CyberKnife with Synchrony (Accuray, Sunnyvale, CA) uses a combination of x-ray imaging and optical tracking in its motion tracking system (1, 2). X-rays are taken periodically during treatment to determine the location of internal fiducials or known anatomical structures; this gives the location of the internal tumor target. Continuous optical tracking of the patient?s skin gives the real-time position of the skin; this is an external marker. Synchrony uses these two data sources to compute a mapping function between the position of the external markers and the position of the internal target. Thus, the motion of the external marker estimates the location of the internal target. By knowing the motion of the internal target, the CyberKnife robot arm can steer the radiation beam to follow that motion.

The goal of this study was to quantitatively measure the ability of the CyberKnife system to track respiratory motion. To accomplish this goal, we used a robotic motion platform to simulate respiratory motion. We also used an optical tracking system to continuously measure the position of both the CyberKnife linear accelerator (linac) and the tumor target mounted on the motion platform. By tracking both of these objects at the same time, we were able to directly measure the precision with which the CyberKnife could follow the moving target.

Materials and Methods

Treatment Planning

The treatment target is a plastic ball-cube (Accuray, Sunnyvale, CA) measuring 3 inches (76.2 mm) on a side and contains embedded gold fiducials that are visible to the CyberKnife x-ray tracking system. Treatment planning was based on a static CT scan of the cube acquired using standard parameters for CyberKnife radiosurgery. The internal spherical target of the ball-cube was treated to 3000 cGy at 100% isodose line using dynamic SRS (stereotactic radiosurgery) treatment with Synchrony respiratory motion compensation. Beams in the treatment plan were distributed evenly in space so that the entire hemisphere of possible treatment beams would be sampled in each treatment.

Respiratory Motion Simulation

Respiratory motion was generated using a computer-controlled 3D motion simulator (Figure 1), which has been described previously (3). The simulator consists of two separate platforms, a tumor motion simulator (TMS) and a skin motion simulator (SMS) which can operate independently. The TMS has two linear slides for motion in the left/right (L/R) and superior/inferior (S/I) directions and a linear actuator for motion in the anterior/posterior (A/P) direction, whereas the SMS uses three linear slides. Having two separate platforms allows us to decouple internal tumor motion (which is primarily in the S/I direction) from external skin motion (which is primarily in the A/P direction) and introduce features such as phase shifts into the simulated motion. The design of the simulator is compact and allows a wide range of simulated motion, yet does not obstruct the x-ray imagers used by the CyberKnife. The axes are controlled by servo motors with high performance optical encoders. All axes of the simulator are independently controlled from a single board (Galil Motion Control, Rocklin, CA) and can be programmed with arbitrary motion paths by downloading a series of moves to the controller. Motion commands are issued to the motors every 64 or 128 ms, depending on the frequency content of the commanded path.



Figure 1: Georgetown University Respiratory Motion Simulator. On the left is the tumor motion simulator (TMS) and on the right is the skin motion simulator (SMS).

We tested the CyberKnife using motion patterns derived from recordings of previously treated CyberKnife patients. These patients were under treatment for thoracic or abdominal tumors. Data for these respiratory motion patterns were collected under an IRB-approved protocol. These paths are denoted as patient A, patient B, and patient C.

Optical Tracking Measurements

We used the Optotrak Certus (Northern Digital, Waterloo, Canada) optical tracking system for measurements in this study. The Optotrak has a manufacturer-stated absolute position accuracy of 0.1 mm (0.15 mm in z-axis) and a position resolution of 0.01 mm. The markers used in this system are small infrared LEDs that are actively pulsed by the Optotrak controller, which avoids some of the problems such as missing or ?phantom? markers that may occur with tracking passive optical markers. Two LEDs are attached to the tumor motion simulator near the target ball-cube. An additional six LEDs are attached to the final collimator of the CyberKnife linac. The six LEDs are arrayed radially around the collimator as shown in Figure 2, which maximized the likelihood that at least one of the LEDs would always be visible, no matter what the orientation of the linear accelerator was.


