Robotic Whole Body Stereotactic Radiosurgery: Clinical Advantages of the CyberKnife® Integrated System
by È Coste-Manière, D Olender, W Kilby, R A Schulz
, Reprinted by permission from The International Journal of Medical Robotics and Computer Assisted Surgery
Robotic Industries Association Posted 03/01/2005
Accuray Europe, Paris, France
D Olender, W Kilby and R A Schulz
Accuray Inc., Sunnyvale, California, USA Correspondence to: R Schulz
RIA EDITOR’S NOTE: Medical applications of commercial robotics have gained widespread acceptance. This paper examines the use of robots in radiosurgery. The definition of an Industrial Robot, according to the Robotic Industries Association is: An automatically controlled, reprogrammable multipurpose manipulator programmable in three or more axes which may be either fixed in place or mobile for use in industrial automation applications.
INTRODUCTION: THE REVOLUTION OF ROBOTICS IN MEDICINE
A robot is defined as a sensor-based tool capable of performing precise, accurate and versatile actions on its environment. (RIA EDITOR’S NOTE: Please see RIA’s definition of an Industrial Robot above.) Robots were initially developed to spare humans from performing burdensome, repetitive or risky tasks. They have yielded substantial productivity benefits within a variety of manufacturing fields. Robots have progressed from the research lab to the factory to medical devices.
In medicine, in their most sophisticated form, robots have recently evolved into complex systems integrating perception (medical images and information) and action (precise spatial positioning and sensory feedback) by mechanically controlled systems and image-guided devices (1) . Robots are becoming revolutionary tools for surgeons in a variety of clinical applications. In addition, they can extend capabilities that transcend human limitations such as tremor reduction, repeatability, precision and accuracy. In doing so, they provide a new level of minimally invasive access to a variety of anatomical targets. These new capabilities have the potential to give better consistency to surgical treatments resulting in improved patient outcome and added economic benefit.
A number of robots have been successfully applied to several areas of medicine. Historically, rigid or rigidly constrained organs were the first to benefit from the geometrical precision of robotic systems. Orthopaedic and neurosurgical applications have utilised robots and image guidance for total joint replacement (hip, knee), brain surgery and spine surgery. Some robots such as Robodoc® (Integrated Surgical Systems, Inc., USA, www.robodoc.com) have had varying degrees of commercial success in this area. Tele-operated robots such as the da Vinci® (Intuitive Surgical® , Inc., USA, www.intuitivesurgical.com) have allowed soft tissue procedures to start benefiting from robotic enhancement. These systems have successfully made their way into cardiac and abdominal applications as an enhancement to traditional minimally invasive surgery.
The delivery of ionising radiation for radiosurgical purposes has been an enhancement to the use of radiation for general therapeutic uses. The CyberKnife® radiosurgery system (Accuray™ , Inc, USA, www.accuray.com), has pioneered the field of robotic radiosurgery by introducing the advantages of medical robots to perform more precise and accurate delivery of ionising radiation. Intracranial robotic radiosurgery was first achieved on neuropathology within the skull. This was followed by radiosurgery of generally static extra-cranial targets such as spine lesions. More recently, the design of innovative, target-tracking technologies has extended clinical applications to tumours and lesions within soft tissues which are affected by respiratory motion.
RADIOSURGICAL REQUIREMENTS: AN OVERALL GEOMETRICAL PROBLEM
Challenges to the adoption of robotics into current clinical practice include clinical benefit, ease of use in the clinical environment and total economic efficacy for society. Without these benefits, new medical devices, including those that involve robotics, will have scant chance of being adopted by the medical community. Both clinical and economic benefits of any revolutionary medical device must be realizable for a technology to be successfully integrated into standard medical practice.
Introduced in the 1950’s by Lars Leksell, MD, radiosurgery has been defined as precisely targeted doses of radiation for tumour ablation as a substitute for procedures using traditional surgical instruments. These early ‘‘virtual scalpel’’ procedures were able to perform ablations for a variety of intracranial abnormalities. Recently, radiosurgery has been extended to the whole body applications. The sequence of product development events for radiosurgery is analogous to the progression of CT in the early 70s and MR in the early 80s. In the former case CT imaging systems were confined to only the head for almost a decade from inception. In the latter case MR systems were not generally used in the body for at least 5 years from inception. The knifeless, minimally invasive, procedure of radiosurgery eliminates the trauma of traditional surgery with little swelling, no blood or tissue loss and no infection. Furthermore, the morbidity and mortality associated with general anaesthesia is eliminated and patients can be treated in ambulatory environments with their associated economic and clinical benefits.
