Precision and accuracy of consumer-grade motion

0 downloads 0 Views 2MB Size Report
pedicle screw placement in pediatric spinal fusion surgery. Andrew Chana ...... Spine 2012;37:E473–8. doi: 10.1097/BRS.0b013e318238bbd9 . [24] Sakai Y ...
ARTICLE IN PRESS

JID: JJBE

[m5G;June 7, 2017;19:54]

Medical Engineering and Physics 0 0 0 (2017) 1–11

Contents lists available at ScienceDirect

Medical Engineering and Physics journal homepage: www.elsevier.com/locate/medengphy

Precision and accuracy of consumer-grade motion tracking system for pedicle screw placement in pediatric spinal fusion surgery Andrew Chan a, Janelle Aguillon b, Doug Hill c,d, Edmond Lou a,c,d,∗ a

Department of Department of c Department of d Alberta Health b

Biomedical Engineering, University of Alberta, Edmonton, Alberta T6G 2V2, Canada Chemical and Materials Engineering, University of Alberta, Edmonton, Alberta T6G 1H9, Canada Surgery, University of Alberta, Edmonton, Alberta T6G 2B7, Canada Services – Glenrose Rehabilitation Hospital, Edmonton, Alberta T5G 0B7, Canada

a r t i c l e

i n f o

Article history: Received 21 November 2016 Revised 12 April 2017 Accepted 16 May 2017 Available online xxx Keywords: Adolescent idiopathic scoliosis Spinal surgery Motion capture Image guidance Pedicle screws Intraoperative imaging Abbreviations: AIS, Adolescent Idiopathic Scoliosis 3D, Three Dimensional CT, Computed Tomography RMS, Root-Mean-Square

a b s t r a c t Adolescent idiopathic scoliosis (AIS) is a 3-dimensional spinal deformity involving lateral curvature and axial rotation. Surgical intervention involves insertion of pedicle screws into the spine, requiring accuracies of 1 mm and 5° in translation and rotation to prevent neural and vascular complications. While commercial CT-navigation is available, the significant cost, bulk and radiation dose hinders their use in AIS surgery. The objective of this study was to evaluate a commercial-grade Optitrack Prime 13W motion capture cameras to determine if they can achieve adequate accuracy for screw insertion guidance in AIS. Static precision, camera and tracked rigid body configurations, translational and rotational accuracy were investigated. A 1-h camera warm-up time was required to achieve precisions of 0.13 mm and 0.10°. A three-camera system configuration with cameras at equal height but staggered depth achieved the best accuracy. A triangular rigid body with 7.9 mm markers had superior accuracy. The translational accuracy for motions up to 150 mm was 0.25 mm while rotational accuracy was 4.9° for rotations in two directions from 0° to 70°. Required translational and rotational accuracies were achieved using this motion capture system as well as being comparable to surgical-grade navigators. © 2017 IPEM. Published by Elsevier Ltd. All rights reserved.

1. Introduction Adolescent idiopathic scoliosis (AIS) is a spinal deformity characterized by lateral curvature, often combined with vertebral rotation. It has an overall prevalence of 0.47%–5.2%, with a higher prevalence and severity in girls than in boys [1]. Surgery is recommended for patients with curvatures greater than 50° who have not yet reached skeletal maturity or have rapid curve progression and functional impairment [2]. Posterior fusion surgery utilizing instrumentation has become the gold standard of surgical treatment. Segmental pedicle screws are often used to attach the rods to the vertebral body [3]. Accuracy in insertion pedicle screws is critical to prevent complications including injury to nerve roots, spinal cord or vascular structures, pedicle fracture, and instrumentation failure [4,5]. Prior studies have found pedicle dimensions to be 4–18 mm in the

∗ Corresponding author. Present address: 6 -110F, Clinical Science Building, 8440 – 112 Street, Edmonton, Alberta T6G 2B7, Canada. E-mail address: [email protected] (E. Lou).

transverse direction and 4–14 mm in the sagittal direction, with T4–T8 having the smallest pedicles [6]. Given the narrow width of the pedicles in thoracic levels, a 1 mm error could easily result in a breach of the pedicle. To maximize accuracy, standardized free hand insertion techniques as well as fluoroscopic-based guidance and computed tomography (CT) navigation have been developed [7,8]. The free-hand method involves insertion of pedicle screws based on visible anatomical landmarks and the tactile feedback, relying heavily on surgeon experience and correct identification of the anatomical landmarks. Fluoroscopy is often used to confirm screw placement when free-hand methods are used and can also be used to guide screw placement. CT navigation uses mobile intraoperative CT systems, to allow for 3D reconstruction of bony anatomy, alongside motion capture cameras to localize surgical tools relative to the bony anatomy [7,9]. CT navigation has been used extensively in spinal fusion for pediatric spinal deformities to good success with breach rates of less than 10% [10–12]. Specifically, for adolescent idiopathic scoliosis, breach rates for free-hand methods have been reported to range from 0.1% to 66.8% [13–16] while for fluoroscopy and CT range

http://dx.doi.org/10.1016/j.medengphy.2017.05.003 1350-4533/© 2017 IPEM. Published by Elsevier Ltd. All rights reserved.

Please cite this article as: A. Chan et al., Precision and accuracy of consumer-grade motion tracking system for pedicle screw placement in pediatric spinal fusion surgery, Medical Engineering and Physics (2017), http://dx.doi.org/10.1016/j.medengphy.2017.05.003

JID: JJBE 2

ARTICLE IN PRESS

[m5G;June 7, 2017;19:54]

