson, M.D.,Ichiro Sakuma, Ph.D., Takeyoshi Dohi, Ph.D., and Dan Stoianovici, Ph.D.

Graduate School of Engineering, The University of Tokyo (K.M., I.S., T.D.), and EngineeringResearch Center (G.F., A.P., R.C.S., R.H.T., J.H.A., D.S.), Brady Urological Institute (L.R.K., D.S.),

and Department of Radiology (R.C.S., J.H.A.), Johns Hopkins University, Baltimore, Maryland

ABSTRACT We present the prototype of an image-guided robotic system for accurate and con-sistent placement of percutaneous needles in soft-tissue targets under CT guidance inside the gantryof a CT scanner. The couch-mounted system consists of a seven-degrees-of-freedom passivemounting arm, a remote center-of-motion robot, and a motorized needle-insertion device. Single-image-based coregistration of the robot and image space is achieved by stereotactic localization usinga miniature version of the BRW head frame built into the radiolucent needle driver. The surgeonplans and controls the intervention in the scanner room on a desktop computer that receives DICOMimages from the scanner. The system does not need calibration, employs pure image-based regis-tration, and does not utilize any vendor-specific hardware or software features. In the open air, wherethere is no needle–tissue interaction, we systematically achieved an accuracy better than 1 mm inhitting targets at 5–8 cm from the fulcrum point. In the phantom, the orientation accuracy was 0.6°,and the distance between the needle tip and the target was 1.04 mm. Experiments indicated that thisrobotic system is suitable for a variety of percutaneous clinical applications. Comp Aid Surg 6:370–383(2001). ©2002 Wiley-Liss, Inc.

Key words: image-guided surgery, percutaneous needle placement, CT guidance, surgical robotics

INTRODUCTIONMotivation

Recent advances in medical imaging have broughtminimally invasive image-guided percutaneousbiopsy and local therapies to public attention.1,2

Intraoperative radiological imaging has become

more accurate, faster, affordable, and less haz-ardous to both patients and surgeons. Computedtomography (CT) provides good tissue differen-tiation from even a single slice acquisition, andhas proven to be an excellent image guidancemodality for percutaneous tumor biopsy and

Received May 24, 2000; accepted November 5, 2001.

Address correspondence/reprint requests to: Gabor Fichtinger, Ph.D., Center for Computer-Integrated Surgical Systems andTechnology, New Engineering Bldg, Room #315, Johns Hopkins University, 3400 North Charles Street, Baltimore, MD21218-2682; Telephone: 410-516-4057; Fax: 410-516-5553; E-mail: gabor@cs.jhu.edu; Ken Masamune, Ph.D., Departmentof Bio-Technology, College of Science and Engineering, Tokyo Denki University, Ishizaka, Hatoyama, Hiki, Saitama, Japan350-0394; Telephone: �81-492-96-2911, ext.5112; Fax: �81-492-96-5162; E-mail: masa@b.dendai.ac.jp.

Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/igs.10024

Computer Aided Surgery 6:370–383 (2001)

©2002 Wiley-Liss, Inc.

drainage,3–5 as well as for neurological pain man-agement.6 –10

Our goal was to meet the ultimate challengeof these procedures and provide seamless integra-tion of intraoperative CT imaging with precise,reliable, and affordable aiming and delivery of per-cutaneous surgical devices.

One could ask why we pursued CT guidanceinstead of fluoroscopic or ultrasound navigation.For nerve blocks and facet joint injections, fluoros-copy and CT are widely accepted image guidancemodalities. In this case, selecting CT over fluoros-copy was primarily an engineering decision, ratherthan an expression of strong clinical preference.We believe that our solution may have a greaterimpact on in-CT procedures than on fluoroscopy-based applications. The presented robotic system(primarily on account of its compact size, highdexterity, light weight, and image guidance tech-nique) represents novel and enabling technologyfor in-CT procedures. At the same time, variants ofthe presented system are also being explored forfluoroscopy- and ultrasound-guided procedures inour laboratories. One of our fundamental goals atthe Johns Hopkins University is to develop modu-lar and factorable image-guided robotic percutane-ous therapy systems that, to a large extent, areinvariant to the imaging modality with which theyare deployed. Although these systems apply grosslydifferent image guidance and calibration tech-niques, the body of the robotic system and the flowof information are very similar across modalitiesand applications.

Prior WorkOver the last 2 decades, extensive work has beendone in biopsy and treatment of intracranial lesionsusing invasive stereotactic head frames.11–18 Earlyexperiences with CT-guided robots were also asso-ciated with invasive head frames.19–21,46

For procedures that require access to the ab-domen or spine, full-body stereotactic frames weredeveloped.22–26 Devices like Elekta’s StereotacticBody Frame� or MedTEC’s BodyFIX� were de-signed for fractionated stereotactic radiotherapyand never caught on in interventional procedures.These devices typically include some form of cra-dle with a mold or vacuum bag that may be difficultto apply in the case of immobile patients. The cradleand fiducials can also interfere with the treatmentsite, leaving insufficient room for the interventioninside the gantry. Economically, the devices wereespecially unpromising for simple procedures likenerve blocks or facet joint injections.

Handheld manual devices27–30 provide valu-able assistance, but do not solve the ultimate prob-lem of the strong coupling between the image spaceand the surgical device. The efficacy of these pre-dominantly manual approaches depends primarilyon the surgeon’s hand–eye coordination and abilityto interpret visual feedback. Dohi, Masamune, andcolleagues presented a couch-mounted isocentricneedle-insertion manipulator that acts inside the CTgantry31,46 and the MRI gantry.47 This system oc-cupies the surface of the couch and reaches into thefield through the far end of the gantry, and istherefore suitable only for intracranial procedures.Loser proposed a remote center-of-motion roboticarm manipulated through visual servoing underCT-fluoroscopy.32 Shortcomings of this approachinclude the need for a CT-fluoroscopy option on thescanner and controlled motion of the CT couch.Neither of these features is widely available, leav-ing the system unsuitable for most CT scanners.

