X-Ray Guided Robotic Radiosurgery for Solid Tumors Mohan Bodduluri Accuray Incorporated 570 Del Rey Avenue Sunnyvale, CA 94085 USA and J. M. McCarthy Department of Mechanical and Aerospace Engineering University of California Irvine, CA 92697 Abstract This paper presents an overview of the x-ray guided robotic radiosurgery system that has been developed for the ablation of solid tumors. A robot mounted linear accelerator is directed through a sequence of positions and orientations designed to deliver high radiation dosages focussed at a specific location. Patient movement during treatment is identified by stereo x-ray measurements and the robotic system adjusts the linear accelerator prior to the delivery of radiation at each location. The result is accurate delivery without rigid fixation of the tumor relative to the treatment system. Introduction For almost 20 years precisely administered large doses of radiation has provided a non-invasive surgical treatment for brain tumors, (Leksell 1951, Larsson et al. 1958). Careful measurements in three dimensions are required to locate a tumor relative to the reference frame of the treatment device. Magnetic resonance imaging (MRI) and computed tomography (CT) scans provide the ability to precisely locate a tumor relative to skeletal landmarks, or relative to implanted fiducial.markers. The challenge is to determine the location of these landmarks relative to the treatment system. One strategy for locating these landmarks is to attach a stereotactic frame to the skeleton of the patient which is then locked in place relative to the treatment device. This allows the tumor to be located at a center of a sphere of radiation sources. The intersection of the beams from these sources results in a high dose of radiation that treats the tumor directly with minimal impact on the surrounding tissue. The approach described here is called frameless radiosurgery (Adler 1993). It uses stereo x- ray imaging to regularly measure the location of landmarks relative to the treatment device. An
industrial robot is used to move a small linear accelerator through a preplanned path to focus an array of beams from many different directions. The robot and imaging system can work together to adjust the focus of these treatment beams in order to accommodate movement of the patient. See Chenery et al. (1998) for a description of initial clinical results, and Murphy et al. (2000) for a discussion of the feasibility of using this approach to treat tumors in the spine and pancreas. In this paper, we describe the overall system and future research directions. Figure 1. A robot mounted linac is guided by a stereo x-ray imaging system. Ceiling mounted x-ray sources illuminate two x-ray cameras mounted orthogonally. The CyberKnife Stereotactic Robotic-radiosurgery System consist of four major components: i) a radiation source mounted on an industrial manipulator; ii) a stereo x-ray imaging system; iii) a patient positioning system (or couch); and iv) the treatment planning and delivery software system. See Figure 1. The Radiation Source The radiation source consists of a pulsed electron gun combined with a waveguide that uses microwave radiation to accelerate the electrons into a tungsten target. The result is an x-ray beam that can be collimated into a diameter as small as 5mm. This radiation source is known as
a linear accelerator and is commonly called a linac. The weight of the linac is approximately 100kg and must be manipulated by the robot to accommodate tumor movement. For this reason, an industrial robot such as the Fanuc 420 or Kuka 210 is used for these systems. Figure 2. A fixed set of beam directions are identified on a sphere centered on the tumor. The trajectories are planned relative to the sphere which moves with the patient. The Stereo X-ray Imaging System The imaging system consists of two x-ray sources mounted to the ceiling of the treatment room and a pair of x-ray cameras mounted orthogonally in a v -shaped frame and fixed to the floor. Both CCD fluoroscopes and amorphous silicon detectors have been used to provide 20cm x 20cm x-ray images. The location of the sources, cameras and patient support system results in a spatial resolution of approximately 0.5mm of patient movement per pixel. The x-ray sources and cameras are precisely located within the room and relative to the robot to ensure the accuracy of the tracking algorithms.
Two cameras located perpendicular to each other provide two images of the skeletal structure surrounding the tumor. The tracking algorithm uses these images to locate the tumor in three dimensions relative to the robot with sub-millimeter accuracy. The Patient Positioning System The patient positioning system (couch) supports the patient relative to the imaging system. Appropriate location of the tumor and landmarks in the field of view of the cameras and patient positioning appropriate x-ray illumination enhances tracking performance. However, once in position the couch remains fixed throughout treatment. Patient movement during treatment is accommodated by adjusting the robot manipulator. The Planning and Delivery Software A Silicon Graphics Octane II is used to provide software integration to the system. It supports both the treatment planning and the treatment delivery software. The system is designed to have a fixed set of nodes on a sphere from which treatment beams can be delivered, Figure 2. A treatment plan consists of a dose rate and duration at each node, Figure 3.. The robot moves sequentially through each of the nodes and delivers the required dose. Figure 3. The treatment planning translates a desired dose distribution into robot path and linac beam plan.
