WO2017027905A1

WO2017027905A1 - Imaging method and system

Info

Publication number: WO2017027905A1
Application number: PCT/AU2016/050593
Authority: WO
Inventors: Marcus PANDY; Shanyuanye GUAN; Farzad KEYNEJAD; Hans GRAY
Original assignee: The University Of Melbourne
Priority date: 2015-08-14
Filing date: 2016-07-07
Publication date: 2017-02-23

Abstract

An imaging system, comprising: one or more imaging units comprising respective emitters for emitting imaging radiation and corresponding detectors for detecting the radiation after interaction of the radiation with a target portion of a specimen located in an image capture volume of the system; a mobile gantry for supporting the one or more imaging units, the one or more imaging units defining the image capture volume of the system and being moveable relative to the gantry; a drive controllable to move the gantry and to move the one or more imaging units relative to the gantry, such that the one or more imaging units are controllably moveable in at least two non-parallel directions in at least one plane; at least one tracking sensor supported by the gantry and configured to track the target portion of the specimen located in the image capture volume and to output location data indicative of a location of the target portion; and a controller configured to receive the location data from the tracking sensor and to control the drive based on the location data to move the gantry and the one or more imaging units relative to the gantry to track the target portion with the one or more imaging units, and related imaging method.

Description

IMAGING METHOD AND SYSTEM

FIELD

The present invention relates to an imaging method and system, of particular application in the X-ray fluoroscopy of joints while in motion.

BACKGROUND

Accurate in vivo measurement of human joint motion is important for understanding normal and pathological joint function and for evaluating different surgical techniques and implant designs used to treat joint osteoarthritis. Video-based motion capture with skin markers is the most common method for non-invasive assessment of 3D joint kinematics in vivo. However, this approach is limited by the errors associated with the non-rigid movement of the soft-tissue interface between the skin markers and the underlying bone. For example, root-mean-squared (RMS) errors as high as 29 mm for translations and 24° for rotations have been reported for markers placed near the knee during flexion-extension movements of the joint.

X-ray fluoroscopy is a dynamic imaging method that allows real-time visualization of bone motion in vivo. A generator produces an X-ray beam that irradiates the target joint while an image intensifier converts the X-ray beam into a light signal (radiograph). Single and biplane fluoroscopy systems have been used to measure 2D and 3D bone motion at various joints, including the intervertebral joints of the cervical and lumbar spine; the hip, knee and ankle in the lower limb; and the shoulder in the upper limb. Owing to the high frequency of knee injury and the prevalence of knee osteoarthritis, most fluoroscopic studies have focused on measuring bone motion at the knee joint during daily activities such as walking and stair ambulation.

Single-plane fluoroscopy has been used to measure the motion of normal and reconstructed joints, but this method is limited to 2D assessment and the results are prone to out-of-plane errors that can be relatively large. Typical accuracies associated with measuring kinematics of the intact knee in vivo using single-plane fluoroscopy are 1° for in-plane rotations, 1 to 2 mm for in-plane translations, 2° for out-of-plane rotations, and 4 to 6 mm for out-of-plane translations. Higher accuracies are possible with biplane fluoroscopic analysis, where maximum standard deviation errors of 0.29 mm for translations and 0.25° for rotations have been reported for in vivo measurement of total knee arthroplasty (TKA) joint motion and 0.73 mm and 0.76° for translations and rotations of the bones in the intact knee measured in vitro. Most fluoroscopy systems are stationary and impose restrictions on the types of activities that can be investigated. For example, fluoroscopic analyses of knee-joint kinematics during locomotion are usually performed with the subject walking or 5 running on a treadmill. It is unknown whether significant differences exist in the relative movements of the bones at the knee during treadmill and overground gait. Tracking a joint by mobilizing the X-ray tubes and image intensifiers not only enables image sequences to be captured for multiple cycles of overground gait, it also reduces motion blur by decreasing the relative movement between the target joint and the0 imaging equipment. This approach has been used to measure TKA joint motion during overground walking, where an X-ray unit was translated horizontally using a portable trolley. However, relatively large mean errors of 2 mm were reported for out-of-plane joint motion. 5 SUMMARY OF THE INVENTION

According to a first broad aspect of the present invention, there is provided an imaging system, comprising:

one or more imaging units comprising respective emitters for emitting imaging radiation and corresponding detectors for detecting the radiation after interaction of the o radiation with a target portion of a specimen located in an image capture volume of the system;

a mobile gantry for supporting the one or more imaging units, the one or more imaging units defining the image capture volume of the system and being moveable relative to the gantry (and therefore typically the ground);

5 a drive controllable to move the gantry and to move the one or more imaging units relative to the gantry, such that the one or more imaging units are controllably moveable in at least two non-parallel directions in at least one plane;

at least one tracking sensor supported by the gantry and configured to track the target portion of the specimen located in the image capture volume and to output o location data indicative of a location of the target portion; and

a controller configured to receive the location data from the tracking sensor and to control the drive based on the location data to move the gantry and the one or more imaging units relative to the gantry to track the target portion with the one or more imaging units.

5

In one embodiment, the system comprises a track for supporting the gantry, the gantry being translatable along the track, wherein the drive includes at least one horizontal drive for driving the gantry along the track.

In an embodiment, the drive comprises a plurality of servomotors. In another embodiment, the one or more imaging units are controllably moveable in two perpendicular directions in a common plane.

For example, the two perpendicular directions may include a generally upright direction.

In an embodiment, the tracking sensor comprises a camera.

In one embodiment, the tracking sensor is supported by the gantry stationary relative to at least one of the one or more imaging units.

In one embodiment, the tracking sensor is configured to detect a predefined tracking marker locatable in or on a specimen. For example, the tracking marker may comprise a reflective marker. In an embodiment, the system is configured for use with a target portion comprising a human or animal knee or other joint.

In one embodiment, the system further comprises a memory including one or more sets of stored data indicative of the evolution of the location or velocity of the target portion (such as in the form of a velocity profile), and the controller is further configured to retrieve the stored data from the memory and to control the drive based on the location data and the stored data, the stored data being used to predict the location or velocity of the target portion. It should be noted that the stored data may be derived from observations of the target portion of the specimen while within the image capture volume or within the field of view of the tracking sensor (or a combination of both).

For example, the controller may include a velocity feed-forward unit and a proportional feedback controller configured to predict the location or velocity of the target portion.

In one example, the controller is further configured to store the location data received from the tracking sensor, or velocity data derived from the location data predict the location of the target portion, in the memory for subsequent use as the stored data.

In one embodiment, the method further comprises a velocity controller intermediate between the controller and the drive.

In an embodiment, the imaging radiation is penetrative radiation with respect to the specimen and the respective emitters and detectors of the one or more imaging units are disposed on opposite sides of the image capture volume.

For example, the system may include a plurality of the imaging units with intersecting beams of imaging radiation.

In one example, the one or more imaging units comprise X-ray fluoroscopy units.

In an embodiment, the imaging radiation is reflective radiation with respect to the specimen.

In an embodiment, the gantry comprises two portions on opposite sides of the image capture volume and controllable to be driven by the drive in unison.

In an embodiment, the system is configured to image a joint of a person or animal engaged in any one or more of: i) overground walking, ii) ascending or descending one or more steps, iii) ascending or descending one or more ramps, iv) standing up from and sitting down into a chair, v) performing squatting movements beginning from a standing position, or vi) performing forward lunges beginning from a standing position.

For example, a sets of steps or a ramp may be located on (or as, or as a part of) a walkway of the system, such that a subject can be imaged while ascending and/or descending.

In an embodiment, the one or more imaging units comprise a plurality of more imaging units configured to perform biplanar imaging and the system is configured to capture and output biplane images of the target portion.

For example, the biplanar imaging of the target portion may be in two imaging planes that are perpendicular to the direction of motion of the gantry. According to a second broad aspect of the present invention, there is provided an imaging method, comprising:

locating a moving specimen in an image capture volume of one or more imaging units comprising respective emitters for emitting imaging radiation and corresponding detectors for detecting the radiation after interaction of the radiation with the target portion, the one or more imaging units being supported by a mobile gantry and being moveable relative to the gantry;

tracking the target portion with at least one tracking sensor supported by the gantry;

the tracking sensor outputting location data indicative of a location of the target portion;

receiving the location data from the tracking sensor at a controller; and the controller controlling a drive based on the location data to move the gantry and the one or more imaging units relative to the gantry to track the target portion with the one or more imaging units.

According to a third broad aspect of the present invention, there is provided an image generated using the imaging method of the second aspect. This aspect also provides biplane pairs of images generated using the imaging method of the second aspect.

It should be noted that any of the various individual features of each of the above aspects of the invention, and any of the various individual features of the embodiments described herein including in the claims, can be combined as suitable and desired.

