gms | German Medical Science

GMS Current Topics in Computer and Robot Assisted Surgery

Deutsche Gesellschaft für Computer- und Roboterassistierte Chirurgie (CURAC)

ISSN 1863-3153

Radiosurgery: identification of efficient treatment beams guided by autostereoscopic visualization

Research Article

  • corresponding author Alexander Schlaefer - Institute for Robotics and Cognitive Systems, University of Lübeck, Lübeck, Germany
  • Oliver Blanck - Institute for Robotics and Cognitive Systems, University of Lübeck, Lübeck, Germany
  • Hiroya Shiomi - Osaka University Graduate School of Medicine, Osaka, Japan
  • Achim Schweikard - Institute for Robotics and Cognitive Systems, University of Lübeck, Lübeck, Germany

GMS CURAC 2006;1:Doc14

The electronic version of this article is the complete one and can be found online at: http://www.egms.de/en/journals/curac/2006-1/curac000014.shtml

Published: October 23, 2006

© 2006 Schlaefer et al.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by-nc-nd/3.0/deed.en). You are free: to Share – to copy, distribute and transmit the work, provided the original author and source are credited.


Abstract

In robotic radiosurgery, an arrangement of cylindrical beams forms a high dose area conformal to the target region. Beams start at various positions around the patient, and are oriented towards arbitrary points in the target. Due to the large number of potential beams, only a small subset of beams can be considered during treatment planning. An efficient heuristics to select promising candidate beams can substantially shorten and improve treatment planning.

We have developed a tool to visualize the 3D dose distribution and the beams. Using a hypsometric color scheme, the visualization toolkit (VTK), and an autostereoscopic display, we established the spatial extent of cold and hot spots and the effect of beams.

With this new visualization tool we assessed the dose distribution and the beam arrangement of two treatment plans for intra-cranial tumors. Based on the visual analysis of the dose distribution and the effect of the beams, we derived a heuristics to place beams. Treatment plans computed for the heuristically placed beams were compared to the original plans. The results suggest that a heuristic pre-selection of potential treatment beams is possible.

The visualization proved useful for 3D analysis of treatment plans and as a tool for rapid development of beam placement heuristics.

Keywords: radiosurgery, inverse planning, auto-stereoscopic visualization, robotics


Introduction

One of the most effective forms of cancer treatment is the delivery of a high dose of radiation to the clinical target volume (CTV). In radiosurgery, the high dose area is shaped to precisely cover the tumor so as to limit damage to surrounding normal tissue, allowing to escalate the dose to the target and to deliver the dose in a single or a few fractions. Examples of systems used for radiosurgery include the Gamma Knife, gantry based linear accelerators in combination with cylindrical or multi leaf collimators, and the robotic CyberKnife.

An early development, the Gamma Knife radiosurgical system (Elekta AB, Stockholm) was designed for treatment of intracranial targets. Using a stereotactical frame attached to the patient’s head, beams from up to 201 Cobalt 60 sources intersect at an iso-center, thus forming a small spherical region of high dose. Repeating this procedure, arbitrarily shaped tumors can be treated by placing a number of spheres with different radii accordingly. Conventionally, the selection of beams and spheres is performed manually, and Olofsson et al. [1] describe the use of haptic feedback and stereoscopic visualization to guide the process. A 3D scene containing the planning target volume (PTV), the organs at risk (OAR), and the dose surfaces is generated and displayed via shutter glasses to provide depth information. The human planner can place and replace spheres until the dose distribution is satisfactory. Since the dose is shown as an iso-surface, the target boundaries and areas of particularly high dose (hot spots) may be invisible. Haptic feedback is used to guide the planner towards points inside the target and to avoid hot spots. Another approach is the use of optimization techniques to find suitable arrangements of spheres [2], [3], [4]. Ferris et al. [2] use a heuristics analyzing the shape of the target to place spheres, and non-linear programming for optimization of the dose distribution.