These advantages become fully realised when radiosurgery, the equivalent of abnormal tissue dissection, is performed. This technique maximises target resection and minimises normal tissue destruction. For example, the radiosensitive spinal cord must be protected when targeting lesions in the vertebrae due to the cord’s proximity to the tumour. The radiation dose has to optimally fit the tumour shape, while reducing the damage to collateral organs. This requires focusing the radiation on the target in a precise, conformal, and accurate manner (2) . From a robotics and medical imaging viewpoint, this process may be seen as an overall ray-tracing geometrical problem. Many parameters have to be adjusted to optimize radiosurgical workflow.
Pre-operative lesion identification phase
The targeted lesion and its surrounding critical tissues must be identified in 3-D space relative to the patient’s reference frame. Sophisticated medical imaging techniques are key to optimum target lesion identification. Enhanced medical imaging technologies such as functional imaging (CT/PET, fMRI and spectroscopy) can help provide precise and thorough modelling of the patient’s radiosurgical target site.
Pre-operative planning phase
The goal of this step is to sculpt a conformal dose volume around the target while minimising the dose delivered to adjacent healthy tissues. To do this, the system uses a combination of beam positions whose relative weights, or dose contributions, have been scaled to volumetrically shape the dose accordingly. In the model known as forward planning, this is done manually by the user specifying the desired weight of the various beams. More commonly, inverse planning is employed. This planing method utilises an algorithm to automatically calculate the optimum combination of beams and weights based upon user-defined dose constraints to the target and healthy tissues.
System registration phase
The lesion and the radiation source must be registered to one another in 3D space, so as to accurately preoperatively plan for optimal radiosurgical results. Six degrees of freedom are necessary to describe the spatial position (x, y, z) and the directional orientation (yaw, pitch, roll) of the target relative to the source. Various hardware and software strategies for positioning the radiation source with respect to the patient have been implemented.
In Leksell’s precursor radiosurgical system, the GammaKnife (Elekta, Sweden, www.elekta.com), the (Cobalt-60) radiation sources are fixed in space and the patient’s head is moved and registered so as to expose the anatomical target to the focal point of the 201 pencil like gamma radiation beams. This scenario requires the rigid fixation of the patient’s skull within a pin based stereotactic frame, with substantial discomfort to the patient. By relocating the frame within the machine, the patient’s skull is repeatedly repositioned relative to the fixed radiation reference frame based on a pre-calculated radiation plan.
In later systems, a linear accelerator producing high energy X-rays is attached to a gantry-based system. The resultant device generates arbitrarily shaped beams of radiation, which intersect one another to define a relatively uniform dose distribution at the target. In these systems, fixed stereotactic frames remain necessary to register pre-operative patient information to the intra-operative confluence of the treatment beams. Both of these technologies have the additional constraint of limiting the treatment to single fractions due to the invasive nature of the frame, eliminating the option of hypo-fractionated radiosurgical strategies.
Relocatable stereotactic frames have been developed to avoid this limitation, but these also achieve lower alignment accuracy than is possible with the fixed frames.
Robotic stereotactic radiosurgery
The CyberKnife robotic stereotactic radiosurgery integrated system provides a clever solution to the problems encountered in other systems. Employing real-time frameless registration (3) with six degrees of robotic positioning, this system provides precise spatial positioning. This results in a variety of advantages over other systems.
Using advanced robotic systems, the treatment beam distribution may be eloquently controlled. Such technologies allow radiation beams to be delivered to one or more irregularly shaped lesions while avoiding irregularly shaped radiosensitive areas. Treatment may involve beam paths with a single isocenter, with multiple isocenters, or with completely non-isocentric approaches. The latter approach will always provide the most conformal treatment for irregularly shaped target lesions. Treatment can be delivered in either a single session (mono-fractionation) or in a small number of sessions (hypo-fractionation) as determined during treatment planning. Treatment plans may be optimized for any combination of treatment time, dose conformality or dose homogeneity, by varying the planning constraints. Integration of the overall treatment workflow is key (4) . The inherent complexity of the information may be managed in an elegantly structured way prior to its adoption and use by multi-disciplinary teams of surgeons, oncologists, physicists and technologists.