BRIEF DESCRIPTION OF THE DRAWING

In order that the invention may be more clearly ascertained, embodiments will now be described, by way of example, with reference to the accompanying drawing, in which:

Figure 1 is a schematic view of an X-ray imaging system according to an embodiment of the present invention;

Figure 2 is a perspective view of the imaging system of figure 1;

Figure 3 is a schematic view of a gantry motion profiler of the imaging system of figure 1 ;

Figure 4 is a schematic plot of velocity versus time, showing an exemplary prediction of tracking marker velocity generated by the imaging system of figure 1;

Figure 5 is a schematic diagram illustrating the steps involved in calculating 3D joint kinematics from X-ray images acquired with the imaging system of figure 1 ; Figures 6A, 6B and 6C are photographs of experimental arrangements used to conduct experiments with the imaging system of figure 1 ;

Figure 7 shows the results of measurements of 3D knee-joint kinematics obtained for a TKA knee and an intact knee performed with the imaging system of figure 1 ;

Figure 8 is a plot of the trajectories found experimentally of the knee-joint centre for one complete cycle of overground walking occurring within the image capture volume of the imaging system of figure 1 ; and

Figure 9 presents plots of 3D knee-joint kinematics for two consecutive cycles of overground walking measured from one representative TKA patient using the imaging system of figure 1.

DETAILED DESCRIPTION OF EMBODIMENTS

An X-ray imaging system 10 according to an embodiment of the present invention is shown schematically in figure 1. System 10 comprises two X-ray generators 12, 14, which— in this embodiment— are in the form of BV Pulsera (trade mark) X-ray generators. System 10 also includes two corresponding, synchronized cameras 16, 18, in this embodiment in the form of high-speed, digital cameras (each in the form, for example, an Optronis (trade mark) CR600x2 (trade mark) video camera). System 10 also includes two image intensifiers 20, 22, to which are mounted the respective cameras 16, 18. Image intensifiers 20, 22 may be in the form of 30 cm image intensifiers, such as those provided by Philips Medical Systems.

System 10 includes a robotic gantry 24, which supports X-ray generators 12, 14, cameras 16, 18 and image intensifiers 20, 22 in fixed relative positions. System 10, in particular facilitated by its gantry drive mechanism (discussed below), is capable of translating robotic gantry 24 and hence X-ray generators 12, 14, cameras 16, 18 and image intensifiers 20, 22 over a range of movement sufficient to permit an ambulatory subject to take multiple strides of overground gait while being imaged. In the example shown in figure 1, this range of horizontal movement is 3.6 m. Similarly, sufficient vertical travel is provided to allow imaging of the relevant portion of the subject over its range of vertical movement during those strides. In the example shown in figure 1, this range of vertical movement is 0.5 m., extending upwards from approximately 25 cm above the surface on which the subjects walks.

Furthermore, system 10 provides sufficient velocity and acceleration to keep the imaged portion of the subject within image capture volume 26. In the example shown in figure 1, the gantry drive mechanism provides peak velocities of 5.0 ms ¹ horizontally and 1.2 ms ¹ vertically, and peak accelerations of 20 ms^~2 horizontally and 12 ms^~2 vertically.

It will be appreciated, however, that other ranges of movement, velocity and acceleration will be more suitable or required with other subjects, such as particular animals (e.g. dogs, horses), according to the speed of the subject (whether human or animal), which may be walking or running at a number of different speeds, and according to the range and speed of motion of the target— as some joints, for example, will move over a greater range according to subject, type of gait, etc. The skilled person will be able to adapt or modify system 10 accordingly, in view of the present description.

The gantry drive mechanism of system 10 controls the positions of X-ray generators 12, 14, cameras 16, 18 and image intensifiers 20, 22 so that a target (such as a joint of a subject) moving within the image capture volume 26 of system 10 (essentially at the intersection of those components of respective X-ray beams 28, 30 emitted by X-ray generators 12, 14 and collected by image intensifiers 20, 22) remains within image capture volume 26 for the imaging period, such as spanning multiple strides of overground walking in the example of a knee. Cameras 16, 18 may sample at a maximum rate of 1000 fps, and are interfaced with image intensifiers 20, 22 and used to capture biplane radiographic images in real- or near-real time. Thus, the two X-ray imaging planes are perpendicular to the plane of figure 1 and, respectively, to the direction of propagation of X-ray beams 28, 30, and intersect within image capture volume 26.

System 10 also includes a tracking sensor in the form of a tracking camera 32 and a controller in the form of a gantry motion profiler 34. Tracking camera 32 is mounted on robotic gantry 24 between image intensifiers 20, 22 and remains fixed relative to X- ray generators 12, 14. Tracking camera 32 may be, for example, a high-speed digital camera (such as a Basler (trade mark) piA640-210gm (trade mark) camera, which samples at a maximum rate of 210 fps). Gantry motion profiler 34 may be a computer or other computing device configured to capture and analyze images in real- or near- real time. The gantry drive mechanism of system 10 includes a velocity controller 36, servo- drives 37 and a series servomotors 38 (of which two are shown in this schematic view). Servomotors 38 drive robotic gantry 24, X-ray generators 12, 14, cameras 16, 18 and image intensifiers 20, 22 under the control of velocity controller 36 (via servo-drives 37). Tracking camera 32 is used to image a tracking marker located on the target (such as a joint). The tracking marker is a marker placed on the target to be imaged so is imagable with tracking camera 32. The tracking marker may be, for example, a skin- mounted marker placed on a subject's knee or other joint when tracking joint motion during gait.

Tracking camera 32 relays images of the tracking marker to gantry motion profiler 34, which generates 2D marker positions in the plane of motion. Gantry motion profiler 34 then sends velocity commands in real-time to velocity controller 36 that then uses servo-drives 37 and servomotors 28 accordingly to control the horizontal motion of robotic gantry 24 and the vertical motion of X-ray generators 12, 14, cameras 16, 18 and image intensifiers 20, 22 so as to maintain image capture volume 26 coincident with the tracking marker and hence the target.

System 10 includes X-ray control panels 40 that are used to activate X-ray generators 12, 14 and to set their X-ray tube voltages and currents. System 10 also includes an operator workstation 42, which may be in the form of a suitably configured computing device, and which receives and stores the biplane radiographs outputted by cameras 16, 18, and provides a Graphical User Interface (GUI) that facilitates user control of and interaction with gantry motion profiler 34, X-ray generators 12, 14, image intensifiers 20, 22, and cameras 16, 18.

Figure 2 is a perspective view of system 10, showing in particular X-ray generators 12, 14, camera 16 (camera 18 is obscured), image intensifiers 20, 22 and robotic gantry 24. Also shown in this view are further components of system 10, including and a walkway 50 for ambulatory subjects (such as patients) and parallel, horizontal gantry guides 52 for supporting robotic gantry 24 and guiding robotic gantry 24 during horizontal motion of robotic gantry 24; horizontal gantry guides 52 flank walkway 50.

Robotic gantry 24 includes first and second mobile columns 54, 56, each of which is provided with a pair of robotic arms 58, 60. (Robotic gantry 24 and robotic arms 58, 60 are termed 'robotic' because their movements are computer controlled in real-time, as is described below.) Robotic arms 58 of first mobile column 54 support X-ray generators 12, 14, respectively; robotic arms 60 of second mobile column 56 support image intensifiers 20, 22, respectively, which in turn support cameras 16, 18, respectively. First and second mobile columns 54, 56 are also provided with respective vertical guides 62, 64 (in each case on opposite sides of the respective mobile columns 54, 56), within which are supported and guided respective linearly actuated vertical mounts (not shown). The respective linearly actuated vertical mounts in turn support the respective pairs of robotic arms 58, 60 and impart vertical motion to robotic arms 58, 60 and hence to X-ray generators 12, 14, cameras 16, 18, image intensifiers 20, 22 and tracking camera 32. Tracking camera 32 is mounted on robotic gantry 24, in a fixed position relative to X-ray generators 12, 14 and image intensifiers 20, 22.

System 10 includes a pair of horizontal servomotors 66, located at a first end 68 of horizontal gantry guides 52, for imparting horizontal motion to— and controlling the horizontal motion of— robotic gantry 24. System 10 also includes a pair of vertical servomotors 70, located at the upper ends of first and second mobile columns 54, 56, for imparting vertical motion to— and controlling the vertical motion of— the vertical mounts and hence robotic arms 58, 60. (Servomotors 66, 70 collectively constitute servomotors 38.)

System 10 also includes an electrical control cabinet 72, an electrical control panel 74 and an emergency panel 76 (for immediately stopping operation of the mobile portions of system 10). Gantry motion profiler 34 and operator workstation 42 are also depicted in this figure.