Instead of using specialized equipment, intensity modulated radiotherapy (IMRT) is based on a traditional gantry based linear accelerator in combination with a multi-leaf-collimator. For practical reasons relatively few (typically up to nine) beams are used with IMRT, and often beams are arranged in one plane (co-planar), oriented towards one iso-center. However, each beam is divided into a large number of small beamlets that can be individually weighted, leading to a precision adequate for radiosurgical treatment. While the target dose primarily depends on the beamlet weights, the choice of beams affects the quality and efficiency of the treatment [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15]. In principle the beam orientation could be included in the optimization problem, but due to the large search space this approach is too computationally expensive. A common way to incorporate the beam orientation in the optimization process is to define a metric to estimate a beam’s value for the optimization. Schreibmann et al. [13] use simulated annealing and genetic optimization to find a set of adequately distributed beams that do not pass through OAR. Pugachev and Xing [9], [10] introduced the beam’s-eye view dosimetrics (BEVD) measure to assess the fitness of treatment beams. Similar to the visual evaluation of beams using the beams-eye-view, i.e. a view along the beam’s axis towards the target volume, they defined an empirical score function based on the calculation of the maximal allowable beamlet weights.

Using a robot arm to move a lightweight linear accelerator (linac), the CyberKnife (Accuray, Sunnyvale, CA) adds even more flexibility with respect to beam placement. Given the six degrees of freedom of the robot, the linac can be placed at virtually any position around the patient, and beams can be oriented towards arbitrary points in the target region. While for practical treatment the number of potential linac positions is restricted to approximately 100, there is no limitation on a beam’s orientation. It has been shown that the resulting non-iso-centric and non-coplanar beam arrangements can lead to very homogenous and conformal dose distributions [16], [17], [18]. Due to the large number of beams, planning is typically done automatically. First, a set of candidate beams is chosen and then the beam weights are determined [19], [20], [21]. Given the large number of potential treatment beams, an optimization including all possible beams is infeasible for practical treatment planning. However, the current treatment planning results in a relatively small subset of the candidate beams being weighted. If it were possible to identify criteria to assess a beam’s potential value for the optimization process, planning could be sped up and improved substantially.

We have developed a tool to support a three-dimensional visualization and analysis of treatment plans for robotic radiosurgery using an auto-stereoscopic display. The use of stereoscopic visualization for radiotherapy treatment plans has been reported by Olofsson et al. [1] and by Hubbold et al. [22], [23], [24]. Shutter glasses are used to achieve 3D vision in the system described by Olofsson, the target dose distribution is represented by an iso-surface, and haptic feedback is used to guide the planner away from hot spots. While Hubbold et al. see potential advantages in the use of autostereoscopic displays for radiotherapy planning they mention disadvantages of direct volume rendering for depth perception. To overcome this limitation for the visualization of hot and cold spots, they propose an automatic tunneling technique. By removing material that occludes regions of interest they achieve a better 3D impression. However, in order to decide what areas to remove it would be necessary to know the location of cold and hot spots. Additional problems could arise if multiple cold and hot regions needed to be shown.

This is the case in our application, where we prefer to define cold and hot spots relative to their neighborhood. A grid-like structure is used to facilitate an accurate visual understanding of cold and hot spots and their three-dimensional extent. Furthermore, the location and direction of beams and their effect on the volumes of interest is visualized. We present an application of the tool for analyzing existing treatment plans for robotic radiosurgery and in the development of new beam placement heuristics.


Methods

A patient’s anatomy with the PTV and the OAR, the beams, and the resulting dose distribution is essentially of three dimensional nature. While for conventional radiosurgery beams are typically restricted to be iso-centric, this is not the case for robotic radiosurgery. We found a visualization of the dose distribution and the beams in the sagittal, coronal, and axial planes to be of limited use, as the main axis of cold or hot regions and the central axis of the beams are virtually never perpendicular to any plane. Thus, it becomes hard to assess the three dimensional shape and extent of those regions and to understand the dose effect of the beams.

However, analyzing a treatment plan involves different criteria including the spatial arrangement of the beams and the resulting dose distribution, namely in the PTV and the OAR. In our tool, we discretize the volumes into voxels and use the dose in the voxel center for both, treatment planning and visualization. Each volume is represented by a set of small 3D shapes – e.g. spheres - that are placed into a 3D scene. The actual dose values are displayed using a hypsometric color scheme, i.e. the spheres are colored according to the voxel’s dose. Depending on the type of analysis, the beam can be visualized in different ways. To obtain a first impression and to compare a large number of beams, each beam is represented by a small cylinder along its central axis. When assessing the dose effect of a beam, a pencil shaped surface is shown which denotes the area where the dose coefficient is above a defined threshold. To assist the manual placement of additional beams, we show the beam’s central axis and two disks perpendicular to the axis. The two disks represent the outer limit of the beam and the areas where the dose coefficient is above a threshold, respectively. They can be slid along the central axis (Figure 1 [Fig. 1]).

