Respiration tracking in radiosurgery

Achim Schweikard
beck, 23538 Lu beck, Germany Informatik, Universitaet Lu

Hiroya Shiomi
Division of Multidisciplinary Radiotherapy, Osaka University Graduate School of Medicine, Japan

John Adler
Neurosurgery, Stanford University Medical Center, Stanford, California 94305

Received 21 July 2003; revised 26 May 2004; accepted for publication 1 June 2004 Respiratory motion is difcult to compensate for with conventional radiotherapy systems. An accurate tracking method for following the motion of the tumor is of considerable clinical relevance. We investigate methods to compensate for respiratory motion using robotic radiosurgery. In this system the therapeutic beam is moved by a robotic arm, and follows the moving target through a combination of infrared tracking and synchronized x-ray imaging. Infrared emitters are used to record the motion of the patients skin surface. The position of internal gold ducials is computed repeatedly during treatment, via x-ray image processing. We correlate the motion between external and internal markers. From this correlation model we infer the placement of the internal target during time intervals where no x-ray images are taken. Fifteen patients with lung tumors have recently been treated with a fully integrated system implementing this new method. The clinical trials conrm our hypothesis that internal motion and external motion are indeed correlated. In a preliminar study we have extended our work to tracking without implanted ducials, based on algorithms for computing deformation motions and digitally reconstructed radiographs. 2004 American Association of Physicists in Medicine. DOI: 10.1118/1.1774132 Key words: whole body radiosurgery, respiration tracking, soft-tissue navigation

I. INTRODUCTION The success of radiosurgical methods for brain tumors suggests that higher precision treatment could dramatically improve treatment outcome with tumor irradiation, and enable tissue ablation.1,2 High precision radiosurgery has been limited to brain tumors, since stereotactic xation is difcult to apply to tumors in the chest or abdomen. To apply radiosurgical methods to tumors in the chest and abdomen, it is necessary to take into account respiratory motion. Respiratory motion can move the tumor by more than 1 cm. Without compensation for respiratory motion, it is necessary to enlarge the target volume with a safety margin. For small targets, an appropriate safety margin produces a very large increase in treated volume. For a spherical tumor of radius 1 cm, a safety margin of 0.5 to 1 cm would have to be added to ensure the tumor remains within the treated volume at all times. The ratio between the radius and volume of a sphere is cubic. Thus a margin of 1 cm will cause an eight-fold increase in treated volume, which primarily involves healthy tissue. Furthermore, observed motion ranges in excess of 3 cm suggest that a 1 cm margin may not be sufcient in all cases. An enlarged margin of 2 cm would result in a 27-fold increase of dose in this example. Thus, an accurate method capable of compensating for respiratory motion would be of utmost clinical relevance. Typical variations in the pattern of respiratory motion can range from 1 to 2 cm for the same patient in pediatrics, and the duration of a single respiratory cycle has been reported to vary between 2 and 5 s.3 We have investigated the design of a new sensor method
1 Med. Phys. 31 8, August 2004

for tracking respiratory motion.4 A method for correlating internal motion to external surface motion has been developed. The treatment beam itself is moved by a robotic arm. The system is based on a modied Cyberknife Accuray Inc., Sunnyvale, CA. A number of patients have recently been treated successfully with this new correlation-based method. II. RELATED WORK Intratreatment displacements of a target due to respiration have been reported to exceed 3 cm in the abdomen, and 1 cm for the prostate.5,3 Webb6 discusses methods for achieving improved dose conformality with robot-based radiation therapy. Specically, trade-offs between treatment path complexity and conformality are analyzed. The experience reported suggests that very high conformality can be achieved with robotic systems by modulating the intensity of the beam. However, this assumes the ideal situation of a stationary target, unaffected by respiration. This assumption holds only for few anatomic sites, such as the brain or spine, but not for tumors close to the lung or diaphragm. Conventional radiation therapy with medical linear accelerators LINAC systems uses a gantry with two axes of rotation movable under computer control.7 This mechanical construction was designed to deliver radiation from several different directions during a single treatment. It was not designed to track respiratory motion. Respiratory gating is a technique for addressing this problem with conventional LINAC radiation therapy. Gating techniques do not directly
2004 Am. Assoc. Phys. Med. 1