In the example shown in figure 1, the overall dimensions of system 10 are 7.2 m χ 4.0 m x 1.9 m (length χ width χ height). Walkway 50 measures 7.2 m χ 2.0 m (length χ width). Each corresponding X-ray generator 12, 14, image intensifier 20, 22 and camera 16, 18, constitutes an X-ray imaging unit or fluoroscope, each of which has three lockable degrees of freedom (two horizontal translations and one rotation about a vertical axis) with respect to its linearly actuated vertical mount, which allows the configurations of the X-ray imaging unit to be altered, thereby changing the inter-beam angle. In the example presented and explained below, an inter-beam angle of -60° was used, but it will be appreciated that this can be set as desired according to application, though constrained by the desire to collect 3D images while preferably providing an obstacle-free path for ambulation.

The X-ray image capture volume 26 is thus translated both horizontally and vertically as a consequence of the configuration shown in figures 1 and 2.

In other embodiments of the invention, 2D images may be satisfactory or desirable, in which case only one X-ray imaging unit may be employed, angled to provide a beam direction relative to the direction of motion of the subject and robotic gantry 24 of any suitable value. In some cases a suitable beam direction will be approximately 90° to this direction of motion, but in others it may have a value between be 90° and 0° (or indeed 0°). When this angle is sufficiently low (including when the beam is substantially parallel to this direction of motion), a single mobile column may be employed, supporting a tracking camera, and supporting robotic arms configured to locate an X-ray generator generally before or behind the subject and an image intensifier with camera on the opposite side of the subject to that of the X-ray generator.

In other variations of the system of figures 1 and 2, one or more additional X-ray imaging units may be employed, such that there are three or more such X-ray imaging units. Accurate joint motion tracking is employed so that the target (e.g., a knee joint) remains within image capture volume 26. The size of image capture volume 26 may be varied by varying the dimensions of system 10 and the specifications of the X-ray imaging units. In the configuration shown in figures 1 and 2, image capture volume 26 had a diameter of approximately 20 cm when projected onto the plane of motion (i.e., in the sagittal plane).

Vibration of robotic gantry 24 is minimized during tracking both to protect the X-ray imaging units from damage and to minimize errors in the kinematic measurements. This is done by having velocity controller 36 control the motion of robotic arms 58, and hence X-ray generators 12, 14, image intensifiers 20, 22 and cameras 16, 18, in a manner that minimizes tracking error while suppressing external disturbances to keep vibrations of the gantry structure to a minimum. That is, unwanted vibrations that cannot otherwise be eliminated are taken into account while positioning X-ray generators 12, 14, image intensifiers 20, 22 and cameras 16, 18 dynamically. The high inertia of the X-ray imaging components (which, in a prototype of this embodiment, have a combined mass of -150 kg) and gantry structure made the task of joint motion tracking particularly challenging. Figure 3 is a schematic view of gantry motion profiler 34 of figure 1. Gantry motion profiler 34 includes a processor 80 and a memory 82. In figure 3, processor 80 is shown implementing a number of modules based on program code 84 and data stored in memory 82. Persons skilled in the art will appreciate that a number of these modules could be implemented in some other way, for example by dedicated circuits.

Referring to figure 3, processor 80 includes a velocity feed-forward unit 86, a proportional feedback controller 88, and a look ahead trajectory predictor 90.

Velocity feed-forward unit 86, proportional feedback controller 88 and look ahead trajectory predictor 90 implement, respectively, a velocity feed-forward method and a proportional feedback control method and a Look-Ahead Trajectory Prediction (LATP) method to track the motion of the target. The command ( v_cmd ) to be sent to velocity controller 36 is generated by velocity feed-forward unit 86, proportional feedback controller 88 and look ahead trajectory predictor 90 in real time and is of the form:

^cmd ^— ^marker ^ p ^camera ( ^ ) where v_marker , the feed-forward component, represents the predicted marker velocity generated by the LATP method; K is a proportional gain; and <?_camera is the camera feedback error, which was calculated as the difference between the position of the tracking marker and the position of the centre of the image capture volume in the plane of motion (e.g., sagittal plane during gait). Look ahead trajectory predictor 90 has been found to generate smooth and delay-free tracking marker velocity signals (i.e., v_marker in equation ( 1)) that compensate for the delay in the marker-camera sensor system and reduce vibrations of the gantry structure. Look ahead trajectory predictor 90 determines a velocity profile of the tracking marker and uses this information to predict the realtime tracking marker velocity, v_marker . The determined velocity profile of the tracking marker (referred to as a 'learned velocity profile' or LVP) comprises a sequence of velocities including both horizontal and vertical components of the velocity of the tracking marker. It is generated by look ahead trajectory predictor 90 based on marker velocity sequences recorded from 3 or 4 'practice trials' of the prescribed task (e.g., walking at the natural speed). During these 'practice trails', the tracking marker is tracked using a proportional control scheme (i.e., ^vmarker ⁼ 0 ^m equation (1)) that maintains the tracking marker within the relatively large image capture volume of the tracking camera (60 cm horizontal χ 40 cm vertical in the plane of motion), but not— in this embodiment— within the smaller X-ray image capture volume (20 cm diameter circle). Look ahead trajectory predictor 90 creates the LVP by first synchronizing and then averaging the tracking marker velocity sequences recorded from the practice trials, and finally smoothing the average profile with a low- pass filter. The recorded marker velocity sequences are synchronized by finding the time offset using a synchronization function:

^T _syx,k = Sync(V_l(t),V_k (t), T_iri , T_amv ) (2) where r_{sync k} is the time synchronizing offset for the k^:h practice trial; V_\ and V_k are the sequences to be synchronized, which were the tracking marker velocity sequences recorded from the 1st and k'" practice trials; r_ini is an initial guess for r_{sync i} ; and r_amp is the amplitude of the search range. The synchronization function calculates the correlation coefficient between V^t) and V_k (t + r) at ail discrete time offsets, τ , such thatr_ini - r < τ < r_ini + r . The time synchronizing offset, τ _k , is given by the value of τ that corresponds to the largest value of the correlation coefficient. The synchronized velocity sequences are then averaged:

yav( = -^-∑_i ⁿL_{1 i} (i + r_synCji) (3) where n_t is the number of practice trials and V_av is the average of the synchronized learned velocity sequences. Look ahead trajectory predictor 90 then smooths the sequence V_av to create the LVP, then stores the LVP in stored velocity profile 92 of memory 82.

In trials for which the joint was imaged with the X-ray tubes activated, look ahead trajectory predictor 90 predicted the velocity of the tracking marker ( v_marker ) at each time instant using the stored LVP. During each cycle of the real-time control algorithm, the marker velocity was calculated and stored in a FIFO (^"first-in, first-out') buffer of n_b elements. Equation (2) was then used to find the time synchronizing offset (τ _t ) between the filtered velocity sequence of n_b elements thus far collected and the LVP (see Fig. 3). Finally, the value of ^, in equation (1) for the present (i ^h) time step was found from: vmarker,i ⁼ LVP(ti); t~ 2^" _{sync> ;} + - - + ?_delay (4)

J s

where f_s is the sampling frequency of the controller, and thus ( n_b I f_s ) is the length in time of the FIFO velocity buffer of n_b elements; and i_delay is the total time delay associated with the gantry control system, which was estimated to be 280 ms for the horizontal axis and 260 ms for the vertical axis. A large portion of this time delay (-190 ms) was caused by filtering the velocity sequence stored in the FIFO buffer. Filtering was needed owing to the noise introduced by the low sampling rate and low resolution of the marker-camera sensor system (100 Hz and 1 mm, respectively).

Figure 4 is a schematic plot of velocity versus time, showing the prediction of the tracking marker velocity, ^, , performed by the LATP algorithm at the instant (i^m) time step. The learned velocity profile (LVP) is shown as a dashed curve, and the smoothed velocity sequence of n_b elements from the FIFO buffer is shown as a solid curve at 94. All variables are as defined in the text. After data collection, X-ray image processing, system calibration, and pose estimation are performed off-line to calculate 3D joint kinematics, as illustrated schematically in figure 5. Figure 5 is a schematic diagram illustrating the steps involved in calculating 3D joint kinematics from X-ray images acquired with system 10, and outlines the steps involved in calculating the relative pose between the tibial and femoral components of a TKA (total knee arthroplasty) implant. Input data consisted of X-ray images 100 of a static calibration phantom and the TKA implant 102 at each time frame during joint motion, as well as the known geometry of the implant components defined by volumetric models 104 obtained from the manufacturer. The X-ray images acquired by cameras 16, 18 are processed 106 to effect flat field correction, spatial distortion correction, and blur and noise reduction, so as to improve the quality of the images. Flat field correction is used to remove noise and artifacts caused by variations in sensitivity between pixels in the sensors of cameras 16, 18. Spatial distortion is corrected by imaging a suitable phantom and determining the required function to correct image distortion. For example, the phantom may be a plate transparent or translucent to the X-rays emitted by X-ray generators 12, 14 (such as of Perspex (trade mark)) embedded with X-ray imagable (e.g. metal) beads. In one specific example, the phantom may be a 46 cm ^χ 46 cm Perspex plate embedded with 17x 17=289 steel beads. The beads may be of any suitable size (e.g. each 1 mm in diameter), provided they are readily locatable in the X-ray image. In one example, a ninth-order polynomial may be used to correct for image distortion. De-blurring may then be applied to enhance certain features of the image (e.g., the edges of the target) and to reduce the effects of noise. X-ray images may be collected at 200 fps using a shutter speed of 1/200 s and a resolution of 1024 ^χ 1024 pixels.