A few commercial robotic needle-insertionsystems, such as the NeuroMate robot from Inte-grated Surgical Systems, CA, also exist. Althoughthis robot has been cleared by FDA for stereotacticneedle punctures, it does not lend itself well toin-scanner applications due to its relatively largesize and heavy build.

Stoianovici et al. developed a compact anddexterous remote center-of-motion robot in con-junction with a radiolucent needle driver.33 Thesystem was used for percutaneous access to thekidney under joystick control with C-arm fluoros-copy. This system has proven to be an excellentpercutaneous robotic aid, but did not provide com-puterized remote control, which is necessary forcomputer-aided path planning and execution.

Susil et al.34 proposed a downsized version ofthe Brown-Roberts-Wells (BRW) head frame forsingle-slice-based registration of manipulators toimage space inside a CT scanner, and providedvaluable robustness and sensitivity analysis of theregistration method.

Present Contribution

The system we present combines the favorableaspects of the work of Stoianovici and Susil, andadds several key features.

Most importantly, a central computer pro-vides intraoperative image processing, motionplanning, remote actuation, and control of the ro-botic components. The interactive software com-pletes an intraoperative control loop, thus imple-menting a simplified variant of the surgical CAD/

Masamune et al.: Robotically Assisted Procedures with CT Guidance 371

CAM paradigm promoted by Taylor et al.36 Thecentral computer also enables the gathering of com-plex intraoperative information. Postoperationalprocessing of these data is expected to becomevaluable for outcome analysis and rehabilitationplanning, as well as for the performance evaluationof the engineered system.

Our system applies purely image-based reg-istration between the robot and the image space byintroducing a modified Susil frame. The registra-tion device is built with the radiolucent needledriver as an integral component. (For details, seesection entitled Novel End-Effector.) The com-bined device offers two unique features: (1) Therobotic insertion system does not need to be cali-brated before it is used, the dimensions of thedevice being known at the time of manufacture. (2)Registration and targeting requires only a singleimage slice. Both these features promote lowerradiation exposure and a shorter procedure.

The system does not depend on any vendor-specific hardware or software feature, and is de-ployable on any scanner that has a DICOM inter-face.

Clinical SignificanceThe proposed system has potential use in a widerange of percutaneous interventions. Figure 1shows the seamless transition from the currentphantom setup to applications in abdominal, pros-tate, and spine procedures. As a “flagship applica-tion,” we selected CT-guided percutaneous painmanagement, initially nerve blocks and facet jointinjections in the lumbar spine.

Spinal disorders are undoubtedly the fastestgrowing musculoskeletal subspecialty, consumingan estimated $120 billion dollars for direct andindirect costs. The acceleration in the number ofnew cases treated annually reflects the aging of oursociety, as well as the dramatic transformation ofthe traditional workplace into offices that imposeextreme stress on the skeletal system, primarily onthe spine. It is estimated that over 70% of ourpopulation experiences significant low-back pain,with 10% reporting sciatica and 1% requiring long-term conservative and/or surgical intervention.Combined with osteoporosis, these problems areexpected to incur an estimated cost of $200 billiondollars within the next 20 years.

Minimally invasive percutaneous pain man-agement in the spinal region has attracted a lot ofinterest lately. CT-guided nerve blocks and facetjoint injections have proven to be safe and effectivemethods for alleviating severe pain and providinglongstanding relief for patients regardless of age orsex.7–10

From a technical point of view, due to therelatively low complexity of the procedure, nerveblocks and facet joint injections are ideally suitedfor robotic assistance. Typically, these proceduresrequire a single puncture with a thin needle acrossreasonably superficial soft tissue, and the use of asingle CT image. The workflow of a manual pro-cedure is practically identical to the steps followedby our robotic system. This parallelism offers aunique opportunity for gradual transition from afully manual procedure to a fully robotic interven-tion.

However simple and easy these proceduresmay look, there is a great need for precise andconsistent aiming and delivery of the needle. Thelongevity of pain relief is thought to be associatedwith the spatial accuracy of needle placement. Eventhese single-puncture procedures present large vari-ability from surgeon to surgeon. For example, thelength of the procedure varies between 15 and 40min, depending on the experience of the interven-tionalist. Longer times are usually associated withmisplacement of the needle, often on several at-tempts, before its correct position is confirmed.Misplacements also include overdriving the needle,which causes sharp pain to the patient and must beavoided at all costs. Very importantly, each mis-placement of the needle necessitates an extra CTimage, each of which exposes the patient to addi-tional radiation.

Fig. 1. The robotic system used in the phantom experi-ment is easily deployable in prostate, spine, and abdominaltreatments.

372 Masamune et al.: Robotically Assisted Procedures with CT Guidance

SYSTEMS DESCRIPTION

Design CriteriaTo provide the required clinical accuracy in nerveblocks and facet joint injections, the robotic systemshould demonstrate accuracy of 0.5 mm at a depthof 5.0 cm in air, and actual needle placement ac-curacy of about 1.5 mm in human tissue. One of thekey design criteria was to preserve, as much aspossible, the natural workflow of manually actu-ated, CT-guided needle insertions. Clinical usabil-ity is heavily influenced by practical features suchas ease of calibration, mounting, and setup, as wellas the weight and dimensions of the robot. Al-though we did not set quantitative requirements forthese features in this highly experimental system,we kept these criteria in focus during the entireproject.