We implemented the tool using the visualization toolkit (VTK) and an autostereoscopic display (SeeReal Technologies GmbH, Dresden) was used to show the 3D scenes (Figure 2 [Fig. 2]). The display has a physical resolution of 1600 by 1200 pixels which results in an effective resolution of 800 by 1200 pixels when used with stereo rendering, making it suitable for the display of relatively small objects like the spheres representing voxels.

We used the new tool for the visualization and analysis of two existing treatment plans and to derive an improved beam placement heuristics. To evaluate the fitness of beams selected for planning, we implemented an optimization algorithm similar to the one described by Schweikard et al. [20]. A linear program is generated to optimize the beam weights given in monitor units. For each beam and each voxel the respective dose coefficient is computed and then constraints are introduced which restrict the total dose in each voxel as prescribed by the dose bounds. We used this setup to calculate optimal plans for different sets of beams.


Results

Using the visualization tool we assessed the dose distribution and the beam placement of two existing treatment plans generated by inverse planning and containing 105 and 102 weighted beams, respectively. Based on the visual display of the dose distribution and the beams effect, we derived the following heuristics for beam placement: (a) use efficient beams that have a high dose effect in the target region, (b) avoid hot spots, e.g. the target center, (c) avoid OAR, and (d) use beams from (many) different directions.

To evaluate how information on the dose distribution and a beam’s effect could be useful to identify potential treatment beams, we manually applied the heuristics to the two aforementioned inter-cranial cases previously treated at Osaka University. We placed sets of 80, 100, and 120 beams for each plan, ran the inverse planning with those beam sets and compared the resulting plans.

The original plans consisted of 105 and 102 weighted beams out of 606 and 404 candidate beams respectively. From the visual evaluation, most of the weighted beams where targeted slightly off the tumor center. Cold spots were located on parts of the tumor surface, specifically towards organs at risk (optic nerve, optic chiasm). When applying the heuristics, we evaluated the plans after placing 80, 100, and 120 beams respectively. The plans contained 79, 89, and 99 weighted beams for the first case and 78, 89, and 98 weighted beams for the second case, see Table 1 [Tab. 1]. For all plans, the dose for the OAR stayed well below the specified bounds, and the target dose increased with the number of candidate beams, see Figures 3 [Fig. 3] and 4 [Fig. 4].


Conclusions

We presented a tool for three-dimensional visualization of radiosurgical treatment plans and its application in the development of beam placement heuristics. The visualization proved to be useful in the analysis and guidance of beam placement. Using a simple placement heuristics, 120 manually placed beams resulted in treatment plans comparable to plans generated from a much larger set of originally selected candidate beams. Manual beam placement in a clinical context is not intended. However, the results indicate that a heuristic pre-selection of candidate beams is possible. The visualization tool enables further studies to derive a more general heuristics for beam placement, which we plan to implement as an automatic procedure.


Acknowledgements

This work was partially supported by Deutsche Forschungsgemeinschaft (DFG).