System 10 is calibrated 108 by performing calibration measurements to calculate the configuration of the biplane X-ray units, that is, to determine the parameters describing the geometric configuration of the X-ray generators 12, 14 and image intensifiers 20, 22 in a global reference frame. A phantom is used for the calibration measurements, such as a nylon block containing steel beads. In one example, the phantom comprises a 85 mm ^χ 85 mm ^χ 130 mm nylon block containing 23 steel beads (each 1 mm in diameter). Images 100 of the phantom may be acquired with the two X-ray units of system 10, and parameters defining the configuration of the X-ray units (expressed in the phantom reference frame) may be determined (such as using the Projection Matrix method). Computer code, such as written in Matlab (trade mark), may be used to integrate these image processing steps and obtain the required calibration parameters, which then were used to calculate the pose of an object (e.g., TKA implant) in a global reference frame. We note that the relative positions of the X-ray units are configurable depending on the joint of interest and the activity to be investigated. An inter-beam angle of approximately 60° may be used in anticipation of the examples described below involving fluoroscopic analyses of knee-joint motion during overground walking and stair ambulation.

The processed radiographs and calibration parameters obtained from X-ray image processing 106 and system calibration 108 are used, together with the volumetric models 104, to calculate 3D target (e.g. knee-joint) kinematics at each time frame. Pose estimation 110 is performed by providing an initial guess for the pose of a subject and minimizing a scalar cost function with six independent variables. The six variables are the three components of position and three Euler angles that define the subject's pose in a global reference frame. The cost function is defined as the root-mean-square distance between the detected edges on the biplane X-ray images and the projected edges of a volumetric model of the subject [20] . An initial guess for the pose of the subject may be found using, for example, an open-source software package called JointTrack (trade mark) [28], and the optimization problem may be solved using, for example, a nonlinear programming solver in Matlab called 'fminsearch' with default options. In the example of human subject, the average time for convergence has been found to be— in one series of experiments— approximately 48 seconds per frame per TKA implant component and 216 seconds per frame per natural bone on a desktop PC (Intel (trade mark) Core 2 Quad CPU Q9400 2.67 GHz, RAM 8.00 GB). In the case of a TKA implant, reference measurements of the pose of each implant component may be obtained using metal (e.g. steel) beads embedded in the distal femur and proximal tibia. A program (such as may be developed in Matlab) may be used to locate the centres of the beads on each X-ray image, from which the 3D coordinates of each bead may be determined. The best-fitting pose for the set of beads in each bone may then be found using, for example, a least-squares method (such as that described in [29]).

The output comprises 3D target kinematics 1 12, such as three translations and three rotations of the target. In the example of a knee during walking, this output comprises three translations and three rotations of the tibia relative to the femur over one cycle of walking; the translations 114 comprise lateral tibia shift, anterior drawer and joint distraction, while the rotations 1 16 comprise flexion, adduction and external rotation. EXAMPLES

A test rig was designed and built to evaluate the accuracy with which system 10 could measure knee-joint kinematics during simulated overground walking. The test rig provided a range of motion in the horizontal direction of 2.0 m and in the vertical direction of 0.5 m.

Three sets of experiments were performed to evaluate the accuracy of 3D kinematic measurements obtained from system 10 (see Table 1). Figures 6A, 6B and 6C are photographs of an experimental arrangements 130a,b,c used in these benchmark experiments. Figure 6A is a photograph of the arrangement 130a used in experiment 1. Arrangement 130a includes a nylon plate 132 attached to a lower carriage 134 of the test rig. The inset in this figure is a close-up of nylon plate 132; the configuration of the three embedded steel beads is apparent in this insert.

Figure 6B is a photograph of the arrangement 130b used in experiment 2, including a saw-bone model 136 of a human leg with implanted TKA components 138. The femur and tibia of the model leg 136 were attached to an upper carriage 140 and lower carriage 134 of the test rig, respectively, and then actuated to simulate the stance phase of overground walking. Figure 6C is a photograph of the arrangement 130c used in experiment 3, and includes a human cadaver knee specimen 142 sealed in a plastic bag. The femur and tibia were attached to the upper carriage 140 and lower carriage 134 of the test rig and then actuated to simulate the stance phase of overground walking. In experiments 2 and 3, a tracking marker in the form of a retro-reflective marker 144 was attached at the approximate center of the knee joint to enable joint motion tracking during simulated gait. The purpose of experiment 1 was to determine the accuracy with which the position of a point in space could be measured under static and dynamic conditions. Three steel beads (each 1 mm in diameter) were embedded in a nylon plate so that the distances between the centers of the beads (hereafter referred to as 'inter-marker distances') measured 30.00±0.01 mm, 40.00±0.01 mm, and 50.00±0.01 mm (see figure 6A). Five trials were performed with the nylon plate attached to the lower carriage 134 of the test rig. In Trial 1, nylon plate 132 was held stationary at different locations within the capture volume, whereas in Trials 2 and 3 it was moved on a circular path of radius 40 mm at velocities of 0.3 ms ¹ and 0.5 ms^_1, respectively. In Trials 4 and 5, nylon plate 132 was moved along a trajectory that simulated the position and velocity of the center of the knee joint in the sagittal plane during the stance phase of walking at speeds of 0.5 ms^_1 and 0.7 ms^_1, respectively. Errors in inter-marker distances were calculated for each trial. X-ray generators 12, 14 were operated in continuous mode with tube voltages of 95 kV and 90 kV and tube currents of 6.97 mA and 6.89 mA.

TABLE 1

Conditions defining benchmark experiments

Expt. Target Trial X-ray units Prescribed motion of target

1 Steel 1 Stationary Stationary

beads 2 Stationary 0.3 ms^" circular motion

3 Stationary 0.5 ms^_1 circular motion

4 Tracking target 0.5 ms^_1 walking^†

5 Tracking target 0.7 ms^_1 walking^†

2 TKA knee Tracking target 0.7 ms^_1 walking^†

3 Intact knee Tracking target 0.7 ms^_1 walking^†

' Trajectories of the knee joint during walking were obtained by scaling gait data reported by Winter [30]

Experiment 2 determined the accuracy with which system 10 could measure the relative pose of the tibial and femoral components of a TKA knee 138 (sourced from Smith and Nephew Pty Ltd, Sydney) implanted in a saw-bone model 136 of a human leg (sourced from Pacific Research Laboratories Inc., Vashon Island, WA) (cf. figure 6B). Four steel beads (each 1 mm in diameter) were embedded in both the distal femur and proximal tibia of the saw-bone leg 136. The positions of the beads relative to the TKA components are needed to determine the accuracy with which system 10 can measure the relative positions of the implant components. A two-step procedure was followed to determine the positions of the beads relative to the TKA components. First, a laser scan (in the form of a 3D Scanner HD, NextEngine Inc., Santa Monica, CA; dimensional accuracy ±0.127 mm) of the surfaces of the bones and TKA components was used to create volumetric models of the implant and bones. Next, a CT scan (using a 5G NewTom (trade mark) CT scanner; voxel size, 0.125 mm ^χ 0.125 mm ^χ 0.125 mm) of the bones was used to create a volumetric model of the bones and beads. Both of these steps were performed because the implant components were not clearly visible on the CT scan while the beads were not clearly visible on the laser scan. The two sets of data were combined with the known geometry of the implant (defined by the volumetric models obtained from the manufacturer) to determine the positions of the beads with respect to the TKA components.

The saw-bone leg was attached to the test rig by first connecting the femoral head to upper carriage with a suitable upper or hip attachment (146 in figure 6C); the tibia was then connected to lower carriage 134 with a suitable lower or ankle attachment 148 (see figure 6B). Elastic bands that approximated the actions of the knee ligaments were used to connect the femur and tibia at the knee. Actuators of the test rig were programmed to simulate knee-joint motion during the stance phase of walking at a speed of 0.7 ms^_1. System 10, which functioned independently of the test rig, tracked and imaged the TKA components as the knee translated during the simulated stance phase of gait. X-ray generators 12, 14 were operated in continuous mode with tube voltages of 90 kV and 85 kV and tube currents of 6.89 mA and 6.81 mA. The accuracy of the measured relative pose of the TKA components at each time frame was determined by comparing the volumetric-model-based results against the bead-based results. Errors arising from system 10 were calculated and described using the Joint Coordinate System.