Tracking Patient Movement The imaging system of CyberKnife consists of two diagnostic x-ray sources mounted in the ceiling, two amorphous silicon cameras mounted in the floor, Figure 4. The cameras are orthogonal to each other, the sources are aimed at the cameras directly. The distance between the camera and patient is approximately 65 cm, which eliminates most of the scatter radiation. The distance between the patient and the x-ray source is approximately 260 cm, which provides the robot ample access to the patient during the patient treatment. Digitally Reconstructed Radiographs Before the treatment begins, the system generates a set of digitally reconstructed radiographs (DRRs) for the patient. A DRR can be viewed as an x-ray image generated from a CT scan of the patient, using a computer model of the x-ray source and camera. By incrementing the alignment of the CT scan through the entire range of expected patient movements, an array of synthetic x-ray images is obtained for matching to a live image. In the present implementation, DRRs cover ±10mm range with step size of 2.5mm. Image Matching The patient is positioned so that real time live images match the nominal DRR defined as zero movement. Movement of the patient, and therefore of the tumor away from this nominal location is detected by matching the x-ray images of the two cameras to the precomputed array of DRRs. Standard image processing techniques are used to subtract the images and obtain a measure of the difference of the images, Figure 5. An interpolating surface is fit to the data to obtain a position estimate to less than 1mm error. Figure 4. Two x-ray sources and a cameras provide orthogonal views of the tumor. Tracking is achieved by matching the x-ray images to a set of synthetic images that correspond to various movements
Figure 5. Synthetic images and real x-ray images are compared from cameras A and B to compute patient position. System Calibration A system centerpoint called the isocenter is identified in order to calibrate the installation of the imaging system and the robot in the treatment facility. The system accuracy from treatment planning through delivery measured by exposing a film cube in a head phantom, and is found to be within 2mm (Murphy and Cox 1996). Imaging System Calibration The alignment of the cameras and location of the sources places the image of the isocenter within 5mm of the center of both cameras. A calibration process identifies variations in location over the face of the camera to account for small variations. The imaging system s accuracy is less than 1 mm. Robot Calibration The installation of the robot determines the orientation of the user frame which has its origin at the isocenter. The tool frame is identified by measuring the position and orientation of the line determined by a laser which is mounted coaxial to the therapeutic beam. Further calibration involves an automated process in which the robot illuminates an isocrystal and searches for the position and orientation that provides the maximum signal. A correction look-up table is maintained by the system. The residual error of the robot positioning system is less than 0.5 mm.
The Treatment Process A typical patient treatment process begins with a CT scan which is generally performed a day or two prior to the scheduled radiation delivery. The medical physicist and physician register the patient on the system, import the CT scans, and generate a treatment plan. On the day of the treatment, the patient is positioned and the treatment plan is executed. A normal time for radiation delivery is approximately one hour. Treatment Planning System The treatment planning system provides an interactive environment that is used to: 1. Display the CT scan in various formats, such as axial, sagittal, coronal, as well as three dimensional surface reconstructions; 2. Outline tumors and other critical structures, such as eyes, brain stem, optic chiasm for identification to the planning system; 3. Direct the beams at the tumor while avoiding critical structures. The following methods are available for directing the beams: a. All the beams are directed at the same point (isocentric), b. Subsets of beams are directed at various points (multi-center), c. The beams are directed so that they are uniformly distributed throughout the tumor conformal shape) d. Manual specification of beam directions. 4. Adjust the amount of radiation for each of the beams. The methods for specifying the radiation are: a. Forward planning the user selects the doses for each beam b. Inverse planning the user specifies dose requirements in the tumor, and maximum allowed dose in the critical structures, and the planning system calculates dose for each of the beams. 5. Calculate and display the resulting dose calculations in terms of isodose curves, color washes, three-dimensional isodose clouds, and dose volume histograms (DVHs). 6. Simulate the treatment. This allows verification of the dose distributions. Figure 6. shows an example of the results of a simulated treatment plan.