A. Chan et al. / Medical Engineering and Physics 000 (2017) 1–11

from 0–50.7% and 7.5–7.9%, respectively [17–21]. In addition to individual studies, four head-to-head studies have shown superiority of CT image guidance over free-hand methods in reducing breach rates [22–25]. However, free-hand methods are still often preferred to usage of navigation systems in adolescent idiopathic scoliosis procedures. Radiation exposure remains a concern for both image guidance methods in the pediatric population [26–29]. Past studies suggest an increased lifetime risk of fatal cancer of 1 in 10 0 0 from abdominal CT scans [30]. Furthermore, in the case of intraoperative CT scans, screw insertional time is increased to allow for registration between image and vertebral landmarks [31]. Lastly, the additional cost and bulk in using intraoperative CT scans remains a barrier to usage of current navigation technologies though this may be mitigated from fewer reoperations or complications [8]. Usage of handheld ultrasound-CT registration for spinal fusion appears to be a promising method of providing image guidance without the bulk, radiation exposure and cost of conventional systems [32]. Further study into the adaptability of commercial-grade mid-range motion capture cameras could broaden the usability of ultrasound-CT registration in the operating room, particularly for adolescent idiopathic patients. Motion capture systems have been used extensively in gait analysis, providing accuracy of within 1% error for 120 mm translations [33]. Motion capture cameras have light emitting diodes that emit light into a capture area. Markers in the capture area reflect the light back to the camera which then determines the size and location of the marker [34]. Using multiple markers and cameras, the motion capture system is then able to triangulate the actual position of markers in 3D space. However, the accuracy of motion capture depends heavily on both quality of cameras and range of space in which cameras are used [35,36]. The required theoretical accuracy of navigation systems was found to be less than 1 mm and 5° by Rampersaud et al. when considering screw trajectories in 3D space [37]. While good accuracy has been achieved by commercially available navigation systems, the question remains whether more inexpensive and flexible motion capture camera systems can be used in navigation with equivalent accuracy. The goal of this study is to determine the usability of off-theshelf motion capture equipment for usage in the operating room and eventual integration with ultrasound-CT navigation system for spinal fusion in adolescent idiopathic scoliosis. The objectives of this study are to evaluate Optitrack Prime 13W motion capture cameras for three attributes: (1) to investigate the static precision of the motion capture system, (2) to determine the optimal camera and marker configuration for highest tracking accuracy and (3) to evaluate the translational and rotational accuracy of the system for varying movement magnitudes.

above the operating space. The positions of cameras for evaluating camera configuration are described under Camera and Rigid Body Configuration Testing. The tracking software, Motive from the manufacturer (Tracker v. 1.10.0, NaturalPoint, United States), was used to obtain motion tracking data. The cameras can record data over a period of time to exported for further processing, but can also continuously output live positional and orientation data for active tracking in separate software. Calibration of the Optitrack system was completed with the Optitrack CW-250 Calibration wand and the Optitrack CS200 Calibration square for setting the origin. The built-in calibration wanding process was used to calibrate the system which involved moving the calibration wand throughout the entire capture volume for 30 s. At the end of calibration, the software displays the accuracy of the current calibration ranging with six ranks ‘Poor’ to ‘Exceptional’. All calibrations that were used met the ‘Exceptional’ ranking, resulting in software-estimated errors of less than 0.15 mm. Cameras were recalibrated either every day prior to testing, or after camera placement was changed. Optitrack 7.9 mm markers were placed on Optitrack M3 9 mm bases. The study evaluated the accuracy and precision of the builtin rigid body recognition system, which is able to create a rigid body from three or more markers that are mounted on a single object. Position and rotation values of the center of the rigid body were exported as XYZ translational coordinates and Pitch, Yaw and Roll rotational angles using the XYZ Euler rotation sequence. 2.2. Static testing

2. Methods

The precision of the cameras was determined using two experiments: ten-minute trials and six-hour trial. Precision was defined as the 95% confidence interval of the standard deviation of positional or rotational data. A right triangle with dimensions of 90 mm x 120 mm x 150 mm was created with a 3D printer (Makerbot Replicator 2X, United States) to hold the markers in a stationary position (Fig. 2). The ten-minute trials involved continuous recording of the position and orientation at 120 FPS for 10 min. To determine if cameras required a heat-up time, two sets of trials were completed: the first with cameras recording data within five minutes of turning on from ambient room temperature (20 °C) and the second with cameras being pre-heated for an hour prior to recording with the mean difference from origin to compare the two trials. Each cold-start test was started on a different day with the pre-heat test completed on the same day, to ensure full cool-down of the cameras. The six-hour trial involved obtaining positional and rotational data every five minutes, acquiring data at 120 frames per second over two seconds to mimic the duration of a long spinal surgery. Cameras and markers were not moved between each of the three six hour trials over three days.

2.1. Camera specifications and software configuration

2.3. Camera and rigid body configuration testing

Optitrack Prime 13W motion capture cameras (Prime 13W, NaturalPoint, United States) were selected specifically due to their wide 82° x 70° field of view and small size at 69 × 69 × 22 mm. The capture rate was 120 frames per second and used 850 nm infrared light to minimize interference from overhead lights. A schematic of camera positions relative to the capture volume is shown in Fig. 1. Cameras were mounted on tripods and placed on one side of the required capture volume with dimensions of 0.8 m deep, 0.6 m wide and 0.6 m high. Cameras were then placed 0.8– 1.2 m vertically from the floor of the capture volume and a horizontal distance of 1.0–1.2 m from the closest face of the capture volume. This setup was based on measurements in an operating room where cameras would be placed at the head of the patient

To determine the optimal camera configuration, multiple camera positions were evaluated with combinations of three or four cameras. Cameras at aligned or staggered at heights to a range of ±150 mm, and cameras aligned or at staggered depths to a range of ±200 mm as shown in Fig. 1 were tested. An object with known dimensions, a rigid body with three adjustable arms to mount markers, was created with a 3D printer (Objet30 Pro, Stratasys, United States) and then attached onto a digital caliper (Mitutoyo, Japan) (Fig. 3a) was used to assess accuracy. The accuracy of the caliper was within 0.01 mm. The rigid body was translated ten times by 40 mm for each of the eight camera configurations, with the most accurate camera positions being selected for further evaluating rigid body configurations.

Please cite this article as: A. Chan et al., Precision and accuracy of consumer-grade motion tracking system for pedicle screw placement in pediatric spinal fusion surgery, Medical Engineering and Physics (2017), http://dx.doi.org/10.1016/j.medengphy.2017.05.003

JID: JJBE

ARTICLE IN PRESS

[m5G;June 7, 2017;19:54]

A. Chan et al. / Medical Engineering and Physics 000 (2017) 1–11

3

Fig. 1. Schematic of camera position relative to capture volume. Blue (translucent) region shows capture volume. For camera configuration testing, depth was varied by 200 mm in either direction and 150 mm in either direction while overall width was kept constant at 600 mm. X, Y and Z translational directions and pitch, roll, yaw rotational directions displayed. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 2. 3D printed right triangle used for holding markers in place for static testing.

