`
`26 1
`
`An Image-Directed Robotic System
`for Precise Orthopaedic Surgery
`
`Russell H. Taylor, Fellow, IEEE, Brent D. Mittelstadt, Howard A. Paul,
`William Hanson, Peter Kazanzides, Member, IEEE, Joel F. Zuhars, Member, IEEE,
`Bill Williamson, Bela L. Musits, Edward Glassman, and William L. Bargar
`
`Abstract-We have developed an image-directed robotic system
`to augment the performance of human surgeons in precise bone
`machining procedures in orthopaedic surgery, initially targeted
`at cementless total hip replacement surgery. The total system
`consists of an interactive CT-based presurgical planning com-
`ponent and a surgical system consisting of a robot, redundant
`motion monitoring, and man-machine interface components. In
`vitro experiments conducted with this system have demonstrated
`an order-of-magnitude improvement in implant fit and placement
`accuracy, compared to standard manual preparation techniques.
`The first generation system described in this paper was used
`in a successful veterinary clinical trial on 26 dogs needing hip
`replacement surgery. It was the basis for subsequent development
`of a second-generation system that is now in human clinical trials.
`
`I. INTRODUCTION AND BACKGROUND
`
`researchers to explore the use of robotic devices to augment
`a surgeon’s ability to perform geometrically precise tasks
`planned from computed tomography (CT) or other image data.
`The pioneering work in the use of general purpose robots
`for surgery was that of Kwoh et al. [l] who used a six-axis
`industrial robot to replace a stereotactic frame in neurosurgery.
`In this case, the robot was mounted in a known position
`relative to the table of a CT scanner and suitable geometric
`calibrations were performed. During surgery, the patient was
`CT-scanned and a desired placement for a biopsy needle probe
`was determined from the image data. The robot then positioned
`a passive needle guide appropriately, brakes were applied, and
`power was turned off. Finally, the surgeon inserted the needle
`through the guide into the patient’s brain. The principal benefit
`gained was the greater convenience and faster positioning
`possible with the robot, compared to the use of a stereotactic
`frame. A number of similar systems have been developed
`subsequently. The most successful to date is that of Lavallee,
`et al., ([2], [3]), who used a stereo pair of intraoperative
`radiographs to register the robot to the patient’s CT data
`(and to the patient) and to plan needle paths that avoid
`blood vessels. Over three hundred cases have been performed,
`although (again) the robot is turned off while the needle is
`inserted. Kelly, et al. [4], [ 5 ] have implemented a specialized
`motorized stereotactic system for laser neurosurgery, in which
`an XYZ table is used to reposition the patient’s head relative to
`the focal point of a surgical microscope. More recently, Drake,
`Goldenberg, et al. [6] have reported several cases in which a
`general purpose robot moved while in contact with the patient,
`although the motions were very simple and highly constrained.
`Further, these cases were performed on an exception basis,
`in which the surgeon had no practical alternative, despite
`somewhat more limited safety checking than would have been
`desirable for more routine use. Several other neurosurgery
`robots are in various stages of development (e.g., [7]).
`A number of active robotic systems for augmentation of
`non-neurosurgical procedures have also been proposed or
`developed. For example, Davies et al. have developed a spe-
`cialized robotic device to assist in laparoscopic prostatectomies
`[8], which has been used clinically. A number of groups (e.g.,
`[9]-[ 121) have developed a variety of other telerobotic devices
`for endoscopic and laparoscopic surgery. McEwen et al. have
`developed and marketed a clinically qualified voice controlled
`limb positioning system for orthopaedics [ 131. Several groups
`(e.g., [ 141, [ 151) have demonstrated in vitro robotic systems for
`1042-296X/94$04.00 0 I994 IEEE
`
`A . Augmentation of Human Skill in Surgery
`HE RESEARCH reported in this paper represents a step
`
`T in an evolving partnership between humans (surgeons)
`
`and machines (computers and robots) that seeks to exploit the
`capabilities of both to do a task better than either can do alone.
`Recent advances in medical imaging technology (CT, MRI,
`PET, etc.), coupled with advances in computer-based image
`processing and modelling capabilities have given physicians
`an unprecedented ability to model and visualize anatomical
`structures in live patients, and to use this information quanti-
`tatively in diagnosis and treatment planning. Further, advances
`in CAD-CAM technology have made it practical to use this
`data to design and precisely fabricate custom surgical implants
`for individual patients.