The purpose of Experiment 3 was to determine the accuracy with which system 10 measured the relative pose of the bones in an intact human cadaver knee. Approval for this part of the study was obtained from the Human Research Ethics Committee at the University of Melbourne. Four steel beads (each 2 mm in diameter) were embedded in both the distal femur and proximal tibia of a human cadaver knee specimen 142 (Donor: female, aged 67; body mass, 66 kg; height, 157 cm). The knee 142 was then attached to upper carriage 140 and lower carriage 134 of the test rig using the aforementioned hip and ankle attachments 146, 148 (cf. figure 6C). Volumetric models of the femur and tibia were created from CT scans (Brilliance CT 64-channel scanner, from Philips, Amsterdam; voxel size, 0.24 mm ^χ 0.24 mm ^χ 0.33 mm). The CT scans were also used to determine the positions of the steel beads in the local reference frames of the bones. Actuators of the test rig were again programmed to simulate knee-joint motion during the stance phase of walking at a speed of 0.7 ms ¹. X-ray generators 12, 14 were operated in continuous mode with tube voltages of 100 kV and tube currents of 14.4 mA. The accuracy of the measured relative pose of the femur and tibia at each time frame was determined by comparing the volumetric -model-based results against the bead-based results. Errors arising from system 10 were calculated and described using the Joint Coordinate System. To demonstrate the in vivo capability of system 10, tracking error (i.e., the distance between the tracking marker 144 and the center of the image capture volume in the plane of motion) was quantified over one gait cycle for 10 TKA patients, and also calculated 3D knee-joint kinematics for one of these patients over two consecutive gait cycles. Approval for these experiments was obtained from the Human Research Ethics Committee at the University of Melbourne and all subjects gave written informed consent. Ten unilateral TKA patients were tracked and imaged as they walked overground at their self-selected speeds (0.8 to 1.2 ms ¹). To demonstrate vertical tracking capability, one knee of a healthy young volunteer was also tracked during stair ascent at a self-selected speed of 0.5 ms^-1. For in vivo data collection, both fluoroscopes were set to 110 kV and 13.1 mA in continuous mode. Based on these settings, it was estimated that, in a typical clinical study, a volunteer would receive 0.17 mSv of radiation for 60 seconds of exposure time (e.g., 4 seconds per activity ^χ 5 activities ^χ 3 repetitions) which, according to the ARPANSA Code of Practice, represents minimal risk, as a dose of 0.2 mSv corresponds to that delivered by natural background radiation over a period of a few weeks.

RESULTS

The RMS error for inter-marker distances on nylon plate 132 was no more than 0.12 mm when the X-ray units remained stationary and the nylon plate 132 moved at speeds of up to 0.3 ms^-1 (Table 2, Trials 1-2). Translating nylon plate 132 at higher speeds increased the RMS error by a factor of 1.5 (Table 2, compare RMS errors for Trials 1-3). The maximum RMS error obtained when the X-ray units remained stationary was 0.18 mm.

TABLE 2

Mean, standard deviation (SD) and root-mean-squared (RMS) errors obtained in Experiment 1; 'Mean' represents the absolute value of the mean error for each trial. Mean SD RMS

Trial X-ray units Nylon plate

(mm) (mm) (mm)

1 Stationary Stationary 0.10 0.04 0.11

2 Stationary 0.3 ms^_1 circular motion 0.17 0.10 0.20

3 Stationary 0.5 ms^_1 circular motion 0.15 0.16 0.22

4 Tracking target 0.5 ms^_1 walking 0.15 0.06 0.16

5 Tracking target 0.7 ms^_1 walking 0.15 0.06 0.16

RMS error decreased by a factor of two when motion-tracking was implemented. When nylon plate 132 was translated to simulate the trajectory of the knee joint during walking and the gantry tracked the motion of the plate, the RMS error was below 0.1 mm and lower than in all three previous conditions when the X-ray units remained stationary (Table 2, compare Trials 1-3 with Trials 4-5).

Joint motion measurements for the TKA knee 138 were more accurate than those for the intact knee (Table 3 and figure 7). RMS errors for the TKA knee 138 ranged from 0.12 mm to 0.33 mm for joint translations and from 0.18° to 0.65° for joint rotations. The largest translational error was associated with joint distraction whereas internal- external rotation produced the largest rotational error. By comparison, RMS errors for the intact knee ranged from 0.35 mm to 0.78 mm for joint translations and from 0.30° to 0.77° for joint rotations. The largest translational and rotational errors were associated with mediolateral shift and abduction-adduction, respectively.

TABLE 3

Mean, standard deviation (SD) and root-mean-squared (RMS) errors calculated from kinematic measurements obtained for the TKA and intact knees. Joint translational and rotational errors represent displacements of the tibia relative to the femur. Positive mean values for translational errors represent lateral shift, anterior drawer and joint distraction; negative values represent medial shift, posterior drawer and joint compression. Positive mean values for rotational errors represent flexion, adduction and external rotation; negative values represent extension, abduction and internal rotation. Raw kinematics data (volumetric- model-based data and bead-based data) were filtered using a fourth-order Butterworth filter with a cut-off frequency of 20 Hz prior to error calculation.

Joint translational errors (mm)

Lateral Anterior Joint

shift drawer distraction Mean -0.10 0.03 -0.32

Experiment 2

SD 0.23 0.12 0.08

(TKA knee)

RMS 0.25 0.12 0.33

Mean 0.54 -0.18 -0.28

Experiment 3

SD 0.56 0.30 0.28

(Intact knee)

RMS 0.78 0.35 0.39

Joint rotational errors (deg)

External

Flexion Adduction rotation

Mean -0.06 -0.45 -0.23

Experiment 2

SD 0.17 0.18 0.61

(TKA knee)

RMS 0.18 0.48 0.65

Mean -0.20 0.68 0.27

Experiment 3

SD 0.23 0.37 0.54

(Intact knee)

RMS 0.30 0.77 0.61

Figure 7 shows the results of measurements of 3D knee-joint kinematics obtained for a TKA knee and an intact knee during the stance phase of simulated overground walking at a speed of 0.7 ms^-1, in the form of plots of lateral tibia shift, anterior drawer, joint distraction, flexion, adduction and external rotation. Measurements were performed using either volumetric models of the TKA components and bones or steel beads embedded directly into the bones. Data shown are from one trial of each experiment. Positive translations of the tibia relative to the femur are represented by lateral shift, anterior drawer, and joint distraction. Positive rotations of the tibia relative to the femur are represented by flexion, adduction, and external rotation.

For the human experiments, tracking error for the entire gait cycle during overground walking was generally less than 40 mm and peaked at approximately 60 mm. Tracking error for stair ascent was comparable to that for overground walking (see figure 8). Measured knee-joint kinematics for two consecutive gait cycles produced repeatable patterns for all rotations and translations except mediolateral shift, where the peak-to- peak displacement was no more than ~2 mm (see figure 9). The results of these in vivo experiments are shown in figures 8 and 9. Figure 8 a plot

150 of the trajectories of the knee-joint centre for one complete cycle of overground walking occurring within the image capture volume 152 (with a diameter of 200 mm) and projected onto the sagittal plane; the data are plotted as vertical tracking error (VTE) versus horizontal tracking error (THE). The data are shown for the 10 TKA patients walking (W) at speeds ranging from 0.8 ms^-1 to 1.2 ms ¹ (shaded lines) and for 1 young healthy subject ascending a staircase (S) (with a step height of 17 cm) at a self- selected speed of 0.5 ms^-1 (black curve 160). The various curves may be somewhat difficult to distinguish in this figure, but it is apparent that tracking errors for all subjects were generally within 40 mm and peaked at -60 mm (e.g. curves 154 and 156— subjects walking at 0.9 ms^"1, curve 158— a subject walking at 0.8 ms^"1).

Figure 9 presents plots of 3D knee-joint kinematics for two consecutive cycles of overground walking measured from one representative TKA patient. In this figure, iFS = ipsilateral foot strike, iFO = ipsilateral foot off, cFS = contralateral foot strike, and cFO = contralateral foot off.

System 10 accurately measured 3D knee-joint motion during simulated overground walking at speeds up to 0.7 ms^-1. Maximum RMS errors were 0.33 mm and 0.65°, respectively, for the relative translations and rotations of the tibial and femoral components of a TKA implant, and 0.78 mm and 0.77° for translations and rotations of the tibia relative to the femur in the intact knee. A unique feature of system 10 was the robotic gantry mechanism that enabled concurrent tracking and imaging of the joint for multiple strides of overground walking. The tracking control system performed well during the in vivo gait experiments, ensuring that the knees remained well within the image capture volume, a requirement for accurate pose estimation.