Rigid body testing involved comparing the triangular marker configuration on the standard 7.9 mm markers (standard rigid body) and the 6.4 mm markers (small markers) as well as a linear configuration (thin rigid body) with 7.9 mm markers. Lastly, a custom 96 mm calibration wand was compared to the standard Optitrack CW-250 wand to determine if use of a smaller calibration wand with a comparable size to the rigid body would affect accuracy. Ten translations of 40 mm were performed for each trial, completed over a single day.

Fig. 3. (a) Rigid body mounted on a digital caliper for translational testing, (b) Mounting of rigid body on a three-directional protractor for rotational testing.

2.4. Motion testing Translation magnitude was tested with the rigid body mounted on a digital caliper and translated three times by 10 mm, 20 mm, 40 mm, 80 mm and 150 mm in X, Y and Z directions separately. The magnitude of these movements was based on the

translations required to move across the dimensions of the lumbar vertebrae of a standardized phantom spine model at approximately 30 × 40 × 40 mm. Testing was completed over a single day. Rotational testing involved determining the accuracy of rotation values with varying magnitudes in different directions.

Please cite this article as: A. Chan et al., Precision and accuracy of consumer-grade motion tracking system for pedicle screw placement in pediatric spinal fusion surgery, Medical Engineering and Physics (2017), http://dx.doi.org/10.1016/j.medengphy.2017.05.003

ARTICLE IN PRESS

JID: JJBE 4

[m5G;June 7, 2017;19:54]

A. Chan et al. / Medical Engineering and Physics 000 (2017) 1–11 Table 1 Rotational variables tested in motion capture system. Variable changed

Description of variable

Single direction

30° and −60° rotations about each axis individually. 30°, 0° and 0° 0°, 30° and 0° 0°, 0° and 30° 60°, 0° and 0° 0°, 60° and 0° 0°, 0° and 60°

Three directions

Combinations of three rotations along three axes: −30°, 45° and −60° 45°, −30° and −60° −60°, 45° and −30° −30°, −60° and 45° 45°, −60° and −30° −60°, −30° and 45°

Small angles

Combinations of three rotations along three axes: 5°, 10° and 5° 10°, 10° and 5° 5°, 5° and 10° 10°, 5° and 10°

Large angles

Combinations of two rotations along three axes: 60°, 60° and 0° 60°, 0° and 60° 0°, 60° and 60° 65°, 65° and 0° 65°, 0° and 65° 0°, 65° and 65° 70°, 70° and 0° 70°, 0° and 70° 0°, 70° and 70° 75°, 75° and 0° 75°, 0° and 75° 0°, 75° and 75°

Because the Optitrack system uses the X, Y, Z Euler angle sequence, the rotational directions were altered from translational directions to match this convention, with the vertical ‘yaw’ direction set as the X direction, the long axis ‘roll’ along the cameras set as the Y direction and the short axis ‘pitch’ along the cameras set as the Z direction (refer to Fig. 2). Yaw was measured with a digital protractor (Model 1702, General Tools & Instruments, United States) with a listed precision of 0.3°. The roll and pitch were measured with 3D-printed analog protractors with a precision of 0.5°. A custom protractor built from Lego (Lego, Denmark) combined with custom 3D-printed components using VeroWhitePlus (Objet30 Pro, Stratasys, United States) was built to provide rotations in each direction as shown in Fig. 3b. Potential deflection of the rotation apparatus was calculated to be less than 0.03 mm and precision of block was found to be within 0.05 mm, with an overall potential error of 0.05° and deemed to be adequate for this application. To test the accuracy of rotation, a set of rotation variables were tested which is summarized in Table 1. Testing was completed over a single day. With each variable, each combination was tested three times. Angle values were selected to cover the range of motions of pedicle screw placement.

3. Results 3.1. Static testing Fig. 4a presents the mean difference from origin of rotation and position values for the ten-minute trials at 0.25 mm and 0.06°, respectively from cold start, and 0.01 mm and 0.01° with a one hour preheat. Fig. 4b compares the mean difference for the six hour trial, with the first hour compared to the origin at time zero, and hours two to six compared to the position and rotation values at the end of hour one. Positional and rotational mean difference within the first hour were 1.78 mm and 0.09° while from hour two and six, were 0.07 mm and 0.03°. The 95% confidence interval from hour two to six was 0.10 mm and 0.07°. A six-hour timeline of static position and rotation from the initial position is shown in Fig. 5. A decrease in position of more than 2 mm was noted in the Y direction, while a decrease of more than 0.2 mm was found in both X and Z directions. Rotational precision was more consistent over six hours, varying between −0.17° and 0.06°. All subsequent testing including camera configuration, rigid body configuration, translational and rotational testing were completed with a one hour pre-heat period.

2.5. Statistical analysis Accuracy for both translational and rotational tests was calculated as a root-mean-square (RMS) error using the equation:



RMS Accuracy =

n=1 N

(XN − Xo )2 N

(1)

In Eq. (1), XN represents the position value being compared to the theoretical caliper measured value Xo while N is the number of samples taken. The confidence interval of the mean value of the position was calculated as:

s Con f idence Interval o f Mean V alue = X ± t √ N

(2)

In Eq. (2), X is the mean value of the position, s is the standard deviation of the sampled position values, N is the number of position values and t is the corresponding t value for that number of position values at a 95% confidence interval. Specifically for rotations, each grouping of errors was combined to find an overall mean RMS accuracy and confidence interval for each group of rotations: Single pitch, single roll, single yaw, multiple angles at 30–60°, multiple angles at 5–10°, multiple angles 60–65° and multiple angles 70–75°.