`One result is that the precision of image-based presurgi-
`cal planning often greatly exceeds the precision of surgical
`execution. Typically, geometrically precise surgery has been
`limited to procedures (such as brain biopsies) for which a
`suitable stereotactic frame is available. The inconvenience
`and restricted applicability of these devices has led many
`
`Manuscript received December 16, 1991; revised March 14, 1994.
`R. H. Taylor and E. Glassman are with the IBM T. J. Watson Research
`Center, Yorktown Heights, New York 10598. USA.
`B. D. Mittelstadt, P. Kazanzides, J. F. Zuhars, B. Williamson, B. L. Musits,
`and W. L. Bargar are with Integrated Surgical Systems, Sacramento, California
`95834, USA.
`H. A. Paul, deceased, was with Integrated Surgical Systems, Sacramento,
`California 95834, USA.
`W. Hanson was with the IBM Palo Alto Science Center, Palo Alto, CA
`95834, USA. He is now with Loral Federal Systems, Gaithersburg, MD 20879,
`USA.
`IEEE Log Number 9403327.
`
`Mako Exhibit 1009 Page 1
`
`
`
`262
`
`IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION, VOL. 10, NO. 3, JUNE 1994
`
`positioning passive instrument guides for knee replacement
`surgery. Of these applications, that of Davies comes closest
`to ours in the sense that it uses an active automatic device
`to perform a tissue removal operation. Important differences
`include an order-of-magnitude difference in the accuracy re-
`quired for the application, the greater complexity of the shapes
`to be cut, the use of a general purpose manipulator rather
`than a specialized devise, and the greater degree of safety and
`consistency checking built into our system, which must move
`safely in a much less constrained volume.
`
`B . Precise Orthopaedic Surgery
`Orthopaedic applications represent a particularly promising
`domain for the integration of image and model-based presur-
`gical planning, CAD/CAM technology, and precise robotic
`execution. For example, about half of the 300000 total hip
`replacement operations performed each year use cementless
`implants. In these procedures, accurate preparation of the
`femoral cavity to match the implant shape and accurate place-
`ment of the cavity relative to the femur can significantly affect
`stress transfer, implant stability, and restoration of proper
`biomechanics, which, in turn, are important factors affecting
`efficacy. For example, Sandbom, et al., [ 161. have reported that
`the size of gaps between bone and implant significantly affects
`bone ingrowth. Furthermore, the present manual broaching
`method’ for preparing the femoral cavity leaves considerable
`room for improvement. In one recent study Paul, Hayes, et al.
`[ 171 found that only about 20% of the implant actually touches
`bone when it is inserted into a manually broached hole. The
`average gap between the implant and the bone was commonly
`1-4 mm and the overall hole size was 36% larger than the
`broach. Furthermore, the exact placement of the implant cavity
`relative to the bone (which affects restoration of biomechanics)
`depends on the surgeon’s ability to line up the broach manually
`and to drive it the right distance into the femur. Driving the
`broach too far can split the femur.
`These considerations have led us to explore the use of
`robotic machining to prepare the femoral cavity for the im-
`plant. Initial feasibility studies by Paul, Mittelstadt, et al. [ 181
`demonstrated that a robot could successfully machine shapes
`in human cadaver bones and that preoperatively implanted
`calibration pins could be used to accurately register CT image
`and robot coordinates for a femur.
`Following these studies, we developed a complete planning
`and execution system suitable for use in an actual operating
`room. In vitro experiments with this first generation system
`demonstrated an order-of-magnitude improvement in surgical
`precision, compared to manual broaching. One of the authors
`(Dr. Paul) conducted a veterinary clinical trial on dogs needing
`hip replacement surgery. This experience provided the basis
`for development of a second-generation system that is now in
`human clinical trials [ 19]-[22].
`Subsequent sections of this paper will summarize the presur-
`gical planning and surgical procedure followed for robotic
`’ Fig. I shows a typical cementless implant and the corresponding broach
`
`used to make the hole for it. Fig. 2 shows the use of a broach on a human
`patient. The procedure in a dog is essentially the same.