ConfigurationA schematic drawing of the overall configuration ispresented in Figure 2. The CT images are trans-ferred across a local area network (LAN) in DICOMformat to a Pentium II 333-MHz PC equipped with

a 17� flat panel display. The computer runs a “sim-ple storage” DICOM server, installed from thepublic domain source at http://www.erl.wustl.edu/DICOM.37 The operator of the scanner pushesDICOM images from the CT console through theLAN to the DICOM server. The computer also runsan interactive software that provides image pro-cessing, registration, and path planning, and alsocontrols the robot and the needle driver. The sur-geon uses an interactive display to execute theintervention step by step. Upon completing eachstep, the computer waits for confirmation beforecontinuing.

The mechanical side of the system is shownin Figure 3. A stiff seven-degrees-of-freedom(7DOF) passive arm is mounted on the CT couch.In the setup presented here, we use an off-the-shelfarm that reaches as far as 60 cm and weighs only3.5 kg. The arm locks and unlocks easily with ahandle. We also experimented with a device devel-oped by Lerner et al.38 that was superior in terms ofstiffness but more difficult to handle. To promoteencapsulation of robotic components, the amplifiersand power supplies are built inside the robot mount.

Fig. 2. Systems configuration.

Masamune et al.: Robotically Assisted Procedures with CT Guidance 373

We experimented with several embodiments, allweighting about 7–8 kg. Mounting on the CTcouch was preferred over mounting on the gantryfor several reasons. The geometry of the couch ismore standardized than that of the gantry, makingthe system more adaptable to various brands. Stifffixation on a gantry is difficult without drillingholes in it, which would raise serious quality-con-trol questions and would also probably void themanufacturer’s warranty on the scanner. Noninva-sive stiff mounting on the gantry would yield astructure considerably larger than a table mount.Finally, assembling and disassembling a gantrymount would require at least two technologists,which would be a most unfavorable feature.

A remote center-of-motion (RCM) robot de-veloped by Stoianovici et al.39–44 is attached to thepassive arm. The end-effector, which is a combi-nation of a motorized needle driver and a miniaturestereotactic frame (for more details, see sectionentitled Novel End-Effector), attaches to the RCM

robot. The end-effector and RCM robot weigh 0.15and 1.6 kg, respectively. This assembly is ideallysuitable for the robotization of manual needle punc-tures, which include the following three decoupledtasks: (1) touch down the needle tip on the entrypoint; (2) orient the needle by pivoting around theentry point; (3) enter the needle into the body alonga straight trajectory. Inserting a needle to an arbi-trary location requires six degrees of freedom. Ifthe skin entry point is prescribed, three degrees offreedom are necessary and sufficient to reach atarget. Two rotations are necessary to orient theneedle, and only one translation is necessary toinsert it. The system addresses safety by employinga low-DOF robot, decoupling needle orientationfrom the needle insertion, and using nonbackdriv-able transmissions. The system presents a minimalarchitecture and implements only the necessarythree degrees of freedom.

The orientation and insertion of the needle areimplemented by different mechanisms that are en-

Fig. 3. Mechanical overview of the couch-mounted system.

374 Masamune et al.: Robotically Assisted Procedures with CT Guidance

abled independently by safety switches. Duringneedle alignment, the surgeon activates only theRCM robot and pivots the needle around the inser-tion point on the skin. When the needle is properlyaligned, the RCM robot is deactivated. Needle in-sertion is enabled by turning on the power for theneedle driver. This scheme prevents insertion be-fore proper alignment is confirmed, and also pre-vents accidental changing of the needle path duringinsertion. The robot applies nonbackdrivable wormtransmission that preserves configuration when therobot is deactivated or in the event of power failure.The RCM robot has been used in multiple clinicalscenarios at the Johns Hopkins University, and itsrelevant mechanical features and parameters havebeen published extensively.39–44 The robot wasfound to be sufficiently small, dexterous, stiff, andaccurate for our purpose.

The combined weight of the system is about13 kg. One reasonably skilled technician can set upand take down the system in 10 min. The mount,passive arm, RCM robot, and needle driver foldconveniently into a carry-on suitcase.

Workflow

The workflow for the manual and robotically as-sisted procedures is conceptually identical. In thefollowing list, italics denote the new activities in-troduced by the robotic assistant. (We assume thatthe insertion point is already identified, cleaned,and prepared. We also assume that the needle isalready inserted in the needle driver and that therobot is properly mounted.)

1. Place the tip of the needle at the entry point.(Unlock the passive arm, position the light-weight robot manually, then lock the passivearm.)

2. Take one image slice. (Transfer the imageto the planning computer that performs ba-sic image processing and registration.)

3. Select the target point in the image. (Clickon the target point with the mouse.)

4. Determine needle angle and insertion depth.(Calculated by the computer.)

5. Aim the needle by pivoting around the entrypoint. (Enable the RCM robot, which aimsthe needle automatically. Disable the robotwhen done.)

6. Enter the needle to the calculated depth.(Enable the needle driver to advance theneedle. Disable the needle driver whendone.)

7. Confirm the location of the needle.8. Release the payload through the needle (in-

ject pain-reliever medication, etc.).9. Retract the needle and clean the site. (En-

able the needle driver to retract the needleor pull out the needle manually.)

It is important to emphasize that the roboticassistant does not alter the workflow of the manualneedle insertion procedure and does not require anycalibration.