3.2. Camera and rigid body configuration testing Fig. 6 compares the position RMS accuracy and standard error between the eight configurations. The configuration with best accuracy used three cameras with staggered height and aligned depth at 0.13 mm while the poorest used four cameras with aligned heights and staggered depth at 0.53 mm. The configuration with best repeatability used three cameras with staggered heights and depths of cameras at 0.03 mm while the poorest used four cameras with staggered height and aligned depth at 0.12 mm. On average, three cameras were superior to four cameras at 0.109 mm vs 0.190 mm, staggered height superior to aligned height at 0.143 mm vs 0.156 mm, and aligned depth superior to staggered depth at 0.122 mm vs 0.177 mm. Fig. 7 compares the positional RMS accuracy and standard error of the standard triangular rigid body on both 7.9 mm and 6.4 mm markers, the linear rigid body with standard 7.9 mm markers, and the custom 120 mm wand for calibration. The RMS accuracy was 0.21 mm for the standard rigid body, 0.21 mm for smaller markers, 0.51 mm for thin rigid body and 0.28 mm for custom wand. The standard error was 0.11 mm for standard, 0.11 mm for small markers, 0.09 mm for thin rigid body and 0.20 mm for custom wand.

Please cite this article as: A. Chan et al., Precision and accuracy of consumer-grade motion tracking system for pedicle screw placement in pediatric spinal fusion surgery, Medical Engineering and Physics (2017), http://dx.doi.org/10.1016/j.medengphy.2017.05.003

JID: JJBE

ARTICLE IN PRESS

[m5G;June 7, 2017;19:54]

A. Chan et al. / Medical Engineering and Physics 000 (2017) 1–11

5

Fig. 4. Mean difference of rotation or position from origin comparing 10-min cold start vs hot start and first hour of cold start vs hours 2–6 from same trial.

3.3. Motion testing Fig. 8 displays the RMS accuracy and 95% standard error of 10–150 mm movements in each direction. The RMS accuracy when translating 10 mm, 20 mm, 40 mm, 80 mm and 150 mm was 0.13 mm 0.14 mm, 0.24 mm, 0.12 mm and 0.21 mm while the standard errors were 0.09 mm, 0.152 mm, 0.127 mm, 0.135 mm and 0.135 mm, respectively. Fig. 9 compares the rotational RMS accuracy and standard error of varying magnitudes and directions of rotation. The accuracy from single rotations was 1.7°, three rotations less than 60° at 3.8°, two rotations less than 10° at 1.71°, two rotations between 60° and 65° at 4.9° and two rotations greater than 70° at 6.7°. The standard error was less than 0.05° for all rotational tests. 4. Discussion 4.1. Technical considerations The motion capture cameras collect spatial information at 120 frames per second. While a lower framerate was tested, the markers tracked more consistently at the maximum framerate. This study found that over ten minutes, the mean difference from origin greatly improved with a one hour pre-heat period at 0.25 mm and 0.06° compared with 0.01 mm and 0.01°. Similarly, the six hour tests showed a mean difference comparing the first hour from a cold start compared with hour two to six at 0.77 mm and 0.04°

compared with 0.03 mm and 0.02°. These results show the importance of a pre-heat period when using these cameras, ideally over one hour to maximize the static accuracy of the cameras. The 95% confidence interval deviation over hours two to six at 0.06 mm and 0.04° show that with a preheat period, the recorded values are extremely precise for off-the-shelf motion capture systems. Focusing on the 2 mm deviation in the Y-direction, it was surprising to see the position consistently decrease with each test, despite no movement of cameras or markers between each static test. Cameras were locked in placed and the rigid body was taped securely to the rigid capture surface. Slight but gradual movement of the cameras was considered, but any downward camera motion due to gravity would have shifted entire frame of reference downwards, resulting in a relative upward motion of markers, which was not found in any of the tests. Considering Fig. 8, the 40 mm translation in the Z direction had the largest error which goes against the increasing-error trends in the X and Y directions, and the decreasing error trends in the Z direction. After repeating the test ten times, the same error was found. However, it is reassuring to see that despite the large error, it is still less than 0.25 mm. Camera configuration also had surprising results, with three cameras superior to four cameras in both accuracy and standard error. While it was expected that increasing the number of overlapping fields was would improve accuracy, it is possible that three overlapping perspectives is already adequate to maximize accuracy in those fields. Adding a redundant image did not contribute new information and may have hampered the accuracy due to

Please cite this article as: A. Chan et al., Precision and accuracy of consumer-grade motion tracking system for pedicle screw placement in pediatric spinal fusion surgery, Medical Engineering and Physics (2017), http://dx.doi.org/10.1016/j.medengphy.2017.05.003

JID: JJBE 6

ARTICLE IN PRESS

[m5G;June 7, 2017;19:54]

A. Chan et al. / Medical Engineering and Physics 000 (2017) 1–11

Fig. 5. Positional and rotational value deviation from initial value over six hours, sampled every five minutes from cold-start.

Fig. 6. Positional RMS accuracy (bar chart) and 95% confidence interval of standard error from the actual mean (diamond with error bars) comparing eight camera configurations. three to four cameras, aligned or staggered heights, aligned or staggered depths, in X, Z and Y directions. Diamond represents the standard error from the actual mean value while bar chart represents root mean square error.

Please cite this article as: A. Chan et al., Precision and accuracy of consumer-grade motion tracking system for pedicle screw placement in pediatric spinal fusion surgery, Medical Engineering and Physics (2017), http://dx.doi.org/10.1016/j.medengphy.2017.05.003

JID: JJBE

ARTICLE IN PRESS

[m5G;June 7, 2017;19:54]

A. Chan et al. / Medical Engineering and Physics 000 (2017) 1–11

7

Fig. 7. Positional RMS accuracy (bar chart) and 95% confidence interval of standard error from the actual mean (diamond with error bars) from 7.9 mm marker rigid body compared with 6.4 mm markers, linear rigid body and custom 120 mm wand for calibration. Diamond represents the standard error from the actual mean value while bar chart represents root mean square error.

Fig. 8. Positional RMS accuracy (bar chart) and 95% confidence interval of standard error from the actual mean (diamond with error bars) at 10 mm, 20 mm, 40 mm, 80 mm and 150 mm in X, Z and Y directions. Diamond represents the standard error from the actual mean value while bar chart represents root mean square error.