`
`Fig. I . Typical cementless hip implant and instrumentation. This figure
`shows a typical cementless hip implant, together with the broach used to
`produce a corresponding hole in the patient’s thigh in conventional manual
`surgery. Proper placement of the implant socket relative to the femur and
`accurate reproduction of the socket shape are very important to assure stability,
`uniform stress transfer, and restoration of the proper biomechanics.
`
`Fig. 2. Manual broaching procedure. This figure shows the use of a broach in
`a human cementless hip replacement. The procedure in a dog is essentially the
`same. One study found that only about 20% of the implant actually touches
`bone when it is inserted into a manually broached hole. The average gap
`between the implant and the bone was commonly 1 4 mm and the overall
`hole size was 36% larger than the broach.
`
`hip replacement surgery and will discuss the requirements
`for robotic systems intended to augment human precision
`in surgery. After providing a brief overview of the system
`architecture, we will provide a fuller discussion of several
`key aspects of the system, including the image-based presur-
`gical planning, geometric calibration, shape cutting and safety
`checking mechanisms. Finally, we will discuss experience of
`the system in actual clinical use (on dogs) and will discuss
`some of the lessons learned.
`
`11. SUMMARY OF PROCEDURE
`Before surgery, three titanium pins are implanted through
`small skin incisions into the greater trochanter and condyles
`of the patient’s femur. A CT scan is made of the leg. The
`presurgical planning system automatically locates the pins
`relative to the coordinate system of the CT images. The sur-
`
`Mako Exhibit 1009 Page 2
`
`
`
`TAYLOR et al.: AN IMAGE-DIRECTED ROBOTIC SYSTEM FOR PRECISE ORTHOPAEDIC SURGERY
`
`263
`
`IMPLNTED BONE
`
`PLANNING
`
`CALIBRATION PIN POS.
`IMPLANT SHAPE DATA
`IMPLANT PLACEMENT
`
`J
`
`MOTION
`MONITOR
`
`I
`
`L
`
`\
`SENSORS
`\
`\ CUTS/
`\PIN
`
`POS
`
`HAND -
`
`HELD
`TERMINAL
`
`IMPLANTED BONE
`Fig. 3. Architecture of hip replacement surgery system. The system consists
`of a presurgical planning component and a surgical component. In the system
`used for the veterinary clinical trial, the motion monitoring and robot control
`functions are subsumed within the robot controller.
`
`geon interactively selects an implant model and determines its
`desired placement relative to CT coordinates. This information
`is written to a diskette for use in surgery.
`Key steps of the intraoperative procedure are shown in Fig.
`5 for an in vitro test on a cadaver femur. Fig. 6 shows the
`operating room scene during the first canine clinical trial in
`May, 1990. Briefly, the procedure is as follows.
`1) The robot is brought into the operating room and pow-
`ered up. A sterile cutting tool is attached to a tool
`interface just below the force sensor, and the robot is
`covered with a sterile drape. The patient data diskette is
`loaded into the robot controller, and the robot is placed
`in a standby mode.
`2) The patient is prepared and draped in the normal manner.
`Surgery proceeds normally until the acetabular compo-
`nent of the implant is implanted and the ball of the femur
`is removed.
`3) The robot is brought up to the operating table, and
`the femur is rigidly attached to the robot base, using
`a specially designed fixator. The three titanium pins are
`exposed manually.
`4) A ball probe “cutter bit” is inserted into the collet of the
`cutting tool. The top center of each pin is then located
`by a combination of manual guiding and autonomous
`tactile search by the robot. Although several modes of
`manual guiding are available, the most commonly used
`is force compliance. The surgeon simply pulls on the
`shaft of the cutter; the robot controller senses the forces
`exerted on the tool and moves the robot in the indicated
`direction.
`
`5) The robot controller uses the pin location information to
`compute an appropriate transformation from CT coordi-
`nates to robot coordinates. The ball probe is replaced by
`a standard cutting bit, and the robot cuts out the desired
`implant shape at the planned position and orientation
`relative to the pins. The surgeon monitors progress both
`by direct observation of the robot and patient and by
`looking at a graphical display depicting successive cuts.