Novel End-EffectorAs in any image-guided surgical system, accurateand robust registration between the surgical instru-ment and the image space is crucial. To achievefully image-based registration, rigid-body fiducialswere applied. As discussed earlier, invasively at-tached head frames and body frames did not farewell in procedures performed inside a CT gantry.The method of Susil and colleagues34 (attaching arigid-body fiducial pattern to the end-effector)seemed to be a viable solution. Because CT imagesare taken in the transverse direction with potentialgaps between them, a fiducial pattern combinedfrom straight rods is the most obvious choice.These fiducial systems, like the BRW or Leksellframes, have been used in similar circumstances fordecades, and their accuracies and error character-istics are well documented.15–18 Susil et al.34 alsoproved that the Z-shaped fiducial motifs of conven-tional stereotactic frames demonstrate favorable er-ror characteristics, accuracy, and reliability in alarge angular range. In our design (Fig. 4), thecombined needle driver and fiducial system form a

Fig. 4. End-effector. A combination of a needle driverand a stereotactic registration frame.

Masamune et al.: Robotically Assisted Procedures with CT Guidance 375

rigid body of known dimensions. In the miniatur-ized frame, two adjacent Z-shaped motifs share acommon fiducial rod, so there are only seven marksin each CT slice instead of the conventional ninemarks. However, this design feature has no effecton the essence and accuracy of the registration.

The kinematics of the RCM robot ensure thatthe remote center of motion is at all times in aconstant and known position with respect to theneedle driver. If, therefore, the needle is insertedinto the driver so that the needle tip and the remotecenter of motion coincide, the system never needscalibration. (During setup, the robot casts a thinlaser light that passes through the pivoting point.We adjust the needle until its tip coincides with thelaser.) Using only a priori geometric informationand the current CT image, we can calculate theorientation of the target with respect to the fiducialframe, as well as the distance of the target from theneedle tip.

The needle injector employs the kinematicprinciple “Friction Transmission with Axial Load-ing” patented by Stoianovici et al. (pending U.S.Patent Application Serial No. 09/026,669).33,35 Thetransmission is uniquely suited for a miniaturizedradiolucent construction intended to provide motor-ized needle actuation. A distinctive feature of thisdevice is that it grasps the barrel of the needleinstead of its head. This solution significantly re-duces the unsupported length of the needle, whichin turn, reduces lateral flexure during injection andincreases the accuracy of insertion.

The main parts of the needle driver areacrylic, which provides sufficient rigidity whileproducing negligible artifacts in CT images. TheZ-shaped markers are assembled from seven cylin-ders 5 mm in diameter. The cylinders are filled withX-ray blocking liquid (HYPAQUE, Nyromed Inc.,NJ, at 65% concentration), so the cylinders producesolid marks in CT. The size of the bounding boxdefined by the central axes of the fiducial rods is60 � 60 � 60 mm.

Axially loaded friction driving does not allowfor predictable depth control due to occasional slip-page. Should the transmission slip, the needle stopsshort and is guaranteed not to overshoot the target.For additional safety, we can fix a sterilized spring-loaded clamp on the needle at the intended inser-tion depth. This fixture indicates whether the needlehas slipped and stopped shallow and, very impor-tantly, provides an extra safety measure againstoverdriving the needle. Should the needle stop shortdue to transmission slippage, we can drive the needle

to the correct depth under direct joystick control,using the needle clamp as a depth indicator.

Registration and Targeting

This section explains the essence of device regis-tration and path planning. In our approach, one CTslice is necessary and sufficient for the calculationof needle orientation and insertion depth, assumingthat the target and all fiducial rods are visible in theimage. Our registration method produces a closedform solution, and there is no need for initial postureestimation or guessing. The method is reliable in awide range of rotations, as reported previously.34

It is important to stress again that, in ourprotocol, the skin entry point and the remote centerof motion are identical. This common point is pre-scribed by the physician before the robotic actioncommenced, and remains constant with respect tothe stereotactic frame. Because the fulcrum point ofthe robot remains constant for the entire lifetime ofthe robot and end-effector, the system never needscalibration. Targeting with the robotic device takesplace in the following steps:

1. Identify the marks of the fiducials and thetarget point in the CT image.

2. Ascertain the pixel size of the CT image.(We also assume that the images are suffi-ciently free of distortion, which is true formost current scanners. Therefore, the regis-tration does not require calibration of theCT scanner.)

3. Calculate the 6DOF transformation matrixthat will take the target point from imagecoordinates to stereotactic frame coordi-nates.

The geometric description of all threeZ-shaped motifs, as well as the spatial relationamong the three motifs, is known a priori. When animage plane intersects with a Z-shaped motif, thethree rods produce three distinctive marks in theimage. In our miniaturized frame, two neighboringZ-shaped motifs share a fiducial rod, so there areonly seven marks in each CT slice, as shown inFigure 5. Standard image processing steps are usedto obtain the positions of the seven markers inimage pixel coordinates. We apply binary thresh-olding and calculate the centroids of the ellipticalmarks in the thresholded image. The thresholdvalue is set interactively.

In the registration, the positions of the markson the Z-shape motifs are calculated, then theirlocations are obtained with respect to the fiducial

376 Masamune et al.: Robotically Assisted Procedures with CT Guidance

frame. The calculation for the intersection pointwith the slanted rod in a Z is shown in Figure 6.Our approach is conceptually identical to that ofSusil, but we present a less complex calculation.The intersections of three rods with the image planeare P1, P2, and P3. The length of the slanted rod isL. The distance measured between marks P1 and P2

is d12 � �P1 3 P2�, and the distance between P2

and P3 is d23 � �P2 3 P3�. From similar trianglesmarked with pattern filling in Figure 6, we gain theformula c � L*d23/(d12 � d23), where c is thedistance of P2 along the slanted rod from the cornerof the Z. Distances in the CT image (d12 and d23)are measured in pixel units, so multiplying them bythe known pixel size makes them available in frame

units (cm or mm). From c, we can calculate the P2

point in stereotactic frame coordinates. Repeatingthe calculation for the remaining two Z-shapedmotifs, we obtain P4 and P6 (refer back to Fig. 5).The three intersection points with the three slantedrods (P2, P4, and P6) are guaranteed not to becollinear. In fact, we could use any three noncol-linear fiducial points out of the seven available, butthe fiducial points on the slanted rods are alwayssufficiently distant from each other and never pro-duce a deformed triangle.