Please cite this article as: A. Chan et al., Precision and accuracy of consumer-grade motion tracking system for pedicle screw placement in pediatric spinal fusion surgery, Medical Engineering and Physics (2017), http://dx.doi.org/10.1016/j.medengphy.2017.05.003

JID: JJBE 8

ARTICLE IN PRESS

[m5G;June 7, 2017;19:54]

A. Chan et al. / Medical Engineering and Physics 000 (2017) 1–11

Fig. 9. Rotational RMS accuracy (bar chart) and 95% confidence interval of standard error from the actual mean (diamond with error bars) comparing rotation in only X, Y and Z directions, three angle combinations at 30°, 45° and 60°, two angle combinations at 60° degrees and two angle combinations at >70°. Error bars are too small to be visualized on the figure.

imprecise matching of camera properties such as gain, focal length and spectral response. Considering previous research, Eichelberger et al. compared six, eight and ten cameras, finding that eight cameras to be superior to six, but ten being no different from eight cameras [38]. Windolf et al. compared a three camera setups in arbitrary positions with two four camera setups, showing it to be superior to one test and inferior to the other [36]. These studies showed that increasing number cameras might not always improve accuracy, depending on additional camera configuration factors. Camera heights staggered being superior to aligned heights, and aligned depths superior to staggered height. The variation in height ranged ±150 mm while depth variation ranged from ±200 mm. The three-camera system configured with staggered height and aligned depth was selected as the superior configuration for the rest of the study. It is important to note that the potential for occlusion of cameras in the operating room may require more cameras to be used. A custom, portable motion capture mounting frame, as well as 3D printed motion capture marker adapters for surgical tools are both currently in development for occlusion testing in the operating room. As surgeons typically stand on either side of the operating table, the frame is designed to mount cameras above the head of the patient to minimize potential occlusion of camera views. The standard rigid body has superior accuracy to the linear rigid body and smaller markers and custom wand. The superiority of the standard rigid body was expected since having markers along two dimensions of the three-marker plane would provide more spatial information for tracking than a linear orientation. In particular, the Z direction has a significantly worse accuracy, likely due to the linear rigid body being aligned to the Z direction during movement. The 7.9 mm markers would be more easily tracked and visible than

6.4 mm markers. Windolf et al. similarly found that larger reflective markers yielded improved accuracy [36]. Translational accuracy and error was superior in the X direction, while accuracy was poorer in the Z and error was poorer in the Y directions, though overall accuracy and error both remained below 0.10 mm, regardless of direction. Accuracy travelling 40 mm in the Z direction appeared to be an outlier. The X direction is orthogonal to all three cameras, allowing for the greatest redundancy in motion capture. The Y and Z directions however are at an angle with respect to the three cameras with motion that is largely parallel to the camera’s orientation. From these translational results, the RMS accuracy of the system was deemed as 0.25 mm, equivalent to the worst RMS accuracy from the translational tests. The standard error of the measurements was 0.1 mm which was less than the static precision of 0.15 mm. The standard error was greater than the ten-minute static precision value of 0.07 mm over a 95% confidence interval, which is expected because the tests involved moving the rigid bodies. Regarding rotation, the worst accuracy was for angles greater than 70° at 6.9°. The standard error was small relative to these angles with the worst error at 0.05° for angles greater than 70°. Angles greater than 70° are considered too inaccurate given the poorest accuracy value of 9.1° compared with the required accuracy of 5.0°. However, it is expected that screw trajectories will be less than 45° during screw insertion. In an adult cadaver study by Chung et al., the maximum required transverse and sagittal screw angles ranged from 9.4 to 29.5° and 5.2 to 25.4°, respectively [39]. Pedicle angles in a skeletal study of pediatrics patients by Zindrick et al. found transverse pedicle angle ranges from −4.2 to 35.3° and sagittal angles from 1.8 to 23.3° [6]. It is intended that the angle of pedicle screw insertion will be measured relative to the position of

Please cite this article as: A. Chan et al., Precision and accuracy of consumer-grade motion tracking system for pedicle screw placement in pediatric spinal fusion surgery, Medical Engineering and Physics (2017), http://dx.doi.org/10.1016/j.medengphy.2017.05.003

JID: JJBE

ARTICLE IN PRESS

[m5G;June 7, 2017;19:54]

A. Chan et al. / Medical Engineering and Physics 000 (2017) 1–11

an ultrasound device directed at the posterior surface of each vertebra. As a result, the range of angles of screw insertion will be the same as the transverse and sagittal pedicle angles reported in the two studies above. Given the maximum angles of 23.3° and 35.3° the 3.9° accuracy level achieved in rotations less than 60° meet the required accuracies presented at the beginning of this study. However, the ‘screw-in’ motion may require further evaluation as the tool will rotate a full 360° during screw placement. Limitations of this study can be grouped into sources of error in camera placement, and measurement sources of error. Cameras were mounted on tripods which provided adequate stability for the duration of the study. However, the relative positions of the cameras to each other were not precisely measured, but were instead placed within the range of distances shown in Fig. 1. Similarly, while translations and rotations of the rigid body were done near the origin of the capture volume, the location of the tests were not precisely measured. Translations and rotations were done in two directions to minimize hysteresis. Given the final setting of the motion capture system being in an operating room, these conditions were deemed adequate and realistic to evaluate accuracy, though further testing in the operating room is planned for future study. Focusing on measurement errors, the digital caliper was not independently recalibrated prior to usage, though they were found to be within 0.01 mm after experimentation. The accuracy of the three-directional protractor was 0.3° for the digital protractor and 0.5° for the analog protractor which would affect repeatability and accuracy of measurements. Considering that errors for rotations between 60° and 65° had an accuracy of 4.9°, the resolution of these measurements devices may preclude usage of motion capture even at 60°. To ensure that motions were in the correct direction, the digital caliper and three-directional protractor was aligned to the calibration square when defining X, Y and Z directions. To further minimize error, the magnitude of translations in any direction was defined as the magnitude of translation overall, not only in the direction of movement. However, the magnitude of rotation in any direction was not taken as the overall magnitude of rotation, potentially underestimating the actual accuracy of orientation measurements.