0094-2405200431814$22.00

Schweikard et al.: Respiration tracking in radiosurgery

compensate for breathing motion, i.e., the therapeutic beam is not moved during activation. Instead the beam is switched off whenever the target is outside a predened window. One of the disadvantages of gating is the increase in treatment time. A second problem is the inherent inaccuracy of such an approach. One must ensure that the beam activation cycles can have sufcient length for obtaining a stable therapeutic beam. Kubo and Hill8 compare various external sensors breath temperature sensor, strain gauge, and spirometer with respect to their suitability for respiratory gating with LINACS. By measuring breath temperature, it is possible to determine whether the patient is inhaling or exhaling. It is veried in Ref. 8 that frequent activation/deactivation of the linear accelerator does not substantially affect the resulting dose distribution. However, the application of such a technique still requires a substantial safety margin for the following reason: The sensor method only yields relative displacements during treatment, but does not report and update the exact absolute position of the target during treatment. Tada et al.9 report using an external laser range sensor in connection with a LINAC-based system for respiratory gating. This device is used to switch the beam off whenever the sensor reports that the respiratory cycle is close to maximal inhalation or maximal exhalation. We investigate a method for tracking a tumor during treatment. Stereo x-ray imaging is combined with infrared tracking. X-ray imaging is used as an internal sensor, while infrared tracking provides simultaneous information on the motion of the patient surface. While x-ray imaging gives accurate information on the internal target location, it is not possible to obtain real-time motion information from x-ray imaging alone. In contrast, the motion of the patient surface can be tracked with commercially available high speed infrared position sensors. The main idea of our approach is to use a series of images from both sensors infrared and x ray where signal acquisition is synchronized. From such a series of sensor readings and corresponding time-stamps, we can determine a motion pattern. This pattern correlates external motion to internal motion. Below we describe an integrated system using the new correlation method internal versus external ducials, and outline a concept for extending this work to tracking without implanted ducials. III. SYSTEM
A. Overview

FIG. 1. System overview. Infrared tracking is used to record external motion of the patients abdominal and chest surface. Stereo x-ray imaging is used to record the three-dimensional position of internal markers gold ducials at xed time intervals during treatment. A robotic arm moves the beam source to actively compensate for respiratory motion.

tomographic CT monitoring. In addition, hardware means for capturing the time points of x-ray image acquisition and infrared sensor acquisition are used to synchronize the two sensors. Prior to treatment, small gold markers are placed in the vicinity of the target organ. Stereo x-ray imaging is used during treatment to determine the precise spatial location of these gold markers. Using stereo x-ray imaging, precise marker positions are established once every 10 s. Limitations on patient radiation exposure and x-ray generator activation times currently prevent the use of shorter intervals. Notice, however, that even much shorter time intervals would clearly be insufcient for capturing respiratory motion. The typical target velocity under normal breathing is between 5 and 10 mm/s. External markers placed on the patients skin can be tracked automatically with optical methods at very high speed. Updated positions can be transmitted to the control

Figures 1 and 2 show the system components. A robot arm modied Cyberknife system moves the therapeutic beam generator medical linear accelerator. The component added to the standard Cyberknife system is an infrared tracking system BGI Inc., Boulder, CO. Infrared emitters are attached to the chest and the abdominal surface of the patient. The infrared tracking system records the motion of these emitters. A stereo x-ray camera system x-ray cameras with nearly orthogonal visual axes records the position of internal gold markers, injected through an 18 gauge biopsy needle into the vicinity of the target area under computerized
Medical Physics, Vol. 31, No. 8, August 2004

FIG. 2. Infrared emitters, attached to the patients chest, are used as external markers Osaka University Medical Center.

Schweikard et al.: Respiration tracking in radiosurgery

dation and Osaka University Hospital. Sixteen patients have since been treated at Osaka University Hospital with the ducial-based method described in Sec. III. Notice that the method in Sec. III still requires the use of implanted ducial markers.
A. Clinical trials

FIG. 3. Total target excursion top curve and correlation error bottom curve in mm for a clinical case. x-axis: treatment beam direction number x-ray live shot number, y-axis: error in mm.