From here, the process of deriving theimage3frame transformation matrix is reduced toa rigid-body registration problem. We look for the6DOF transformation matrix, which takes the (P2,P4, P6) rigid triangle from image coordinates tostereotactic frame coordinates. There are severalways to produce this transformation matrix. In oursolution, we first calculate the gravity center of thetriangle, expressed in both image and frame coor-dinates. The offset between the two points repre-sents the 3DOF translation component of theimage3frame transformation. From the (P2, P4,P6) triangle, we can also derive an ortho-normalbase that is rigidly attached to the triangle. Weexpress this ortho-normal base in both CT imageand stereotactic frame coordinates. The rotationbetween the two bases represents the 3DOF rota-tional component of the image3frame transforma-tion.

Once the image3frame transformation is

Fig. 5. Fiducial marks, as seen in frame coordinates (a) and in a CT image (b).

Fig. 6. Calculation of the intersection point with a slantedfiducial rod.

Masamune et al.: Robotically Assisted Procedures with CT Guidance 377

available, the targeting algorithm proceeds as sum-marized below.

1. Apply the image3frame transformation tothe target point and obtain the target in thecoordinate space of the stereotactic frame.

2. Calculate the 2DOF rotation matrix that willbe applied to rotate the needle around its tip,so that the needle is aligned with the target.First we calculate the straight line that con-nects the (already transformed) target pointand the remote center of motion. This linerepresents the desired orientation of the nee-dle. Then we calculate the two rotations thattake the needle from its current orientationto its desired orientation.

3. Calculate the insertion depth, which is avail-able as the distance from the target point tothe remote center of motion (i.e., the tip ofthe needle).

Error AnalysisThis section provides error analysis for our regis-tration and targeting method. In this analysis, weconcentrate on the planning, rather than on theactual execution of the intervention.

In the open air, where there is no needle–tissue interaction, we systematically achieved anaccuracy better than 1 mm in hitting targets 5–8 cmfrom the fulcrum point. This result meets the re-quirement set in the Design Criteria section above.There are two main sources of error: hardware, andimage guidance. Regarding the hardware, the RCMrobot has proven to be sufficiently accurate formicrosurgery procedures43,44 and the needle-driv-ing mechanism was also analyzed and publishedpreviously.33,35 In conclusion, the robot hardwarerepresents negligible error.

Errors related to the calculation of insertiondepth are of less concern to us, because this processis straightforward and does not involve stereotacticregistration. The accuracy of the calculation de-pends solely on the resolution and slice thickness ofthe CT; an issue that has been extensively pub-lished in prior work.18

Errors related to the aiming of the needlehave two sources: the fiducial frame and its regis-tration to the CT space. The frame itself was accu-rately manufactured and carefully measured, and itscontribution to systemic error is negligible. As seenin the previous section, the registration softwareproduces two rotations for the robot joints. Wecompared these values to the actual readings fromthe joint encoders. We brought the robot to an

arbitrary starting position and took a CT image.Then we moved the robot arbitrarily several timesand took a CT image in each position. The regis-tration software calculated the relative rotation be-tween every two consecutive measurement posi-tions, and we also queried the same informationfrom the joint encoders. The average angular dif-ference was consistently under 1°, indicating anaverage aiming error of 0.75 and 1.5 mm at depthsof 50 and 100 mm, respectively. We suspect thatmuch of this error can be attributed to slice thick-ness, but more experiments are needed to separateall individual sources of the image processing error.

PHANTOM EXPERIMENT ANDRESULTSWe performed a series of phantom experimentswith the system, as shown in Figure 7. A melonrepresented the patient, and a 2-mm steel ball wasimplanted in the melon to serve as the target. Theexperiment followed the workflow described ear-lier. Figure 8 shows a transverse image taken be-fore insertion (left) and a thresholded binary imageof the same field (right). We performed a series ofneedle punctures using the system. One of theconfirmation images is shown in Figure 9. In thetransverse image (left), the tip of the needle accu-rately hits the target. In the corresponding projected“scout” view (right), a slight bending of the needlecan also be observed inside the phantom. By mea-suring the image acquired after needle insertion, theorientation accuracy was determined to be 0.6°, andthe distance between the needle tip and the targetwas 1.04 mm.

Fig. 7. Overview of the phantom experiment setup. Ahoneydew melon is used as a phantom. The robot ismounted on the scanner bed.

378 Masamune et al.: Robotically Assisted Procedures with CT Guidance

DISCUSSIONThe presented phantom experiments were an over-all success: We demonstrated the necessary accu-racy set and the ability to execute the workflow setin the sections entitled Design Criteria and Work-flow, respectively. Besides highlighting numerouspositive aspects of the system, the experiments alsounderlined several issues where further researchand development are necessary.