4.2. Clinical considerations The aim of this study was to determine if commercial-grade motion capture systems had adequate accuracy for surgical application. The lower cost and greater flexibility in using multiple cameras at different areas in the operating room make these systems attractive to use for navigation systems, particularly in combination with novel imaging methods including intraoperative ultrasound. However, their accuracy needs to be comparable to current surgical grade navigators. In the surgical field, current engineering measurement error of surgical optical navigators has been shown to range from 0.2–1.9 mm, whether for placement of surgical instruments or measurements of landmarks [40–42]. The most precise evaluation of motion capture was completed by Koivukangas et al. on the StealthStation S7 (Medtronic, Louisville, CO, USA) system was tested on an industrially verified phantom, resulting in an accuracy of 0.2 ± 0.1 mm for up to 120 mm translations. A head phantom was also tested using the Polaris Vicra (NDI, Waterloo, Canada) navigation camera resulting in errors of 0.3–0.6 mm for registration. Holloway et al evaluated the O-arm for both measurement error and accuracy of lead positioning in deep brain stimulation procedures, finding an accuracy of 0.7 mm and 2.1 mm, respectively. Navigation in human cadavers and in patients have found registration errors of 0.5–3 mm [43,44].

9

Comparing to current results from the Optitrack 13W system, these accuracies are comparable with the reported studies. However, it is important to note that in the cadaver and patient studies, the registration error is reported, not the actual positional error of screw placements. In the case of Papadopoulos et al., the mean registration error as estimated by the computer is reported, while Holly et al. used a more accurate measure by calculating the Euclidean distance of a stereotactic probe at the apparent probe position in the patient space. However, both of these depend on the cameras to measure error instead of an external validating device. A strength of this study is the usage of an external device, whether digital calipers or 3D protractor, to determine accuracy of position and rotation. Still, translation of these engineering measures directly into screw trajectories remains a shortcoming of this current study. Further in vivo studies have compared intra-operative predicted screw trajectories with post-operative screw placement. Oertel et al. found that predicted screw trajectories differed from postoperative trajectories by 2.8° [45]. Scheufler et al. had similar findings, with trajectories differing by less than 2° in 98% of patients [46]. Because these were in vivo studies, usage of post-operative CT to confirm screw position is likely the best method to validate placement, though these studies did not consider positional accuracies. From this study, accuracies ranged from 1.7° to 3.8° for rotations less than 60° and 4.9° for rotations between 60° and 65°. While numerical accuracies are inferior in this commercial system, further comparison of desired and final trajectories of screw placements would be needed to ensure comparability. It is also important to note that this study reports the error in three dimensions while current in vivo studies compared axial scans of screws on single planes, potentially underestimating the actual angular error. Still, the desired standard of 5° was achieved for rotations less than 65°. Focusing on the final application, accuracy of pedicle screw placement has typically been presented as a breach rate, with breaches defined as either a magnitude in millimetres, or according to the proportion of the screw penetrating the cortex of the pedicle [15,21]. Both medio-lateral translation and transverse trajectory can affect the extent of a breach. However, what is not captured by reporting a breach rate is the cause of the breach, whether it is medial translation of the entire screw, or a laterally translated screw that is angled medially into the spinal canal. The previously presented approximate screw trajectories at each spinal level are commonly used as a guide to place screws, but are rarely presented as a reference for accurate screw placement. A recent study by Guha et al. suggested that clinical accuracy and engineering accuracy are often not well correlated and recommended that absolute navigation accuracy be reported for true evaluation and navigation system accuracy [47]. For this study, the engineering accuracies relative to a static flat surface, have been confirmed to be adequate for screw placement. However, further study into evaluating accuracies relative to anatomical transverse and sagittal planes needs to be completed. Ultrasound imaging of individual vertebrae to provide these reference planes will be evaluated in the future. 5. Conclusion The Optitrack Prime 13W system was evaluated for its accuracy in usage for pedicle screw placement in spinal surgery. The system exceeded translational accuracy requirements at 0.25 mm compared to the required 1 mm standard. However, rotational accuracy met the 5° requirement only in rotations less than 65° at 4.9°. A three-camera configuration with each camera aligned at the same depth while having varying heights to cover the camera area was selected as the configuration to be used in the operating

Please cite this article as: A. Chan et al., Precision and accuracy of consumer-grade motion tracking system for pedicle screw placement in pediatric spinal fusion surgery, Medical Engineering and Physics (2017), http://dx.doi.org/10.1016/j.medengphy.2017.05.003

JID: JJBE 10

ARTICLE IN PRESS

[m5G;June 7, 2017;19:54]