THE CYBERKNIFE SYSTEM
The CyberKnife® system is a non-invasive radiosurgical device for tumour ablation. It is a unique device that integrates robotics with image-guidance technology. It is used to destroy lesions or tumours with large doses of accurately targeted megavoltage X-radiation. During treatment, multiple radiation beams are delivered according to a pre-defined treatment plan. A 6MV linear accelerator mounted on a robotic positioning arm (KUKA, Germany, www.kuka.de) accurately targets the beams at tumours and other lesions in the head and body. The radiation beams, and their resultant dose distribution, are designed to destroy the tumour while minimising exposure to nearby healthy tissue. Prior to and during treatment, a system composed of two orthogonal imaging chains made of diagnostic (kV) X-ray sources and digital amorphous silicon detectors provides a continuous update of the patient’s position. This system allows the robotic manipulator to correct for changes in patient position during treatment beam delivery. A five-degree of freedom treatment table (Axum™ ) is also available for automatic patient (re)positioning prior to or during treatment. A real time optical tracking subsystem is used for dynamic compensation of tumour movement due to respiration during treatment delivery. Figure 1 illustrates the overall system hardware within the radiosurgical suite. Planning, monitoring and system control equipment are located outside the shielded treatment room.
The CyberKnife system was developed to overcome both the geometrical limitations of gantry-based systems and the invasive aspect of frame-based stereotactic systems. A detailed history of the CyberKnife system can be found in ‘‘Accuray: Tightly Targeting Tumours’’ (5) .Using the CyberKnife system, the radiosurgical procedure requires no anaesthesia and the technology is available for ambulatory patients. Patients typically resume normal activities immediately after treatment.
CYBERKNIFE TREATMENT WORKFLOW
The adoption of medical robots in regular clinical practice has been limited by a dearth of formalized clinical workflow. Too often workflow is inefficient and disseminated into many steps lacking integration and synchronization (6) . The CyberKnife system’s workflow incorporates automated subsystems, technologies and techniques which efficiently integrate the different steps found in traditional radiosurgical procedures. This workflow employs a multi-dis-ciplinary clinical team composed of surgeons, radiation oncologists, medical physicists and radiation therapists working together prior to and during treatments.
In the preoperative phase, the workflow begins with patient preparation. In intracranial surgery, a soft retention mask is moulded on the patient’s face prior to their medical imaging examination. In extracranial procedures, radiopaque fiducials may be implanted in close proximity to, or inside, the tumour or lesion in a minimally invasive interventional procedure. Patients may also be fitted with a custom moulded body cradle to stabilize the patient on the treatment table.
The second step of the workflow consists of medical imaging studies. These examinations require at least a CT scan. Additional imaging procedures such as MRI, 3D angiography or functional imaging techniques (fMRI, PET or PET/CT, etc.) may be required to better define specific pathologies. These imaging examinations are used as inputs for segmentation (contouring) of both the target and radiosensitive normal anatomy.
In the normal workflow, the surgeon is in charge of anatomical contouring. Models of the lesion and surrounding tissues, displayed on the medical images, are used as inputs to the pre-operative planning phase. The treatment beam arrangement is then optimised to deliver the required dose distribution at the target lesion(s) while minimising normal tissue irradiation. Medical physicists and radiation oncologists are both involved in planning the radiosurgical treatment including determining the optimal number of treatment fractions. The system optimises the robot’s positional sequencing and those orientations required to render that treatment. Once this treatment plan is complete, the radiosurgical treatment can begin.
For treatment, the patient is positioned on the treatment table. Their face mask or other immobilisation device is used to limit movement while maintaining the patient in a comfortable position. Next, real time X-ray images of the patient in the surgical position are acquired. Target localisation information based on skull or fiducial identification, is extracted automatically and compared with information available from the preoperative medical data. Geometrical transformations are computed to guide the operator to manually or automatically manoeuvre the patient and the treatment table so as to approximately align the patient’s real position with the position of the preoperative imaging data. This procedure continues iteratively until an initial alignment threshold is met. The final, precise alignment is then achieved by compensating with the robotic manipulator.
The robot is positioned around the patient and therapeutic beam delivery can start in a sequential manner. The manipulator follows a given path around the patient as computed by the planning system. Prior to the delivery of each beam in the path, the imaging system rechecks patient alignment. Following the same image registration process as employed for patient positioning, the system calculates the patient’s position to 0.1 mm and 0.1 degrees. Patient displacement information is returned to the manipulator, allowing the robot to automatically modify its delivery path to compensate for patient movement during the treatment. In cases where patient movement is minimized, the operator can reduce the imaging frequency (typically to every third beam) in order to expedite the treatment. Should patient motion exceed an acceptable displacement threshold, the treatment is stopped and the patient can be repositioned manually or using the automatic couch. Beam delivery, localization and co-registration processes are performed continuously until the treatment is completed. For radiosurgical treatments that require more than one fraction, the same workflow is repeated by the clinical team during subsequently scheduled treatment sessions.
THE CYBERKNIFE SUBSYSTEMS AS USED IN THE TREATMENT WORKFLOW
The individual elements of the CyberKnife radiosurgical system are detailed in their order of appearance in the clinical workflow.