Our experiments suggest that the primarycause of needle placement error is needle–tissueinteraction. Inhomogeneities in the penetrated bodytissues may cause substantial needle placement er-ror. Ideally, the needle should be perpendicular tothe skin surface to avoid slippage and deflection ofthe needle. Needle bending is also a significantsource of error. In our controlled experiments wedetected only negligible bending, but this problemcan emerge as prohibitive in real-life applications ifnot addressed properly. Bending can be reduced inseveral ways. Applying a thicker and stronger nee-dle may be an option, but it may not be preferable,as it increases invasiveness. Symmetrical needleswith diamond or conic tips tend to bend signifi-cantly less than beveled needles. (In fact, manyphyisicians prefer beveled needles for manual in-sertion, because they can be steered by applyingtransverse force.) Recently, Kataoka et al. haveproposed a simple force–deflection function formodeling the insertion of beveled needles, and they

performed experimental validation of the model.45

Although this model underestimated real deflectionas measured in animal-tissue experiments, it rein-forced the observation that bending can be con-trolled most effectively by reducing exertion forceson the needle tip. This objective can be achieved byreducing friction forces between the tissue and theneedle. Early observations suggest that friction isinfluenced by the shape of the needle tip, and thatthe diamond tip is preferable over beveled andconic shapes. Friction can be reduced most signif-icantly by rotating the needle while advancing it. Inother words, one can drill the needle into the tissue,as opposed to merely pushing it. This techniquerequires a sophisticated needle driver with 2 de-grees of freedom, the development of which is amajor work in progress in our laboratories.

Friction transmission is associated with amaximum exertion force. If the resistance is greaterthan the maximum transmission force, the needlewill slip and stop short of the target. Even worse,multiple insertions increase the likelihood of slip-page, because body fluids can get into the trans-mission mechanism during retraction of the needle.In our experiments, to reduce the possibility ofslippage, we skinned the melon under the needletip. This is conceptually equivalent to incising thepatient’s skin at the needle tip, which often happensin clinical practice. In many treatments, however,incision is not a viable option. For these cases, a

Fig. 8. CT image with the end-effector frame and target point, as seen on the scanner’s console (a) and as seen in thetreatment planning software after binary thesholding (b).

Masamune et al.: Robotically Assisted Procedures with CT Guidance 379

needle driver with frictionless transmission needsto be developed.

During insertion, we also detected consider-able mechanical vibration of the robot, which wasmost probably caused by the friction between theneedle and the penetrated tissue. Strong vibrationmay displace the entry point of the needle. In thisstudy, the RCM robot was held by a flexible armjoint that may not have sufficient rigidity in allcases. Although the problem was not prohibitive inthe current setup, it needs to be investigated in afocused experiment.

In the present system, the needle must bemanually touched down at the entry point on thebody surface. If, however, an active 3DOF Carte-sian motion stage held the passive arm, then theentry point could also be picked from the CT im-ages, and the Cartesian stage could move the needleto the entry point. The use of a Cartesian stagewould fit naturally with the workflow. To maintain

modularity and safety, the Cartesian motion stagewould be decoupled from the other stages of therobotic system.

The 60-mm cube that encompasses the end-effector frame was found to be relatively small.Occasionally, it was difficult to adjust the robot tofind the target point in a CT slice in which we couldobserve all of the seven fiducial rods. The planningsoftware needs to be enhanced to deal with incom-plete fiducial patterns. Other possible correctionswould include using a bigger frame or more thanone CT slice, but these are not attractive choices.

The current image-processing software calcu-lates the centroids of the fiducial marks in a thresh-olded image. In the next generation of software, wewill calculate the mass center of each fiducial markin the original gray-scale image, where pixels willbe weighted by their Hounsfield numbers.

The current end-effector cannot release itsgrasp on the needle while the needle is inside the

Fig. 9. Confirmation images after insertion. a: An image slice with target, needle tip, and end-effector frame. b: Scout viewwith target, needle, and robot.

380 Masamune et al.: Robotically Assisted Procedures with CT Guidance

body, which may be a problem in long proceduresor when involuntary movement of the patient isanticipated.

Our end-effector, which is made of acrylicand rubber, becomes MRI compatible if one fillsthe fiducial rods with MRI-opaque solvent. MRIprovides exquisite tissue resolution in arbitrary im-age planes, offering a highly promising image-guidance environment for our end-effector.

Clinical safety is a crucial issue, which, un-fortunately, could be addressed only tangentially inthis paper. In general, clinical trials on human sub-jects cannot be planned without a detailed safetyevaluation of the entire system, including the robot,the end-effector, and the control software. This is amajor work in progress.

CONCLUSIONWe have demonstrated a novel system for roboti-cally assisted needle insertion inside a CT scanner.We also demonstrated an X-ray-compatible needledriver that can also be used for the registration ofthe robot and imager, based on a single CT image.Phantom experiments indicated that this roboticsystem is suitable for a variety of percutaneousclinical applications. Compared to other knownrobotic systems, our system appears to be smaller,simpler, easier to use, and more cost-effective. Fur-ther experiments are needed to evaluate the safetyof the system in particular applications.

ACKNOWLEDGMENTThe authors gratefully acknowledge the support ofthe Japanese government under grant #JSPS-RFTF99I00904 and the support of the NationalScience Foundation under grant #EEC9731478 forthe Engineering Research Center. This work hasalso been supported, in part, by Johns HopkinsUniversity internal funds.

REFERENCES1. Jain RK. The next frontier of molecular medicine:

delivery of therapeutics. Nat Med 1997;4:655–657.2. Klausner RD. The Nation’s Investment in Cancer Re-

search: A Budget Proposal for Fiscal Year 1999. Be-thesda, Maryland: National Cancer Institute, 1997.