The errors in the measurements of inter-marker distances obtained from Experiment 1 are comparable to those reported by Kaptein et al. [33] . These investigators evaluated different methods for calibrating a biplane fluoroscopy system by measuring inter- marker distances on a digital caliper and comparing the results against reference data. They reported maximum mean and standard deviation errors of 0.15 mm and 0.24 mm, respectively (Table 4). By comparison, we obtained mean and standard deviation errors of 0.08 mm and 0.09 mm, respectively, for inter-marker distances when both the gantry and nylon plate remained stationary, and mean and standard deviation errors of 0.03 mm and 0.09 mm, respectively, when the gantry tracked the motion of the nylon plate at a simulated walking speed of 0.7 ms^-1 (Table 2, compare Trials 1 and 5).

TABLE 4 Comparison of literature data illustrating the accuracy of knee-joint kinematic measurements obtained from biplane x- ^■ray fluoroscopy.

Study Object Activity Frame CalcMaximum Error

Ref. Imaged Rate ulated for Joint for Joint

(fps) Error Translation Rotation

(mm) (deg)

[35] Intact knee Walking ^a 250 RMS 0.77 3.86

[20] TKA knee Lunge ^b 30 SD 0.29 0.25

[34] Intact knee Knee 30 Mean 0.24^ 0.37^FE

flexion ^c SD 0.18^ 0.91^FE

_{0 2 1}ML

[22] Intact knee Standing ^d 250 Mean 0.60^FE

SD 0.19^ 0.21^FE

RMS 0.26^ 0.85^FE

[33] Digital caliper Static ^e Mean 0.15

SD 0.24

[21] Intact knee Knee 500 Mean 0.24^ 0.17^IE

flexion ^f SD 0.73^ 0.76^IE

Present Experiment 1 Walking ^g 200 Mean 0.03

Nylon plate SD 0.09

RMS 0.09

Experiment 2 Walking ^h 200 Mean 0.32^ro 0.45^

TKA knee SD 0.23^ 0.61^IE

RMS 0.33^s3 0.65^IE

_{0 54}ML

Experiment 3 Walking ¹ 200 Mean 0.68^

Intact knee SD 0.56^ 0.54^IE

_{0 77}AA

RMS 0.78^ anterior-posterior translation; flexion-extension;

^ mediolateral translation; ^ abduction-adduction;

^JD joint distraction; ^IE internal -external rotation;

a Measurements obtained from a canine hind-limb during treadmill walking at 1.5 ms^_1; 5 ^b Forward lunge performed by TKA patients at their self-selected speed (Dr. Guoan Li, Harvard University, personal communication);

c Cadaver knee specimens were translated at 17 mms and flexed slowly (with a peak knee flexion velocity of -15 deg/s);

d In vivo static measurements obtained with subjects standing on level ground;

l o ^e Small tantalum beads mounted on a digital caliper;

f Cadaver knee specimens were swung freely at slow speeds to simulate the stance phase of walking; knee flexion angle remained nearly constant during simulated gait motion; g Nylon plate translated to simulate the trajectory of the knee joint during the stance phase of walking at a speed of 0.7 ms^_1, with a plate velocity of 0.33 ms^_1 to 1.80 ms^_1; ^h Saw-bone leg embedded with TKA components used to simulate the stance phase of walking at a speed 0.7 ms^_1;

¹ Human cadaver knee specimen used to simulate stance phase of walking at 0.7 ms^_1. The error measurements for TKA joint motion during simulated walking (Experiment

2) are comparable to in vivo data reported by Bingham and Li [20]. These researchers calculated maximum standard deviation errors of 0.29 mm and 0.25°, respectively, for implant component translations and rotations compared with the maximum standard deviations of 0.23 mm for translations and 0.61° for rotations (Table 4). Differences in the magnitude of error measured for rotation (i.e., the error associated with rotation in Bingham's study was a factor of two smaller than that obtained here) may be due to differences in the experimental protocol used in these two studies. Bingham and Li [20] measured knee kinematics during a forward lunge performed by a TKA patient whereas we measured TKA kinematics in vitro during the stance phase of overground walking. Also, the knee-joint angular velocities in Bingham's study were likely lower than those observed in the current study, which reached as high as 160 deg/s.

The errors in the present measurements of joint position for the intact knee (Experiment

3) are also consistent with those reported previously by others. Li et al. [34] acquired biplane X-ray images from two human cadaver knee specimens as each knee was moved manually into flexion and extension through its full range of motion at a maximum angular velocity of -15 deg/s. They reported maximum mean errors of 0.24 mm for joint translations and 0.37° for joint rotations with maximum standard deviations of 0.18 mm for translations and 0.91° for rotations (see Table 4). Giphart et al. [21] simulated knee-joint motion during walking by swinging cadaver knees (at a relatively slow speed with the knee flexion angle remaining nearly constant) through the image capture volume of a biplane fluoroscopy system while recording images at 500 fps. They measured maximum mean errors of 0.24 mm for translations and 0.17° for rotations with maximum standard deviations of 0.73 mm and 0.76° for translations and rotations, respectively (Table 4). By comparison, the maximum magnitudes of the mean errors of the present measurements were 0.54 mm for translations and 0.68° for rotations with maximum standard deviations of 0.56 mm and 0.54° for translations and rotations, respectively. You et al. [35] acquired biplane radiographic images at 250 fps of a canine hindlimb during treadmill walking at 1.5 ms^_1 and reported maximum RMS errors of 0.77 mm for translations and 3.86° for rotations of the intact knee (Table 4). The present maximum RMS errors for measurements performed on an intact human cadaver knee were 0.78 mm for translations and 0.77° for rotations. When the X-ray units were stationary in Experiment 1, the accuracy of the present kinematic measurements were better when nylon plate 132 was moved (Trials 2, 4 and 5) than when it was held stationary (Trial 1). This unexpected result occurred because a greater extent of the capture volume was utilised in Trial 1 than in Trials 2, 4, and 5, and also because image distorsion was greater closer to the outer edges of the images.

The accuracy of the present kinematic measurements increased when the motion of the target object was tracked, for two possible reasons. First, tracking reduced the measurement errors introduced by motion-blur artifact. In Trials 2 and 3 of Experiment 1, where the X-ray units remained stationary and the nylon plate moved through the image capture volume, relative motion between the plate and the image intensifiers caused the beads to appear blurred on the radiographs. Tracking reduced this motion- blur by reducing the relative motion between the plate and the image intensifiers. Second, tracking ensured that the plate remained near the center of the image capture volume, where image distortion was lower than at the edges. More accurate measurements of 3D joint position were obtained with the TKA implant than the intact knee. This is because the fluoroscopic images of the TKA components had well- defined edges that were more accurately detected than those of the natural bones, particularly in the condylar region. In addition, the volumetric models of the TKA components were known precisely from the manufacturer, whereas the volumetric models of the natural bones were created from CT scans and were likely less accurate.

Overall, the largest standard deviations and RMS errors occurred in the mediolateral direction for joint translations and in the internal-external direction for joint rotations (Table 3). The errors obtained in the mediolateral direction are likely to be due to inter- axis angle of the two X-ray cones, which was fixed at approximately 60°. If system 10 were configured using a larger inter-axis angle, for example 90°, then one would expect to obtain similar errors in the mediolateral and anteroposterior directions. Relatively large errors were also obtained for internal-external rotations of the knee because these movements were nearly out-of-plane for both X-ray units in the present experiments. In addition, the geometric shapes of the tibial and femoral condyles of the intact knee and that of the tibial component of the TKA knee were more symmetric about the long axes of the tibia and femur than about any other direction. Taken together, these two factors made it difficult to distinguish between small rotational differences on the basis of edge information alone. There are limitations of the present experiments. First, the in vitro experiments were performed at relatively low walking speeds. Knee-joint motion was measured at a simulated walking speed of 0.7 ms^_1, which is slightly below the range of mean self- selected walking speeds reported for TKA patients (0.8 ms ¹ to 1.1 ms ¹) [36] and approximately one-half the natural walking speed for healthy people [37]. However, the maximum speed of 0.7 ms^-1 utilized in this study was imposed by the test rig, not system 10. The servomotors 38 of system 10 are readily adapted or selected to be capable of translating the X-ray units at speeds high enough to track knee-joint motion during normal human gait.

The results of the present in vivo experiments (cf. figures 8 and 9) demonstrate that system 10 is capable of tracking and measuring knee-joint motion for walking speeds in the range of 0.8-1.2 ms^-1 in TKA patients. Because the accuracy of in vivo kinematic measurements cannot be established without inserting metal beads into the bones of living subjects, the test rig was used to simulate 3D knee-joint motion using a cadaver model.