`6) When cutting is complete, the femur is unclamped from
`the fixator, and the robot is moved out of the way. The
`rest of the procedure proceeds in the normal way, with
`the added step of removing the locator pins from the
`patient.
`
`111. REQUIREMENTS AND ISSUES
`
`A. Human-Machine Interaction in a Surgical Situation
`Our goal is not to replace the surgeon. Instead, we are
`concemed with developing a surgical tool that can assist the
`surgeon by precisely executing a tissue removal task under
`the surgeon’s supervision. Although the robot’s geometric
`accuracy is much greater than the surgeon’s, the surgeon’s
`understanding of the total situation is clearly much greater
`than any computer’s, and he or she is responsible for what
`goes on in the operating room. Suitable interfaces must be
`provided to allow the surgeon to monitor the robot’s actions,
`to pause execution at any time, initiate error recovery actions,
`and provide positional guidance to the robot. There is also the
`related problem of human-computer interaction in presurgical
`planning. Convenient and naturally understood interfaces must
`be provided to allow the surgeon to specify what implant shape
`is to be cut and where it is to go. Furthermore, the interfaces
`used intraoperatively to report progress of the surgery should
`be as consistent as possible with those used to plan it.
`
`B. Registration of Plan Data with Intraoperative Reality
`The surgical plan is based on anatomical information de-
`rived from CT images taken prior to surgery. Reliable and
`accurate methods to locate the corresponding anatomical struc-
`tures relative to the robot are essential if the plan is to be
`executed successfully.
`
`C. Verification
`It is very important to verify that the greater potential
`geometric accuracy offered by the use of a robotic surgical
`system is in fact achieved in practical use. Suitable methods
`must be developed for verifying the performance of individual
`system components and of the system as a whole.
`
`D. Operating Room Compatibility and Sterility
`It must be easy to incorporate the robot into a hospital’s
`normal routine. It may be difficult for a hospital to dedicate
`an operating room to robotic surgery, and even if it does so
`it is important that maintenance not be disruptive.2 Gener-
`
`2These considerations led us to rule out some configurations (such as a
`Cartesian manipulator suspended from the ceiling) that might otherwise have
`been attractive.
`
`Mako Exhibit 1009 Page 3
`
`
`
`264
`
`IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION, VOL. IO, NO. 3, JUNE 1994
`
`Optical Tracking System
`
`one Motion MO
`
`Fig. 4. Operating room system architecture. The operating room system consists of 1) a surgical robot with its associated controller, tooling, and
`safety interlocks, 2) a fixator to hold the bone securely to the robot, 3) a redundant motion-monitoring subsystem consisting of a checking computer,
`optical tracking system, and bone motion detector, and 4) a human-machine interface with an online display, display computer, and a hand-held terminal
`interfaced to the robot controller.
`
`ally, the system should be easily brought into the operating
`room and set up as part of the normal presurgical routine.
`Similarly, removal, sterilization, and reattachment of the end
`effector and other critical components should be easy and
`suitable sterile drapes must be developed for the manipulator
`arm and other structural components that cannot be easily
`sterilized.
`
`E. Safety, Et-sos Recovesy, and Backup
`Clearly, redundant safety mechanisms are very important,
`both for the protection of the patient and of the surgeon. Man-
`ual pause and emergency power-off functions are essential.
`Wherever possible, potential error conditions must be antici-
`pated and checked for, and adequate recovery procedures must
`be available. Although the robot often may be able to continue
`with the procedure following a pause, it is also prudent to
`provide a reliable means of stopping the robot, removing it
`from the surgical field, and continuing the operation with
`manual backup.
`Since the surgeon must rely on the precision of the robot,
`it is extremely important that no single failure cause an
`undetected loss of accuracy. The system must monitor the
`position of the robot's cutting tool relative to the shape that
`it is supposed to cut and stop cutting if it strays out of
`the desired volume for any reason. It is especially important
`that systematic shifts (such as might arise from the bone
`slipping relative to the fixator) be detected promptly. A single
`misplaced cut can usually be repaired, but it may much harder
`to correct for misplacing the entire cavity.
`
`IV. SYSTEM ARCHITECTURE
`The system (Fig. 3 ) consists of a presurgical planning
`component and an intraoperative (surgical) component. These
`components are summarized below and discussed at greater
`length in subsequent sections.