3. Yousem DM, Sack MJ, Scanlan KA. Biopsy of para-pharyngeal space lesions. Radiology 1994;193:619–622.

4. Yousem DM, Sack MJ, Weinstein GS, Hayden RE.Computed tomography-guided aspirations of parapha-ryngeal and skull base masses. Skull Base Surg 1995:5:131–136.

5. Sack MJ, Weber RS, Weinstein GS, Chalian AA,Nisenbaum HL, Yousem DM. Image-guided fine-nee-

dle aspiration of the head and neck: five years’ expe-rience. Arch Otolaryngol Head Neck Surg 1998;124(10):1155–1161.

6. Jacob AL, Messmer P, Kaim A, Suhm N, RegazzoniP, Baumann B. A whole-body registration-free navi-gation system for image-guided surgery and interven-tional radiology. Invest Radiol 2000;35(5):279–288.

7. McDonald JS. Computer driven needle probe enablestherapy for painful neuropathies. Stud Health TechnolInform 2000;70:202–206.

8. Gangi A, Dietemann JL, Mortazavi R, Pfleger D,Kauff C, Roy C. CT-guided interventional proceduresfor pain management in the lumbosacral spine. Radio-graphics 1998;18(3):621—633.

9. Pelz DM, Haddad RG. Radiologic investigation oflow back pain. CMAJ 1989;140(3):289–295.

10. Proske M, Steinbrich W. CT-guided pain treatment.Schweiz Med Wochenschr 1996;126(29):1270–1273.

11. Brown RA. A stereotactic head-frame for use with CTbody scanner. Invest Radiol 1979;14:300–304.

12. Brown RA, Roberts TS, Osborne AG. Stereotacticframe and computer software for CT-directed neuro-surgical localization. Invest Radiol 1980;15:308–312.

13. Goers S, et al. A computed tomographic stereotacticadaptation system. Neurosurgery 1982;10:375–379.

14. Leksell L, Jerenberg B. Stereotaxis and tomography: atechnical note. Acta Neurochir 1980;52:1–7.

15. Galloway RL Jr, Maciunas RJ, Latimer JW. The ac-curacies of four stereotactic frame systems: an inde-pendent assessment. Biomed Instrum Technol 1991;25(6):457–460.

16. Maciunas RJ, Galloway RL Jr, Latimer J, Cobb C,Zaccharias E, Moore A, Mandava VR. An indepen-dent application accuracy evaluation of stereotacticframe systems. Stereotact Funct Neurosurg1992;58(1–4):103–107.

17. Maciunas RJ, Galloway RL Jr, Latimer JW. The ap-plication accuracy of stereotactic frames. Neurosur-gery 1994;35(4):682–694; discussion 694–695.

18. Zylka W, Sabczynski J, Schmitz G. A Gaussian ap-proach for the calculation of the accuracy of stereo-tactic frame systems. Med Phys 1999;26(3):381–391.

19. Young RF. Application of robotics to stereotactic neu-rosurgery. Neurol Res 1987;9:123–128.

20. Kwoh YS, Hou J, Jonckeere EA, Hayati S. A robotwith improved absolute positioning accuracy for CTguided stereotactic brain surgery. IEEE Trans BiomedEng 1988;35(2):153–160.

21. Darke JM, Joy M, Goldenberg A, et. al. Computer androbotic assisted resection of brain tumors. Proceedingsof 5th International Conference on Advanced Robot-ics, 1991. p 888–892.

22. Lax I, et al. Stereotactic radiotherapy of malignanciesin the abdomen. Methodological aspects. Acta Oncol1994;33(6):677–683.

23. Erdi Y, Wessels B, DeJager R, Erdi A, et al. A newfiducial alignment system to overlay abdominal CT or

Masamune et al.: Robotically Assisted Procedures with CT Guidance 381

MR images with radiolabeled antibody SPECT scans.Cancer 1994;73(3 Suppl):923–931.

24. Blomgren H, Lax I, Goranson H, Kraepelien T, et al.Radiosurgery for tumors in the body: clinical experi-ence using a new method. J Radiosurg 1998;1(1):63–75.

25. Takacs I, Hamilton AJ. Extracranial stereotactic radio-surgery: applications for the spine and beyond. Neu-rosurg Clin N Am 1999;10(2):257–270.

26. Lohr F, Debus J, Frank C, Herfarth K, Pastyr O, RheinB, Bahner M, Schlegel W, Wannenmacher M. Non-invasive patient fixation for extracranial stereotacticradiotherapy. Int J Radiat Oncol Biol Phys 1999;45(2):521–527.

27. Pelastrant AM. Comprehensive approach to TC-guided procedures with a hand-held device. Radiology1990;174:270–272.

28. Ishizaka H, Katsuya T, Koyama Y, Ishijima H, MotekiT, et al. CT-guided percutaneous intervention usingsimple laser director device. AJR 1998;170(3):745–746.

29. Gangi A, Kastler B, Arhan JM, Klinkert A, et al. Acompact laser beam guidance system for interven-tional CT. J Comp Assist Tomogr 1994;18:326–328.

30. Frahm C, Kloeess W, Gehl HB, et al. First experi-ments with a new laser-guidance device for MR- andCT-guided punctures. Eur Radiol 1994;5:315.

31. Masamune K, Ji LH, Suzuki M, Dohi T, Iseki H,Takakura K. A newly developed stereotactic robotwith detachable drive for neurosurgery. In: WellsWM, Colchester A, Delp S, editors: Proceedings ofFirst International Conference on Medical ImageComputing and Computer-Assisted Intervention(MICCAI’98), Cambridge, MA, October 1998. Lec-ture Notes in Computer Science 1496. Berlin:Springer, 1998. p 215–222.