Second, the test rig restricted horizontal movement of the knee joint to a maximum distance of 2.0 m, allowing only the stance phase of gait to be simulated. Nonetheless, the present error estimates should be valid for the entire gait cycle as peak translational velocities of the knee occur during terminal stance [30]. Furthermore, the results shown in figure 9 demonstrate that system 10 is capable of measuring 3D knee-joint kinematics over multiple strides of overground walking. Third, the error estimates given here pertain only to measurements of knee-joint motion during walking; the results of this study cannot be extrapolated to determine the accuracy of our system in measuring the 3D motion of other anatomical and replaced joints. Fourth, joint tracking performance was affected by the relatively low sampling rate and time delay associated with the marker-camera sensor system used to determine the position of the knee joint during simulated overground walking. One advantage of the marker-camera sensor system is that it avoids direct contact with the body; changes in joint motion may arise from the application of an external force when a contact sensor, such as a retractable wire transducer, is used. Unfortunately, the lower sampling rate and longer time delay associated with the marker-camera sensor system made accurate joint tracking more difficult. The LATP algorithm compensated for these deficiencies by predicting the real-time velocity of the tracking marker based on the LVP, but the success of this approach depends on the extent to which the LVP is repeatable. The relatively small tracking errors shown in figure 8 for walking and stair ambulation indicate that these activities are sufficiently repeatable for data collection with system 10.

Fifth, the effect of the contralateral leg was ignored in Experiments 2 and 3. In the present human experiments for overground walking, the contralateral leg occluded the test knee during early stance (anterior image intensifier view) and late stance (posterior image intensifier view). Although this effect was relatively small for our experiments with the unilateral TKA patients, where the implant edges were still detectable on the occluded images, it may be more pronounced for the natural knee. The configuration of the X-ray units guarantees that at any given time at least one image is free from occlusion, in which case the pose estimation will be based on a single plane if no edge information can be obtained from the occluded image. However, the availability of some edge information on the occluded image is likely, and this would significantly improve on the worst case of pose-estimating from single-plane images. The problem of occlusion is not unique to this embodiment (see [21], [33], [35]), but in the background art, occlusion has been minimized by positioning the two X-ray units (with a smaller inter-beam angle) to provide anteroposterior views [22] . For reasons of safety, however, system 10 of this embodiment advantageously keeps moving parts clear of the walkway, as apparent from figures 1 and 2.

An important advantage of tracking and imaging a joint concurrently is the increase in workspace volume that this approach yields; that is, tracking increases the volume of space in which a moving target can be imaged by both X-ray units simultaneously. For the configuration of the X-ray units used in the current experiments, the dimensions of the workspace volume were 3.6 m ^χ 0.5 m ^χ 0.2 m, enabling the kinematics of various joints such as the knee, hip, and shoulder to be studied for a wide range of activities, including overground and treadmill walking, stair ambulation, squatting, and lunging. Tracking also reduces the relative motion between the target joint and the image intensifiers, thereby reducing motion-blur artifact even at lower shutter speeds. Reduced motion blur improves the accuracy of the pose estimation calculations whereas lower shutter speeds allow for the X-ray beam to be set at lower current levels, which in turn reduces the radiation exposure for subjects. However, tracking also potentially introduces errors in the kinematic measurements that are difficult to quantify. Small relative movements between the X-ray units occur from vibrations induced in the gantry structure during tracking and from inaccuracies in the guides in which the X-ray units move. This relative motion can potentially change the imaging geometry away from the calibrated configuration and cause errors in pose estimation that are difficult to isolate and correct.

5

In summary, embodiments of the present invention provide a novel mobile imaging system, which find application— for example— in measuring dynamic joint motion at high frame rates for multiple strides of overground walking. Mean, standard deviation and RMS errors can be calculated for joint position measurements obtained, such as for i o an intact knee and a TKA implant during simulated overground walking at a speed of 0.7 ms^_1. In experiments described above, maximum RMS errors were 0.33 mm and 0.65° for translations and rotations of the TKA knee and 0.78 mm and 0.77° for translations and rotations of the intact knee. Measurement accuracy was enhanced by the ability to track and image the joint concurrently. System capability for collecting in

15 vivo data was also demonstrated for stair ambulation and level walking over ground.

Modifications within the scope of the invention may be readily effected by those skilled in the art. It is to be understood, therefore, that this invention is not limited to the particular embodiments described by way of example hereinabove.

20

In the claims that follow and in the preceding description of the invention, except where the context requires otherwise owing to express language or necessary implication, the word "comprise" or variations such as "comprises" or "comprising" is used in an inclusive sense, that is, to specify the presence of the stated features but not to preclude 25 the presence or addition of further features in various embodiments of the invention.

Further, any reference herein to prior art is not intended to imply that such prior art forms or formed a part of the common general knowledge in any country.

REFERENCES

30 [1] A. Cappozzo, F. Catani, A. Leardini, M. G. Benedetti, and U. Delia Croce,

"Position and orientation in space of bones during movement: experimental artefacts," CLIN BIOMECH, vol. 1 1, no. 2, pp. 90-100, Jul, 1996.

[2] M. Akbarshahi et al. , "Non-invasive assessment of soft-tissue artifact and its effect on knee joint kinematics during functional activity," / BIOMECH, vol. 43, no. 7,

35 pp. 1292-1301, Jan, 2010. [3] J. D. Auerbach, B. P. Wills, T. C. Mcintosh, and R. A. Balderston, "Evaluation of spinal kinematics following lumbar total disc replacement and circumferential fusion using in vivo fluoroscopy," Spine, vol. 32, no. 5, pp. 527-536, Mar, 2007.

[4] W. J. Anderst, W. F. Donaldson Iii, J. Y. Lee, and J. D. Kang, "Continuous cervical spine kinematics during in vivo dynamic flexion-extension," SPINE J, vol. 14, no.

7, pp. 1221-1227, Aug, 2014.

[5] H. Tanino et al., "An In Vivo Model for Intraoperative Assessment of Impingement and Dislocation in Total Hip Arthroplasty," / ARTHROPLASTY, vol. 23, no. 5, pp. 714-720, Jul, 2008.

[6] J. Stiehl, R. Komistek, D. Dennis, R. Paxson, and W. Hoff, "Fluoroscopic analysis of kinematics after posterior-cruciate-retaining knee arthroplasty," / BONE JOINT SURG, vol. 77-B, no. 6, pp. 884-889, Nov, 1995.

[7] S. A. Banks, and W. A. Hodge, "Accurate measurement of three-dimensional knee replacement kinematics using single-plane fluoroscopy," IEEE T BIO-MED ENG, vol. 43, no. 6, pp. 638-649, Jun, 1996.

[8] S. Tashman, and W. Anderst, "In-Vivo Measurement of Dynamic Joint Motion Using High Speed Biplane Radiography and CT: Application to Canine ACL Deficiency," J BIOMECH ENG, vol. 125, no. 2, pp. 238-245, Apr, 2003.

[9] R. D. Komistek, D. A. Dennis, and M. Mahfouz, "In Vivo Fluoroscopic Analysis of the Normal Human Knee," CLIN ORTHOP RELAT R, vol. 410, pp. 69-81, May,

2003.

[10] S. Tashman, P. Kolowich, D. Collon, K. Anderson, and W. Anderst, "Dynamic Function of the ACL-reconstructed Knee during Running," CLIN ORTHOP RELAT R, vol. 454, pp. 66-73, Jan, 2007.

[11] J. W. Fernandez, M. Akbarshahi, H. J. Kim, and M. G. Pandy, "Integrating modelling, motion capture and x-ray fluoroscopy to investigate patellofemoral function during dynamic activity," COMPUT METHOD BIOMEC, vol. 11, no. 1, pp. 41-53, Feb, 2008.

[12] M. Kozanek et al., "Tibiofemoral kinematics and condylar motion during the stance phase of gait," / BIOMECH, vol. 42, no. 12, pp. 1877-1884, May, 2009.

[13] C. A. Myers et al., "In Vivo Tibiofemoral Kinematics During 4 Functional Tasks of Increasing Demand Using Biplane Fluoroscopy," AM J SPORT MED, vol. 40, no. 1, pp. 170-178, Jan, 2012.

[14] M. Akbarshahi, J. W. Fernandez, A. G. Schache, and M. G. Pandy, "Subject- specific evaluation of patellofemoral joint biomechanics during functional activity,"

MED ENG PHYS, vol. 36, no. 9, pp. 1122-1133, Jun, 2014. [15] S. Yamaguchi, T. Sasho, H. Kato, Y. Kuroyanagi, and S. A. Banks, "Ankle and

Subtalar Kinematics during Dorsiflexion-Plantarflexion Activities," FOOT ANKLE

INT, vol. 30, no. 4, pp. 361-366, Apr, 2009.

[16] N. Nishinaka et al., "Determination of in vivo glenohumeral translation using fluoroscopy and shape-matching techniques," / SHOULDER ELB SURG, vol. 17, no. 2, pp. 319-322, May, 2008.

[17] Z. Zhu, D. F. Massimini, G. Wang, J. J. P. Warner, and G. Li, "The accuracy and repeatability of an automatic 2D-3D fluoroscopic image-model registration technique for determining shoulder joint kinematics," MED ENG PHYS, vol. 34, no. 9, pp. 1303-1309, Dec, 2012.