`
`A . Presurgical Planning
`This component [23] implemented on an IBM workstation,
`permits the surgeon to select an implant model and size and
`to specify where the corresponding shape is to be machined
`in the patient's femur.
`The system maintains a library of computer-aided-design
`(CAD) models of implant designs and accepts computed
`tomography data for individual patients. It automatically de-
`termines the CT coordinates of the preoperatively implanted
`locator pins and provides a variety of interactive graphics
`tools for the surgeon to examine the CT data, to select an
`appropriate model and size from the implant design library,
`and to manipulate the position and orientation of the selected
`implant shape relative to CT coordinates.' The output consists
`of files containing 1) patient identification data 2 ) the position
`of the locator pins relative to CT coordinates, 3) the implant
`specification, 4) the desired implant placement relative to CT
`coordinates, and 5) processed image and model data that will
`be used for a realtime animation of the progress of the surgery.
`
`31n the future, we anticipate the use of computer optimimtion techniques
`to assist the surgeon in determining the best implant placement and also in
`the design of custom implants.
`
`Mako Exhibit 1009 Page 4
`
`
`
`TAYLOR et a/ : AN IMAGE-DIRECTED ROBOTIC SYSTEM FOR PRECISE ORTHOPAEDIC SURGERY
`
`265
`
`(e)
`(f)
`Fig. 5.
`Surgical procedure for hip surgery. (a) Fixated cadaver bone. (b) Manual guiding to approximate pin position. (c) Tactile search for a pin. (d)
`Cutting the shape. (e) Online display. (f) Final result.
`
`B . Operating Room System
`The operating room system (illustrated in Fig. 4) consists
`of several components. The five-axis robot is an IBM 7576
`SCARA manipulator with an added pitch axis, six degree-
`of-freedom force sensor, and a standard high-speed surgical
`cutting tool. During surgery, all but the robot’s cutting tool is
`covered by a sterile drape; the cutting tool is separately steril-
`ized. A sterile fixaror rigidly attached to the robot’s base holds
`the bone during the robotic part of the procedure. The robot
`controller provides servocontrol, low-level monitoring, sensor
`interfaces, and higher-level application functions implemented
`in the AML/2 language. During surgery, the force sensor is
`
`used to support redundant safety checking, tactile search to
`find the locator pins, and compliant motion guiding by the
`surgeon.
`The redundant motion monitoring subsystem 1241 is im-
`plemented on an IBM PC/AT with specialized IO hardware.
`It relies on independent sensing to track the position and
`orientation of the robot end effector during the cutting phase
`of the surgery, and checks to verify that the cutter tip never
`strays more than a prespecified amount outside of the defined
`implant volume. It also monitors strain gauges that can detect
`possible shifts of the bone relative to the fixation device. If
`either condition is detected, a “freeze motion” signal is sent
`to the robot controller. After motion is stopped. application
`
`Mako Exhibit 1009 Page 5
`
`
`
`266
`
`IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION, VOL. 10. NO. 3, JUNE 1994
`
`monitoring
`code in the robot controller queries the motion
`system for more information and then enters an
`appropriate
`error recovery procedure under the surgeon’s supervision.
`The human-machine interface includes an online display
`system that combines data generated in presurgical plan-
`ning with data transmitted from the robot controller to show
`progress of the cutting procedure superimposed on the CT-
`derived image views used in planning. A gas-sterilized hand-
`held ter-mina1 allows the surgeon to interact with the system
`during the course of the operation. This terminal supports
`manual guiding, motion enable, emergency power on/off,
`and menu selection functions. It may also be used to pace
`transitions from one major application step to the next and to
`select appropriate pre-programmed error recovery procedures
`should the need arise. Each of the major control components
`(robot control and motion checker) is able to freeze all
`robot motion or to rum off manipulator and cutter power in
`response to recognized exception conditions. If this happens,
`the surgeon must explicitly re-enable motion from the hand-
`held terminal.