Figure 6. The radiation dose distributions are planned on the CT slices directly. Treatment Delivery On the day of the treatment the patient lies on the couch of the patient positioning system. The treatment plan is selected and the system assists the operator in aligning the patient in the center of the imaging system. This is done by acquiring x-ray images and evaluating offsets from the nominal position. Once the patient is aligned, the treatment begins. During the patient treatment, robot moves the linear accelerator through each beam position and orientation called a node. At each node, a pair of orthogonal x-ray images of the patient is obtained and the movement of the patient is calculated. This information is transmitted to the robot, which compensates for the patient movement before the therapeutic beam is turned on. This process is repeated through all the nodes to complete an entire treatment. Future Research Currently imaging accuracy and tracking algorithms are the primary focus of research. Imaging accuracy depends on the quality and performance of the x-ray sources, cameras, and room calibration. The tracking algorithms depend on quality of the CT scan, the DRR generation, and image processing techniques. Six Dimensional Tracking Current tracking algorithms focus on translational movement and assume that changes in orientation due to patient movement are small. However, in the initial alignment stage, larger rotational movements can provide the convenient treatment configurations. In addition, there are
tumor locations where small rotations result in large translations. For this reason, a tracking algorithm that identifies full six-degree of freedom movement has been implemented, and refinements are the focus of continuing research. Fiducial Tracking Recently, a capability has been released for tracking tumors that are in locations where the skeletal structure is not present, or not visible enough, to identify the movement of the patient. In this situation metallic landmarks called fiducials are implanted and CT scans generated. The x-ray system can compare the location of the fiducials to the synthesized DRRs and determine patient movement. The size, shape and composition of these fiducials are being refined to improve performance in certain anatomic locations. Compensating for Respiratory Motion The body and internal organs of the patient move with the breathing cycle. Therefore tumors in the lungs, liver, and other structures can move during treatment. One strategy is to have the patient hold a breadth while a beam of radiation is applied. Another is to track the patient movement during the application of the radiation and compensate for the respiratory movement. This places severe demands on the quality of the imaging, tracking and robot system. This motion-tracking feature is an important direction for research. Conclusion This paper presents an overview of a robotic system for frameless stereotactic radiosurgery. It relies on stereo x-ray imaging to identify the position of the patient and tumor relative to the treatment system. The accuracy of this process has been shown to be within the current specifications of radiosurgery devices. The elimination of rigid fixation framework to locate the tumor relative to the treatment device increases patient comfort and reduces the procedure to an outpatient treatment. This technology is increasingly finding a range of clinical applications. To date over 2000 patients have been treated with robotic radiosurgery around the world. See Figure 7. Acknowledgments The authors acknowledge the contributions of James Wang, John Dooley, Greg Glosser, Uisik Ro, Gopi Kuduvalli, Andy Irish, Shehrzad Qureshi, Bill Main, Chris Lee, and Eric Earnst to the successful performance of the CyberKnife system.
Spine 6% Prostate <1% Pancreas 1% Lung 4% Primary Intracranial Tumor 15% Benign/ Functional Tumor 43% Metastatic Brain Tumor 31% Figure 7 The distribution of patients treated by the CyberKnife in the US. References J. R. Adler, ``Frameless Radiosurgery,'' in, S. J. Goetsch and A. A. F. DeSalles (ed): Sterotactic Surgery and Radiosurgery, Medical Physics Publishing, Wisconsin, 17:237-248, 1993. J. R. Adler, M. J. Murphy, S. D. Chang, and S. L. Hancock, ``Image-guided Robotic Radiosurgery,'' Neurosurgery, 44(6):1299-1307, June, 1999. S. G. Chenery, H. H. Chehabi, D. M. Davis, J. R. Adler, ``The CyberKnife: Beta System Description and Initial Clinical Results,'' Journal of Radiosurgery, 1(4):241--249, 1998. B. Larsson, L. Leksell, B. Rexed, ``The high energy proton beam as a neurosurgical tool,'' Nature, 182:1222-3, 1958. L. Leksell, ``The stereotaxic method and radiosurgery of the brain,'' Acta Chir Scand, 102:316-19, 1951. M. J. Murphy and R. S. Cox, ``The accuracy of dose localization for an image-guided frameless radiosugery system,'' Medical Physics, 23(12):2043-2049, 1996. M. J. Murphy, J. R. Adler, M. Bodduluri, J. Dooley, K. Forster, J. Hai, Q. Le, G. Luxton, D. Martin, J. Poen, ``Image-Guided Radiosurgery for the Spine and Pancreas,'' Computer Aided Surgery, 5:278-288, 2000. A. Schweikard, J. R. Adler, J. C. Latombe, ``Motion Planning in Stereotaxic Radiosurgery,'' Proc. International Conference on Robotics and Automation, 9:1909-1916. IEEE Press, 1993.
A. Schweikard, R. Z. Tombropoulos, J. R. Adler, J. C. Latombe, ``Treatment Planning for a Radiosurgical System with General Kinematics,'' Proc. International Conference on Robotics and Automation, 10:1720-1727. IEEE Press, 1994. S. B. Tatter, ``History of Stereotactic Radiosurgery,'' http://neurosurgery.mgh.harvard.edu/histpb.htm, MGH Neurological Service, 1998.