[18] J. W. Fernandez, and M. G. Pandy, "Integrating modelling and experiments to assess dynamic musculoskeletal function in humans," EXP PHYSIOL, vol. 91, no.

2, pp. 371-382, Mar, 2006.

[19] D. C. Ackland, F. Keynejad, and M. G. Pandy, "Future trends in the use of X- ray fluoroscopy for the measurement and modelling of joint motion," P I MECH

ENG H, vol. 225, no. 12, pp. 1 136-1 148, Dec, 201 1.

[20] J. Bingham, and G. Li, "An Optimized Image Matching Method for

Determining In-Vivo TKA Kinematics with a Dual-Orthogonal Fluoroscopic

Imaging System," J BIOMECH ENG, vol. 128, no. 4, pp. 588-595, Jan, 2006.

[21] J. E. Giphart, C. A. Zirker, C. A. Myers, W. W. Pennington, and R. F. LaPrade,

"Accuracy of a contour-based biplane fluoroscopy technique for tracking knee joint kinematics of different speeds," / BIOMECH, vol. 45, no. 16, pp. 2935-2938, Aug,

2012.

[22] W. Anderst, R. Zauel, J. Bishop, E. Demps, and S. Tashman, "Validation of three-dimensional model-based tibio-femoral tracking during running," MED ENG

PHYS, vol. 31, no. 1, pp. 10-16, Mar, 2009.

[23] M. S. Zihlmann et al., "Three-dimensional kinematics and kinetics of total knee arthroplasty during level walking using single plane video-fluoroscopy and force plates: A pilot study," GAIT POSTURE, vol. 24, no. 4, pp. 475-481, Dec, 2006.

[24] R. List et al. , "Three -Dimensional Kinematics of an Unconstrained Ankle Arthroplasty: A Preliminary In Vivo Videofluoroscopic Feasibility Study," FOOT ANKLE INT, vol. 33, no. 10, pp. 883-892, Oct, 2012.

[25] E. H. Garling, B. L. Kaptein, K. Geleijns, R. G. H. H. Nelissen, and E. R.

Valstar, "Marker Configuration Model-Based Roentgen Fluoroscopic Analysis," / BIOMECH, vol. 38, no. 4, pp. 893-901, Apr, 2004. [26] K. Dabov, A. Foi, V. Katkovnik, and K. Egiazarian, "Image restoration by sparse 3D transform-domain collaborative filtering," in Proc. SPIE, 2008, pp. 681207-681207-12.

[27] A. Verri, and E. Trucco, Introductory Techniques for 3-D computer vision, Upper Saddle River, NJ, USA: Prentice Hall, 1998.

[28] S. Mu, "JointTrack: An Open-source, Easily Expandable Program for Skeletal Kinematic Measurement Using Model-image Registration," M.S. thesis, University of Florida, Gainesville, FL, USA, 2007.

[29] I. Soderkvist, and P. -A. Wedin, "Determining the movements of the skeleton using well-configured markers," J BIOMECH, vol. 26, no. 12, pp. 1473-1477, Mar,

1993.

[30] D. A. Winter, "Kinematic, kinetic, and energy data," in Biomechanics and motor control of human movement, 4th ed. Hoboken, NJ, USA: Wiley, 2009, APPENDIX A, pp. 296-360.

[31] E. S. Grood, and W. J. Suntay, "A Joint Coordinate System for the Clinical

Description of Three-Dimensional Motions: Application to the Knee," J BIOMECH

ENG, vol. 105, no. 2, pp. 136-144, May, 1983.

[32] Exposure of humans to ionizing radiation for research purposes, ARPANSA,

Yallambie, VIC, 2005, pp. 36.

[33] B. L. Kaptein, K. B. Shelburne, M. R. Tony, and J. Erik Giphart, "A comparison of calibration methods for stereo fluoroscopic imaging systems," /

BIOMECH, vol. 44, no. 13, pp. 251 1-2515, Jul, 2011.

[34] G. Li, S. K. Van de Velde, and J. T. Bingham, "Validation of a non-invasive fluoroscopic imaging technique for the measurement of dynamic knee joint motion," J BIOMECH, vol. 41, no. 7, pp. 1616-1622, Jan, 2008.

[35] B. M. You, P. Siy, W. Anderst, and S. Tashman, "In vivo measurement of 3-D skeletal kinematics from sequences of biplane radiographs: Application to knee kinematics," IEEE T MED IMAGING, vol. 20, no. 6, pp. 514-525, Jun, 2001.

[36] J. A. McClelland, K. E. Webster, J. A. Feller, and H. B. Menz, "Knee kinematics during walking at different speeds in people who have undergone total knee replacement," KNEE, vol. 18, no. 3, pp. 151-155, Apr, 201 1.

[37] Y. P. Lim, Y.-C. Lin, and M. G. Pandy, "Muscle function during gait is invariant to age when walking speed is controlled," GAIT POSTURE, vol. 38, no. 2, pp. 253-259, Nov, 2013.

Claims

CLAIMS:

1. An imaging system, comprising:

one or more imaging units comprising respective emitters for emitting imaging radiation and corresponding detectors for detecting the radiation after interaction of the radiation with a target portion of a specimen located in an image capture volume of the system;

a mobile gantry for supporting the one or more imaging units, the one or more imaging units defining the image capture volume of the system and being moveable relative to the gantry;

a drive controllable to move the gantry and to move the one or more imaging units relative to the gantry, such that the one or more imaging units are controllably moveable in at least two non-parallel directions in at least one plane;

at least one tracking sensor supported by the gantry and configured to track the target portion of the specimen located in the image capture volume and to output location data indicative of a location of the target portion; and

2. A system as claimed in claim 1, comprising a track for supporting the gantry, the gantry being translatable along the track, wherein the drive includes at least one horizontal drive for driving the gantry along the track.

3. A system as claimed in claim 1, wherein the drive comprises a plurality of servomotors.

4. A system as claimed in claim 1, wherein the one or more imaging units are controllably moveable in two perpendicular directions in a common plane.

5. A system as claimed in claim 4, wherein the two perpendicular directions include a generally upright direction.

6. A system as claimed in claim 1, wherein the tracking sensor comprises a camera.

7. A system as claimed in claim 1, wherein the tracking sensor is supported by the gantry stationary relative to at least one of the one or more imaging units.

8. A system as claimed in claim 1, wherein the tracking sensor is configured to detect a predefined tracking marker locatable in or on a specimen.

5 9. A system as claimed in claim 8, wherein the tracking marker comprises a reflective marker.

10. A system as claimed in claim 1, configured for use with a target portion comprising a human or animal knee or other joint.

0

11. A system as claimed in claim 1, further comprising a memory including one or more sets of stored data indicative of the evolution of the location or velocity of the target portion, wherein the controller is further configured to retrieve the stored data from the memory and to control the drive based on the location data and the stored data,5 the stored data being used to predict the location or velocity of the target portion.

12. A system as claimed in claim 11, wherein the controller includes a velocity feedforward unit and a proportional feedback controller configured to predict the location or velocity of the target portion.

0

13. A system as claimed in claim 11, wherein the controller is further configured to store the location data received from the tracking sensor, or velocity data derived from the location data predict the location of the target portion, in the memory for subsequent use as the stored data.

5

14. A system as claimed in claim 1, further comprising a velocity controller intermediate between the controller and the drive.

15. A system as claimed in claim 1, wherein the imaging radiation is penetrative o radiation with respect to the specimen and the respective emitters and detectors of the one or more imaging units are disposed on opposite sides of the image capture volume.

16. A system as claimed in claim 15, including a plurality of the imaging units with intersecting beams of imaging radiation.

5

17. A system as claimed in claim 15, wherein the one or more imaging units comprise X-ray fluoroscopy units.

18 A system as claimed in claim 1, wherein the imaging radiation is reflective radiation with respect to the specimen.

19. A system as claimed in claim 1, wherein the gantry comprises two portions on opposite sides of the image capture volume and controllable to be driven by the drive in unison.

20. A system as claimed in claim 1, configured to image a joint of a person or animal engaged in any one or more of: i) overground walking, ii) ascending or descending one or more steps, iii) ascending or descending one or more ramps, iv) standing up from and sitting down into a chair, v) performing squatting movements beginning from a standing position, or vi) performing forward lunges beginning from a standing position.

21. A system as claimed in claim 1, wherein the one or more imaging units comprise a plurality of more imaging units configured to perform biplanar imaging and the system is configured to capture and output biplane images of the target portion.

22. A system as claimed in claim 21, wherein the biplanar imaging of the target portion is in two imaging planes that are perpendicular to the direction of motion of the gantry.

23. An imaging method, comprising:

24. An image generated using the imaging method of claim 23.