Figure 3 shows representative results for one clinical case. The gure shows the total correlation error. Thus, based on the correlation model Sec. III, we compute the current position of the target based on the external infrared sensor signal alone. At this same time point, we also acquire a pair of x-ray images. We then plot the distance in mm from the placement inferred by the correlation model and the actual placement determined from the implanted ducial markers in the image. The top curve in this gure shows the corresponding target excursion. Two more representative cases are shown in Figs. 4 and 5. V. DISCUSSION AND CONCLUSIONS

computer more than 20 times per second. However, external markers alone cannot adequately reect internal displacements caused by breathing motion. Large external motion may occur together with very small internal motion, and vice versa. We have observed 2 mm external emitter excursion in combination with 20 mm internal target excursion. Similarly, the target may move much slower than the skin surface. Since neither internal nor external markers alone are sufcient for accurate tracking, x-ray imaging is synchronized with optical tracking of external markers. The external markers are small active infrared emitters Flashpoint FP 5000, BGI Inc., Boulder, CO attached to the patients skin Fig. 2. The individual markers are allowed to change their relative placement. The rst step during treatment is to compute the exact relationship between internal and external motion, using a series of x-ray snapshots showing external and internal markers simultaneously. Each snapshot has a time-stamp which is used to compute the correlation model. IV. EXPERIMENTS AND CLINICAL TRIALS Clinical trials of the method in Sec. III have been carried out at several institutions, including Cleveland Clinic Foun-

The above described technique combines information from two sensors: x-ray imaging and infrared tracking. Both sensors have characteristic advantages in our application. Stereo x-ray imaging can determine the exact location of ducial markers via automatic image analysis. Our tracking method x-ray imaging with time-stamps, combined with infrared sensing does not require the location of infrared emitters be computed in the x-ray images. The infrared emitters do not even have to be visible in the x-ray images. This avoids occlusions and simplies image processing considerably. The observed motion of external infrared emitters 4 to 20 mm is large when compared to the noise range of infrared position computation below 0.01 mm. Thus, the ratio between noise and signal is adequate for following the respiratory cycle. Clinical testing of the above described method demonstrates an overall correlation error of below 2 mm throughout an entire 70 min treatment, while the total target motion was over 10 mm. This suggests that our method meets the denition of radiosurgcial accuracy and is capable of reducing safety margins by a very substantial amount. Preliminary results obtained with an extended tracking method, which obviates the need for implanted ducials, will

FIG. 4. Treatment of a hepato-cellular carcinoma HCC with ducial-based respiration compensation as described in Sec. III left image: before treatment, right image 3 months after treatment. Total treatment time was 40 min per fraction for three fractions with 80 beam directions and 39 Gy.

Medical Physics, Vol. 31, No. 8, August 2004

Schweikard et al.: Respiration tracking in radiosurgery

FIG. 5. Treatment of an adeno carcinoma with ducial based respiration tracking Osaka University Hospital treatment with 39 Gy/3 fractions.

be described in a forthcoming article. This new method is very closely related to the method above, but adds several elements: computation of 3D deformation images for internal organs, computation of digitally reconstructed radiographs, and a technique termed 7D registration, which aims at both registering the position and orientation of an organ, and the corresponding respiration stage.
1

A. Schweikard, M. Bodduluri, and J. R. Adler, Planning for cameraguided robotic radiosurgery, IEEE Trans. Rob. Autom. 146, 951962 1998. 2 K. R. Winston and W. Lutz, Linear accelerator as a neurosurgical tool for stereotactic radiosurgery, Neurosurgery 22, 454 464 1988. 3 M. R. Sontag et al., Characterization of respiratory motion for pedriatic conformal 3D therapy, Med. Phys. 23, 1082 1996.

A. Schweikard, G. Glosser, M. Bodduluri, M. Murphy, and J. R. Adler, Robotic motion compensation for respiratory movement during radiosurgery, Special Issue on Planning and Image-Guidance in Radiation Therapy, Computer-Aided Surgery, September, 2000. 5 S. M. Morrill, M. Langer, and R. G. Lane, Real-time couch compensation for intra-treatment organ motion: Theoretical advantages, Med. Phys. 23, 1083 1996. 6 S. Webb, Conformal intensity-modulated radiotherapy IMRT delivered by robotic linac-conformality versus efciency of dose delivery, Phys. Med. Biol. 45, 17151730 2000. 7 S. Webb, The Physics of Three-dimensional Radiation Therapy Institute of Physics, Bristol, 1993. 8 H. D. Kubo and B. C. Hill, Respiration gated radiotherapy treatment: a technical study, Phys. Med. Biol. 41, 9391 1996. 9 T. Tada et al., Lung cancer: Intermittent irradiation synchronized with respiratory motionresults of a pilot study, Radiology 207, 779783 1998.

Medical Physics, Vol. 31, No. 8, August 2004