A. Chan et al. / Medical Engineering and Physics 000 (2017) 1–11

theatre. The system also has comparable accuracy to conventional CT-navigation systems, showing that off-the-shelf mid-range motion capture has similar capabilities to current navigation technologies. Further study in generating clinically relevant measures including screw placement and breaches need to be used to further validate usage of these systems. Funding sources This work was supported by Alberta Spine Foundation, Alberta Innovates: Technology Futures, Natural Sciences and Engineering Research Council of Canada. Conflict of interest No conflict of interest. References [1] Konieczny MR, Senyurt H, Krauspe R. Epidemiology of adolescent idiopathic scoliosis. J Child Orthop 2013;7:3–9. doi:10.1007/s11832- 012- 0457- 4. [2] Weiss H-R, Moramarco M. Indication for surgical treatment in patients with adolescent idiopathic scoliosis – a critical appraisal. Patient Saf Surg 2013;7:17. doi:10.1186/1754- 9493- 7- 17. [3] Maruyama T, Takeshita K. Surgery for idiopathic scoliosis: currently applied techniques. Clin Med Pediatr 2009;3:39–44. [4] Helenius I. Anterior surgery for adolescent idiopathic scoliosis. J Child Orthop 2013;7:63–8. doi:10.1007/s11832- 012- 0467- 2. [5] Li G, Lv G, Passias P, Kozanek M, Metkar US, Liu Z, et al. Complications associated with thoracic pedicle screws in spinal deformity. Eur Spine J 2010;19:1576–84. doi:10.10 07/s0 0586- 010- 1316- y. [6] Zindrick MR, Knight GW, Sartori MJ, Carnevale TJ, Patwardhan AG, Lorenz MA. Pedicle morphology of the immature thoracolumbar spine. Spine 20 0 0;25:2726–35. [7] Puvanesarajah V, Liauw JA, Lo S, Lina IA, Witham TF. Techniques and accuracy of thoracolumbar pedicle screw placement. World J Orthop 2014;5:112– 23. doi:10.5312/wjo.v5.i2.112. [8] Helm PA, Teichman R, Hartmann SL, Simon D. Spinal navigation and imaging: history, trends, and future. IEEE Trans Med Imaging 2015;34:1738–46. doi:10. 1109/TMI.2015.2391200. [9] Allam Y, Silbermann J, Riese F, Greiner-Perth R. Computer tomography assessment of pedicle screw placement in thoracic spine: comparison between free hand and a generic 3D-based navigation techniques. Eur Spine J 2013;22:648– 53. doi:10.10 07/s0 0586- 012- 2505- 7. [10] Larson AN, Polly DW, Guidera KJ, Mielke CH, Santos ERG, Ledonio CGT, et al. The accuracy of navigation and 3D image-guided placement for the placement of pedicle screws in congenital spine deformity. J Pediatr Orthop 2012;32:e23– 9. doi:10.1097/BPO.0b013e318263a39e. [11] Larson AN, Santos ERG, Polly DW, Ledonio CGT, Sembrano JN, Mielke CH, et al. Pediatric pedicle screw placement using intraoperative computed tomography and 3-dimensional image-guided navigation. Spine 2012;37:E188– 94. doi:10.1097/BRS.0b013e31822a2e0a. [12] Jin M, Liu Z, Qiu Y, Yan H, Han X, Zhu Z. Incidence and risk factors for the misplacement of pedicle screws in scoliosis surgery assisted by O-arm navigation—analysis of a large series of one thousand, one hundred and forty five screws. Int Orthop 2016:1–8. doi:10.10 07/s0 0264- 016- 3353-6. [13] Ovadia D, Korn A, Fishkin M, Steinberg DM, Wientroub S, Ofiram E. The contribution of an electronic conductivity device to the safety of pedicle screw insertion in scoliosis surgery. Spine 2011;36:E1314–21. doi:10.1097/BRS. 0b013e31822a82ec. [14] Dede O, Ward WT, Bosch P, Bowles AJ, Roach JW. Using the freehand pedicle screw placement technique in adolescent idiopathic scoliosis surgery: what is the incidence of neurological symptoms secondary to misplaced screws? Spine 2014;39:286–90. doi:10.1097/BRS.0 0 0 0 0 0 0 0 0 0 0 0 0127. [15] Abul-Kasim K, Ohlin A. The rate of screw misplacement in segmental pedicle screw fixation in adolescent idiopathic scoliosis. Acta Orthop 2011;82:50–5. doi:10.3109/17453674.2010.548032. [16] Fiorillo P, Demonti HH. Postoperative radiographic control of instrumentation with thoracic pedicle screws in adolescent idiopathic scoliosis. Coluna/Columna 2013;12:282–4. doi:10.1590/S1808-185120130 0 040 0 0 03. [17] Chen J, Yang C, Ran B, Wang Y, Wang C, Zhu X, et al. Correction of Lenke 5 adolescent idiopathic scoliosis using pedicle screw instrumentation: does implant density influence the correction? Spine 2013;38:E946–51. doi:10.1097/ BRS.0b013e318297bfd4. [18] Upendra BN, Meena D, Chowdhury B, Ahmad A, Jayaswal A. Outcome-based classification for assessment of thoracic pedicular screw placement. Spine 2008;33:384–90. doi:10.1097/BRS.0b013e3181646ba1. [19] Abul-Kasim K, Strömbeck A, Ohlin A, Maly P, Sundgren PC. Reliability of lowradiation dose CT in the assessment of screw placement after posterior scoliosis surgery, evaluated with a new grading system. Spine 2009;34:941–8. doi:10.1097/BRS.0b013e31819b22a4.