References

1.
Olofsson I, Lundin K, Cooper M, Kjäll P, Ynnerman A. A haptic interface for dose planning in stereo-tactic radio-surgery. In: IV '04: Proceedings of the Eighth International Conference on Information Visualisation. 2004. p. 200-5.
2.
Ferris MC, Lim J, Shepard DM. An optimization approach for radiosurgery treatment planning. Siam J Optim. 2003;13(3):921-37.
3.
Ferris MC, Lim J, Shepard DM. Radiosurgery treatment planning via nonlinear programming. Annals of Operations Research. 2003;119:247-60.
4.
Shepard DM, Chin LS, DiBiase SJ, Naqvi SA, Lim J, Ferris MC. Clinical implementation of an automated planning system for gamma knife radiosurgery. Int J Radiat Oncol Biol Phys. 2003;56(5):1488-94.
5.
Söderström S, Brahme A. Selection of suitable beam orientations in radiation therapy using entropy and fourier transform measures. Phys Med Biol. 1992;37:911-24.
6.
Haas OCL, Burnham KJ, Mills JA. Optimization of beam orientation in radiotherapy using planar geometry. Phys Med Biol. 1998;43:2179-93.
7.
Rowbottom CG, Oldham M, Webb S. Constrained customization of non-coplanar beam orientations in radiotherapy of brain tumours. Phys Med Biol. 1999;44:383-99.
8.
Rowbottom CG, Webb S, Oldham M. Beam-orientation customization using an artifcial neural network. Phys Med Biol. 1999;44:2251-62.
9.
Pugachev A, Xing L. Computer-assisted selection of coplanar beam orientations in intensity-modulated radiation therapy. Phys Med Biol. 2001;46:2467-76.
10.
Pugachev A, Xing L. Incorporating prior knowledge into beam orientaton optimization in IMRT. Int J Radiat Oncol Biol Phys. 2002;54(5):1565-74.
11.
Das S, Cullip T, Tracton G, Chang S, Marks L, Anscher M, Rosenman J. Beam orientation selection for intensity-modulated radiation therapy based on target equivalent uniform dose maximization. Int J Radiat Oncol Biol Phys. 2003;55(1):215-24.
12.
Woudstra E, Heijmen B. Automated beam angle and weight selection in radiotherapy treatment planning applied to pancreas tumors. Int J Radiat Oncol Biol Phys. 2003;56(3):878-88.
13.
Schreibmann E, Lahanas M, Uricchio R, Theodorou K, Kappas C, Baltas D. A geometry based optimization algorithm for conformal external beam radiotherapy. Phys Med Biol. 2003;48(12):1825-41.
14.
Wang X, Zhang X, Dong L, Liu H, Wu Q, Mohan R. Development of methods for beam angle optimization for IMRT using an accelerated exhaustive search strategy. Int J Radiat Oncol Biol Phys. 2004;60(4):1325-37.
15.
Engel K, Tabbert E. Fast simultaneous angle, wedge, and beam intensity optimization in inverse radiotherapy planning. Optimization and Engineering. 2005;6(4):393-419.
16.
Webb S. Conformal intensity-modulated radiotherapy (IMRT) delivered by robotic linac-testing IMRT to the limit? Phys Med Biol. 1999;44:1639-54.
17.
Webb S. Conformal intensity-modulated radiotherapy (IMRT) delivered by robotic linac-conformality versus effciency of dose delivery. Phys Med Biol. 2000;45:1715-30.
18.
Yu C, Jozsef G, Apuzzo ML, Petrovich Z. Dosimetric comparison of cyberknife with other radiosurgical modalities for an ellipsoidal target. Neurosurgery. 2003;53(5):1155-63.
19.
Schweikard A, Tombropoulos R, Kavraki L, Adler JR, Latombe JC. Treatment planning for a radiosurgical system with general kinematics. In: Proc. of the IEEE International Conference on Robotics and Automation, San Diego, CA, 1994. p. 1720-7.
20.
Schweikard A, Bodduluri M, Adler JR. Planning for cameraguided robotic radiosurgery. IEEE transactions on robotics and automation. 1998;14(6):951-62.
21.
Tombropoulos RZ, Adler JR, Latombe JC. Carabeamer: A treatment planner for a robotic radiosurgical system with general kinematics. Medical Image Analysis. 1999;3(3):237-64.
22.
Hubbold R, Hancock D. Stereo display of nested 3D volume data using automatic tunnelling. In: Merritt JO, Bolas MT, Fisher SS, editors. Stereoscopic Displays and Applications. 1999. p. 200-7.
23.
Hubbold RJ, Hancock DJ, Moore CJ. Autostereoscopic display for radiotherapy planning. In: Proceedings Electronic Imaging '97 Symposium, Conference on Stereoscopic Displays and Virtual Reality Systems IV. Volume 3012. 1997.
24.
Moore C, Hubbold R, Hancock D. 3D sculpting and auto-stereoscopic display of ICRU-50 volumes and anatomy in conformal radiotherapy treatment planning. In: Proc IEEE Engineering in Medicine and Biology Society Conference. 1999.