`
`V. PRESURGICAL PLANNING SYSTEM
`
`A . Input Processing
`One mundane, but nevertheless essential, task is to load
`the image data into the computer. The CT scanner used for
`the veterinary clinical trial of this system produced images on
`magnetic tape in GE 9800 format. The voxel size for typical
`scans was 0.39 x 0.39 mm x 1.5 mm thick. Multiple cross-
`sectional images spaced 3 mm apart were taken throughout
`the proximal femur. In the vicinity of the locator pins, the
`images were spaced only 1.5 mm apart (i.e., they were
`contiguous). The input software includes facilities for tape
`reading, previewing image slices, selecting a region of interest
`to reduce the size of data sets, maintaining patient information,
`etc.
`
`B. Pin Location Algorithms
`A key problem is determining the location of the top
`center point of each locator pin relative to CT coordinates.
`This is by no means trivial. Although the density of the
`pins is much higher than that of bone, simple segmentation
`based on thresholding is complicated by blooming and other
`artifacts associated with the image formation process, so that
`the images are rather noisy. In particular, edge information is
`very ~nreliable.~ The pins are not nicely aligned with the CT
`slices, and the CT voxels are not cubes. Even in the absence of
`noise, CT cross-sections that pass through the screw threads,
`hexagonal drive hole, and the pin head and shaft can produce
`images that are rather difficult to analyze. To overcome these
`problems, a robust three phase method has been developed.
`In the first phase, simple density thresholding is used to
`distinguish the metallic pin voxels from surrounding “tissue”
`
`‘Experiments by one ot the authors [18], 125) with various materials
`showed that titanium and ceramic yielded the best contrast without excessive
`blooming. However, thc rehulting images were still far from clean. Titanium
`was chown for reasons of biocompatibility and because it is more commonly
`used in orthopaedic implants than is titanium.
`
`Fig. 6. Operating room scene from first canine clinical trial in May, 1990.
`The surgeon is Dr. Paul. The patient was a family pet needing hip replacement
`surgery.
`
`voxels. Unfortunately, blooming causes many “tissue” voxels
`to be mislabeled “pin,” giving the pins a ragged “starburst”
`appearance. These artifacts are cleaned up by first dilating and
`then eroding the binary thresholded image with standard 3D
`morphology filters, using spherical structural elements. This
`process also smooths out the screw threads and fills in the
`drive socket of the pin image.
`In the next phase, the approximate position and orientation
`of the pin are determined by calculating the first and second
`moments of the binary pin image.
`
`and
`
`where the p j are the coordinates of all voxels j classified
`“pin.” Since the pin is cylindrically symmetric, two of the
`eigenvectors of M2 will be practically equal. The other
`eigenvector, a, represents the principal axis of the pin.5
`’ This method would not work if the length of the pin shaft was such that
`
`all three eigenvectors had the same length. In this case, it would be necessary
`to use higher order moments to disambiguate the axes. However, our locator
`pin design precludes this possibility.
`
`Mako Exhibit 1009 Page 6
`
`
`
`TAYLOR et al.: AN IMAGE-DIRECfED ROBOTIC SYSTEM FOR PRECISE ORTHOPAEDlC SURGERY
`
`-
`
`261
`
`Fig. 7 Robot’s wrist during shape cutting experiment. The LED beacon
`plates used by the motion monitoring system are clearly visible. The force
`sensor is just visible behind the top of the plates.
`
`;
`
`l e y M2 ellipse
`
`Fig. 9. Presurgical planning display.
`
`information. The interactive display screen is shown in Fig. 9.
`Three orthogonal sections through rhe CT data set representing
`the bone are shown, together with a simple graphic view
`showing the location of the three cutting planes relative to the
`data set. Standard resampling techniques are used to generate
`undistorted cross-sectional images, which may be displayed
`in one of three modes. Grey-scale mode simply displays the
`CT densities of each (resampled) voxel. Color-map mode
`uses different hues (red, blue, etc.) to represent different
`tissue classes (cortical bone, trabecular bone, etc.), which are
`presently computed by relatively simple intensity thresholding
`techniques.6 Surface contour mode shows a graphic represen-
`tation of boundaries between tissue types. This graphic data
`can be manipulated very quickly, and is most useful when
`the surgeon is identifying the desired cross-sectional views
`through the CT data.
`In use, the surgeon typically selects boundary mode and
`uses the mouse to position and orient the cutting planes
`relative to the CT data. The surgeon then selects either grey-
`scale or color-map mode. Again using the mouse, the surgeon
`selects the desired implant model from a library of available
`designs, and manipulates the position and orientation of the
`implant relative to the CT coordinate system. As he does
`this, the computer automatically generates the cross-sections
`corresponding to the selected orthogonal cross sections and
`displays them superimposed on the corresponding 2D images.