[20] Takahashi J, Hirabayashi H, Hashidate H, Ogihara N, Kato H. Accuracy of multilevel registration in image-guided pedicle screw insertion for adolescent idiopathic scoliosis. Spine 2010;35:347–52. doi:10.1097/BRS. 0b013e3181b77f0a. [21] Akazawa T, Kotani T, Sakuma T, Minami S, Tsukamoto S, Ishige M. Evaluation of pedicle screw placement by pedicle channel grade in adolescent idiopathic scoliosis: should we challenge narrow pedicles? J Orthop Sci Off J Jpn Orthop Assoc 2015;20:818–22. doi:10.10 07/s0 0776- 015- 0746- 0. [22] Su P, Zhang W, Peng Y, Liang A, Du K, Huang D. Use of computed tomographic reconstruction to establish the ideal entry point for pedicle screws in idiopathic scoliosis. Eur Spine J 2012;21:23–30. doi:10.10 07/s0 0586- 011- 1962- 8. [23] Ughwanogho E, Patel NM, Baldwin KD, Sampson NR, Flynn JM. Computed tomography-guided navigation of thoracic pedicle screws for adolescent idiopathic scoliosis results in more accurate placement and less screw removal. Spine 2012;37:E473–8. doi:10.1097/BRS.0b013e318238bbd9. [24] Sakai Y, Matsuyama Y, Nakamura H, Katayama Y, Imagama S, Ito Z, et al. Segmental pedicle screwing for idiopathic scoliosis using computerassisted surgery. J Spinal Disord Tech 2008;21:181–6. doi:10.1097/BSD. 0b013e318074d388. [25] Vissarionov S, Schroeder JE, Novikov SN, Kokyshin D, Belanchikov S, Kaplan L. The utility of 3-dimensional-navigation in the surgical treatment of children with idiopathic scoliosis. Spine Deform 2014;2:270–5. doi:10.1016/j.jspd.2014. 03.004. [26] Frush DP, Applegate K. Computed tomography and radiation: understanding the issues. J Am Coll Radiol JACR 2004;1:113–19. doi:10.1016/j.jacr.2003.11.012. [27] Pearce MS, Salotti JA, Little MP, McHugh K, Lee C, Kim KP, et al. Radiation exposure from CT scans in childhood and subsequent risk of leukaemia and brain tumours: a retrospective cohort study. Lancet 2012;380:499–505. doi:10. 1016/S0140- 6736(12)60815- 0. [28] Ul Haque M., Shufflebarger H.L., O’Brien M., Macagno A. Radiation exposure during pedicle screw placement in adolescent idiopathic scoliosis: is fluoroscopy safe? Spine 2006;31:2516–20. doi:10.1097/01.brs.0 0 0 0238675.91612.2f. [29] Patel AA, Whang PG, Vaccaro AR. Overview of computer-assisted image-guided surgery of the spine. Semin Spine Surg 2008;20:186–94. doi:10.1053/j.semss. 20 08.06.0 05. [30] Brenner DJ. Estimating cancer risks from pediatric CT: going from the qualitative to the quantitative. Pediatr Radiol 2002;32:228–44. doi:10.1007/ s0 0247-0 02-0671-1. [31] Shin M-H, Ryu K-S, Park C-K. Accuracy and safety in pedicle screw placement in the thoracic and lumbar spines : comparison study between conventional C-arm fluoroscopy and navigation coupled with O-Arm® guided methods. J Korean Neurosurg Soc 2012;52:204–9. doi:10.3340/jkns.2012.52.3.204. [32] Yan CXB, Goulet B, Pelletier J, Chen SJ-S, Tampieri D, Collins DL. Towards accurate, robust and practical ultrasound-CT registration of vertebrae for imageguided spine surgery. Int J Comput Assist Radiol Surg 2011;6:523–37. doi:10. 1007/s11548-010-0536-2. [33] Thewlis D, Bishop C, Daniell N, Paul G. Next generation low-cost motion capture systems can provide comparable spatial accuracy to high-end systems. J Appl Biomech 2013;29:112–17. [34] Richards JG. The measurement of human motion: a comparison of commercially available systems. Hum Mov Sci 1999;18:589–602. doi:10.1016/ S0167-9457(99)0 0 023-8. [35] Feng Y, Max L. Accuracy and precision of a custom Camera-based system for 2D and 3-D Motion tracking during speech and nonspeech motor tasks. J Speech Lang Hear Res 2014;57:426–43813. doi:10.1044/2014_JSLHR- S- 13- 0 0 07. [36] Windolf M, Götzen N, Morlock M. Systematic accuracy and precision analysis of video motion capturing systems–exemplified on the Vicon-460 system. J Biomech 2008;41:2776–80. doi:10.1016/j.jbiomech.2008.06.024. [37] Rampersaud YR, Simon DA, Foley KT. Accuracy requirements for imageguided spinal pedicle screw placement. Spine 2001;26:352–9. doi:10.1097/ 0 0 0 07632-20 0102150-0 0 01. [38] Eichelberger P, Ferraro M, Minder U, Denton T, Blasimann A, Krause F, et al. Analysis of accuracy in optical motion capture – a protocol for laboratory setup evaluation. J Biomech 2016;49:2085–8. doi:10.1016/j.jbiomech.2016.05.007. [39] Chung KJ, Suh SW, Desai S, Song HR. Ideal entry point for the thoracic pedicle screw during the free hand technique. Int Orthop 2008;32:657–62. doi:10.10 07/s0 0264-0 07- 0363- 4. [40] Holloway K, Docef A. A quantitative assessment of the accuracy and reliability of O-arm images for deep brain stimulation surgery. Neurosurgery 2013;72:47– 57. doi:10.1227/NEU.0b013e318273a090. [41] Koivukangas T, Katisko JP, Koivukangas JP. Technical accuracy of optical and the electromagnetic tracking systems. SpringerPlus 2013;2. doi:10.1186/ 2193- 1801- 2- 90. [42] Wittmann W, Wenger T, Loewe E, Lueth TC. Official measurement protocol and accuracy results for an optical surgical navigation system (NPU). In: Proceedings of the 2011 annual international conference of the IEEE engineering in medicine and biology society; 2011. p. 1237–40. doi:10.1109/IEMBS.2011. 6090291. [43] Holly LT, Bloch O, Johnson JP. Evaluation of registration techniques for spinal image guidance. J Neurosurg Spine 20 06;4:323–8. doi:10.3171/spi.20 06.4.4. 323. [44] Papadopoulos EC, Girardi FP, Sama A, Sandhu HS, Cammisa FP. Accuracy of single-time, multilevel registration in image-guided spinal surgery. Spine J Off J North Am Spine Soc 2005;5:263–7 discussion 268. doi:10.1016/j.spinee.2004. 10.048.

Please cite this article as: A. Chan et al., Precision and accuracy of consumer-grade motion tracking system for pedicle screw placement in pediatric spinal fusion surgery, Medical Engineering and Physics (2017), http://dx.doi.org/10.1016/j.medengphy.2017.05.003

JID: JJBE

ARTICLE IN PRESS

[m5G;June 7, 2017;19:54]

A. Chan et al. / Medical Engineering and Physics 000 (2017) 1–11 [45] Oertel MF, Hobart J, Stein M, Schreiber V, Scharbrodt W. Clinical and methodological precision of spinal navigation assisted by 3D intraoperative O-arm radiographic imaging. J Neurosurg Spine 2011;14:532–6. doi:10.3171/2010.10. SPINE091032. [46] Scheufler K-M, Franke J, Eckardt A, Dohmen H. Accuracy of image-guided pedicle screw placement using intraoperative computed tomography-based navigation with automated referencing. Part II: Thoracolumbar Spine. Neurosurgery 2011;69:1307–16. doi:10.1227/NEU.0b013e31822ba190.

11

[47] Guha D, Jakubovic R, Gupta S, Alotaibi NM, Cadotte D, da Costa LB, et al. Spinal intraoperative three-dimensional navigation: correlation between clinical and absolute engineering accuracy. Spine J Off J North Am Spine Soc 2016. doi:10. 1016/j.spinee.2016.10.020.

Please cite this article as: A. Chan et al., Precision and accuracy of consumer-grade motion tracking system for pedicle screw placement in pediatric spinal fusion surgery, Medical Engineering and Physics (2017), http://dx.doi.org/10.1016/j.medengphy.2017.05.003