Figure 2: Arrangement of Optotrak active LEDs on the final CyberKnife collimator. LEDs face radially outward and are glued to a ring that clamps onto the collimator. Control wires for the LEDs are run along the linac housing and the robot arm to a controller box on the floor of the room.

The Optotrak is placed in the treatment room in the vertical orientation so that the view of the simulator is roughly a right-superior oblique (Figure 3) and the central axis of the Optotrak Certus points toward the room isocenter. This setup allows us to continuously and simultaneously monitor the positions of both the treatment target and the CyberKnife linear accelerator in a common coordinate system. Optotrak measurements are taken at a rate of 30 Hz.


Figure 3: Photograph showing setup of the Optotrak Certus (arrow) in the CyberKnife treatment room. The Optotrak Certus is elevated off the floor so that all three of its internal cameras have a clear line of sight to the room isocenter.

Data Analysis

The data stream from the Optotrak contains 3D (x,y,z) position readings for each LED at each time point (roughly every 33 ms) over the entire treatment path. However, Synchrony only moves the linac while the treatment beam is on, so we selected those times out of the data stream. First, we removed the times when the linac was moving between beam positions -- these times are easily identified because the motion of the CyberKnife is smooth and covers several cm. Next, we searched the remaining data for times when the linear accelerator exhibited periodic or quasi-periodic motion, which can also be visually identified.

While Synchrony is active, the linac should follow the motion of the target exactly, so the distance between the collimator and the target should remain constant. The angular position of the linac relative to the target should also be constant, indicating that the linac stays directly aimed at the target. We denoted the linac-to-target distance as r[i,j] where i is the beam number and j is an individual time point during that beam. Similarly, θ[i,j] is the elevation angle and φ[i,j] is the azimuthal angle for a particular beam number and time point. The elevation angle is measured relative to the z-axis, and the azimuthal angle is measured in the x-y plane relative to the x-axis. We also denoted m as the total number of tracked beams. We then defined six metrics of Synchrony performance.

Tracking Precision: We calculated the standard deviation of each of the three parameters during each beam; for example, the standard deviation of θ[i,j] for all values of j that occurred within the i^th beam. These values are σ_r[i], σ_θ[i], and σ_φ[i], respectively. Then, the overall precision of the simulated treatment can be characterized with the mean values of σ_r[i], σ_θ[i], and σ_φ[i] over all m tracked beams. For a system with perfect robotic arm tracking, the angles and distance would be constant at all times, so the values of all of these computed parameters would be zero.

Tracking Variability: We defined the overall tracking variability as the standard deviation of σ_r[i], σ_θ[i], and σ_φ[i] over all m tracked beams.

Results and Discussion

In a typical experimental session, we were able to track both the linac and the motion simulator for approximately 60% of the treatment time. Inability to track occurred when the Optotrak view was blocked by the linac or the robot arm, or when the linac was oriented in such a way that the markers on the end of the linac were not visible to the Optotrak. We experimented with placing the Optotrak in different locations in the room, but the location used for this study was the best compromise between the number of viewable beams and being able to ensure that the CyberKnife did not collide with the Optotrak.

Figure 4 shows a sample of the data recorded from the Optotrak and illustrates the tracking capabilities of CyberKnife with Synchrony. The upper plot shows the distance between the target and the linac along one axis, whereas the lower plot shows the recorded position of the linac during the same time period. In the left half of the figure, Synchrony is not active and the linac is stationary; the variation in distance is due to the motion of the respiratory simulator. In the right half of the figure, Synchrony activates and begins to track motion. We, therefore, observe that the position of the linac changes sinusoidally and the linac-to-target distance essentially is held constant. At the far right of the figure, the large change in both plots indicates that the CyberKnife is moving to a new beam position.

An example histogram showing the distribution of r[i,j] (distance) values from a single beam is shown in Figure 5.