`All manipulations, whether of the implant or of the cross-
`sectional CT views are specified relative to one of the three
`2D views. Thus, complex 6D reorientations are accomplished
`by breaking them down into a sequence of simpler transfor-
`
`6The threshold values used to distinguish between different bone classes
`were qualitatively determined by the co-author who is a surgeon (Dr. Paul),
`and reflect his best judgement as to what is useful. Any such distinctions are
`to some extent arbitrary.
`
`M1
`Density Profile
`(b)
`Fig. 8. Projected pin profile. (a) Projected geometry. (b) Density profile.
`
`In the third phase, a cross-sectional volume profile h(d)
`is computed as a function of the distance d along the axis
`ml + da (Fig. 8). The intercept do of the “leading edge” of
`the pin profile is computed, and the top center point ptc of the
`pin is then readily computed from
`
`Ptc = ml+ doa
`C . Interactive Docking Subsystem
`The interactive docking subsystem integrates 3D image dis-
`play and computer graphics techniques to support positioning
`of a 3D CAD model of the desired prosthesis shape relative to
`the CT image of the patient’s anatomy. Since 3D perspective
`projections inherently distort distance and shape, we chose
`to use orthogonal 2D cross-sections to represent the 3D
`
`Mako Exhibit 1009 Page 7
`
`
`
`268
`
`IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION, VOL. IO, NO. 3, JUNE 1994
`
`mations. When the surgeon is satisfied, the coordinates of each
`locator pin, the implant specification, and the desired implant
`position and orientation relative to CT coordinates are written
`to a file.
`In the future, we expect that the computer will assist the
`surgeon by computing and displaying appropriate goodness-
`of-fit measures and eventually proposing optimized positions
`and custom implant designs. Even in its present state of
`development, however, this system has proved to be very
`effective and quite easy to use. The 2D cross sectional dis-
`plays are intuitively attractive to and easily leamed by the
`orthopaedic surgeons who are the targeted end-users. The
`restriction to one 2D rotation or translation at a time has
`similarly proved to be inconsequential since our users tend to
`think of rotations and translations that are easily perceivable
`in a single display-namely
`the ones that the system allows
`on a single interaction.
`
`VI. GEOMETRIC CALIBRATION
`Geometric calibration (e.g., [26]-[28]) is a crucial compo-
`nent of any practical robotic application, especially one in
`which geometrically accurate paths are an important factor.
`This is equally true of surgical applications. At the same
`time, it is important to define methods that are simple, robust,
`do not require elaborate equipment, and are appropriate for
`the accuracies required by the task. In this section, we will
`describe our approach to these tradeoffs.
`
`A . Find Pin Routine
`The methods used in the calibration and in the actual
`surgical execution are very similar to methods earlier used
`in training a robot to copy pilot hole positions for automatic
`drilling of aircraft wing panels [29]. A ball probe cutter is
`inserted into the collet of the cutting tool and the force sensor
`is used to determine points of contact with the object being
`located (typically, a cylindrical pin). Points of contact are
`located by moving the ball to the proximity of the surface and
`then executing a slow guarded motion in a specified direction.
`As soon as the force exceeds a specified threshold, the motion
`is stopped. Since there may be an unpredictable amount of
`overshoot, a sequence of very small steps z; are then taken
`in the reverse direction, and the forces fi along the motion
`direction are measured at each point. The apparent compliance
`is estimated by a straight line approximation
`f; = K ( z ; - 20)
`
`The point z o where the force goes to 0 is assumed to be the
`contact point. Experience has shown that this method, while
`somewhat tedious, is in practice very robust. Repeatabilities
`of the order of 25 pm are routinely obtained. A cylindrical
`object like a pin or cup is then easily located by locating three
`points on the top surface and three points on the side.
`
`six degree-of-freedom force sensor and a high speed revolute
`surgical cutter. The nominal kinematics are given by
`
`where R(a, 0) is a rotation by angle I9 about axis a and
`Pwrist = R(z, 01) (11% + R(z, I 9