32. Loser M, Navab N. A new robotic system for visuallycontrolled percutaneous interventions under CT. In:Delp SL, DiGioia AM, Jaramaz B, editors: Proceed-ings of Third International Conference on MedicalImage Computing and Computer-Assisted Interven-tion (MICCAI 2000), Pittsburgh, PA, October 2000.Lecture Notes in Computer Science 1935. Berlin:Springer, 2000. p 887–897.

33. Stoianovici D, Cadeddu JA, Demaree RD, Basile HA,Taylor RH, Whitcomb LL, Sharpe WN Jr, KavoussiLR. An efficient needle injection technique and radio-logical guidance method for percutaneous procedures.In: Troccaz J, Grimson E, Mosges R, editors: Proceed-ings of First Joint Conference on Computer Vision,Virtual Reality and Robotics in Medicine and MedicalRobotics and Computer-Assisted Surgery (CVRMed-MRCAS’97), Grenoble, France, March 1997. LectureNotes in Computer Science 1205. Berlin: Springer,1997. p 295–298.

34. Susil RC, Anderson JH, Taylor RH. A single imageregistration method for CT-guided interventions. In:Taylor C, Colchester A, editors: Proceedings of Sec-

ond International Conference on Medical Image Com-puting and Computer-Assisted Intervention (MIC-CAI’99), Cambridge, UK, September 1999. LectureNotes in Computer Science 1679. Berlin: Springer,1999. p 798–808.

35. Stoianovici D, Kavoussi LR, Whitcomb LL, TaylorRH, Cadeddu JA, Basile HA, Demaree RD. FrictionTransmission with Axial Loading and a RadiolucentSurgical Needle Drive. Provisional US Patent of In-vention No. 60/038,115. Filed as regular US utilityand PCT application by Johns Hopkins University onFebruary 20, 1998.

36. Taylor RH, Fichtinger G, Jansen P, Riviere C. Medicalrobotics and computer-integrated surgery: informa-tion-driven systems for 21st century operating rooms.J Jpn Soc Comp Aid Surg 2000;2(2):47–53.

37. DICOM Web home page: http://www.erl.wustl.edu/DICOM/.

38. Lerner GA, Stoianovici D, Whitcomb LL, KavoussiLR. A passive positioning device for surgical robotsand instrumentation. In: Taylor C, Colchester A, edi-tors: Proceedings of Second International Conferenceon Medical Image Computing and Computer-AssistedIntervention (MICCAI’99), Cambridge, UK, Septem-ber 1999. Lecture Notes in Computer Science 1679.Berlin: Springer, 1999. p 1052–1061.

39. Cadeddu JA, Stoianovici D, Chen RN, Moore RG,Kavoussi LR. Stereotactic mechanical percutaneousrenal access. J Endourol 1998;12(2):121–126.

40. Cadeddu JA, Stoianovici D, Chen RN, Moore RG,Kavoussi LR. Mechanical percutaneous renal access.J Urol 1998;159(5):110.

41. Stoianovici D, Cadeddu JA, Demaree RD, Basile HA,Taylor RH, Whitcomb LL, Kavoussi LR. A novelmechanical transmission applied to percutaneous renalaccess. Proceedings of the ASME Dynamic Systemsand Control Division, DSC-Vol. 61, 1997. p 401–406.

42. Stoianovici D, Cadeddu JA, Demaree RD, Basile HA,Taylor RH, Whitcomb LL, Sharpe WN Jr, KavoussiLR. An efficient needle injection technique and radio-logical guidance method for percutaneous procedures.In: Troccaz J, Grimson E, Mosges R, editors: Proceed-ings of First Joint Conference on Computer Vision,Virtual Reality and Robotics in Medicine and MedicalRobotics and Computer-Assisted Surgery (CVRMed-MRCAS’97), Grenoble, France, March 1997. LectureNotes in Computer Science 1205. Berlin: Springer,1997. p 295–298.

43. Taylor R, Jensen J, Whitcomb L, Barnes A, Kumar R,Stoianovici D, Gupta P, Wang ZX, deJuan E,Kavoussi L. A steady-hand robotic system for micro-surgical augmentation. Int J Robotics Res1999;18(12):1201–1210.

44. Stoianovici D, Whitcomb LL, Anderson JH, TaylorRH, Kavoussi LR. A modular surgical robotic systemfor image-guided percutaneous procedures. In: WellsWM, Colchester A, Delp S, editors: Proceedings of

382 Masamune et al.: Robotically Assisted Procedures with CT Guidance

First International Conference on Medical ImageComputing and Computer-Assisted Intervention(MICCAI’98), Cambridge, MA, October 1998. Lec-ture Notes in Computer Science 1496. Berlin:Springer, 1998. p 404–410.

45. Kataoka H, Washio T, Audette M, Mizuhara K. Amodel for relations between needle deflection, force,and thickness on needle penetration. Proceedings ofFourth International Conference on Medical ImageComputing and Computer-Assisted Intervention(MICCAI 2001), Utrecht, The Netherlands, October

2001. Lecture Notes in Computer Science 2208. Ber-lin: Springer, 2001. p 966–974.

46. Dohi T, Hata N, Miyata K, Hashimoto D, Takakura K,Chinzei K, Yamauchi Y. Robotics and mechatronicsin computer aided surgery. Jpn J Comp Aid Surg1995;1(1):4–10.

47. Masamune K, Kobayashi E, Masutani Y, Suzuki M,Dohi T, Iseki H, Takakura K. Development of anMRI-compatible needle insertion manipulator for ste-reotactic neurosurgery. J Image Guid Surg 1995;1(4):242–248.

Masamune et al.: Robotically Assisted Procedures with CT Guidance 383