Patient modeling: The CyRIS™ InView image fusion and contouring station
The CyRIS™ InView platform is an interactive software package enabling clinicians to complete the critical step of visualization and segmenting (contouring) anatomical structures in an efficient manner. This procedure (Figure 2) may be performed remotely from the physician’s practice, clinic, or hospital via computer network connections, thus maximising clinical efficiency.
A set of slices from medical images (CT, MRI scans, 3D Angiography (7), PET, SPECT,…) are used as inputs to the CyRIS InView subsystem. State-of-the-art multi-modality image fusion algorithms (8) are integrated in the treatment planning system to make optimal use of the capability of each imaging modality. Sophisticated drawing tools allow the clinician to quickly and accurately define the target and other relevant anatomical volumes on any of the co-registered image sets, providing a series of structures to be used for dose planning. Three-dimensional visualisation features allow an efficient viewing of patient anatomy and appropriate dose calculation of the target volume and surrounding normal tissue structures.
The Treatment Planning System’s function is to design the optimal radiation dose distribution plan for the accelerator-manipulator system. The planning system enables various treatment beam geometries and their resulting dose distributions to be simulated. This allows an optimal treatment plan to be formulated, based on clinician defined dose constraints to the target and normal tissues (9) .
The CyberKnife system irradiates a target by directing beams of X-rays from many directions. A treatment beam can be viewed as a cone beam of therapeutic X-rays originating from a source point, and aimed at a target point in the anatomy, with the size of the beam determined by a variable collimator. These beams need only intersect with the target volume and do not necessarily converge on a single point within the target. Dosimetric beam data is carefully measured for each individual installation during commissioning of the system to enable accurate dose simulation to be performed by the Treatment Planning System. The planning system allows the user to specify the attributes of all the possible beams from the system.
The beam selection process is carried out as follows: The source points are chosen from a preselected set of positions (or nodes) from which radiation is delivered to the treatment target. Currently the number of nodes is approximately 100, with 12 different angular orientations of the head of the robot for each node (Figure 3) resulting in 1200 potential beam trajectories per plan.
The user may select the beam target points as follows:
- Single Centre Planning: A single target point is to be used by all the beams. This process results in an approximately spherical dose distribution.
- Multiple Centre Planning: Multiple (up to ten) target points are to be used by the multiple beams. This process is used for the treatment of multiple lesions with spherical distributions or multiple spherical treatments for a single irregular lesion.
- Conformal Planning: Multiple independent target points are distributed according to the geometry of the lesion (i.e. each beam is targeted at a different point within the target volume). This process results in highly conformal dose distributions.
One or more secondary collimators can be used to give the beams the diameter required to achieve the desired clinical result. Beam trajectories and relative weights must then be assigned. In forward planning, the beam-weights for each beam are selected by the user and an algorithm computes the dose distribution. In inverse planning , the algorithm incorporates a two step process to arrive at the desired dose distribution. First, the system adjusts the robot’s angular orientation for each nodal position such that the beam’s central axis intersects the target volume. Next, the algorithm computes the weights for each beam so that the composite dose from all beams complies with the dose constraints for each volume of interest (target and critical structures). The resulting dose distribution is displayed on the monitor and can be fine tuned and then stored in the patient’s files as shown (Figure 4)
The treatment plan computed as a result of the above beam selection and weighting processes is then converted into a set of paths that the robotic manipulator can safely follow. The treatment information is in turn communicated to the delivery system for treatment.
Registration: Patient alignment and target localisation
The graphical user interface (GUI) of the main treatment delivery software is the primary control point of the CyberKnife operational system. It is at the heart of the integrated architecture approach. The treatment delivery software initiates and monitors operations in the different subsystems. During treatment delivery, the software monitors the system status and safety controls, reports errors, manages the patient database and records treatment data log files for post-treatment assessment and analysis.
Registration makes use of the localization system composed of two diagnostic X-ray beams and two amorphous silicon plates, orthogonally positioned. They provide near real-time digital X-ray images of the patient in the treatment position. After verification of the target position, this subsystem provides the requested information to the control algorithms allowing them to dynamically reposition the patient table and/or the manipulator arm. Afterwards, the treatment will then be activated and the 6MV beams can be triggered iteratively as calculated during the planning phase.
The system uses a movable table to support the patient. The table, with imbedded safety switches, can be moved automatically in five orientations (three translations and two rotations). The final rotational movement can be applied manually. Figure 1 illustrates the overall system configuration for patient alignment and target localization.
The comparison of the real time orthogonal [left anterior oblique (LAO) and right anterior oblique (RAO)] X-ray images and the digitally reconstructed radiographs (DRRs) in the same LAO and RAO positions from the preoperatively obtained CT scans is highly accurate and eliminates the requirement for frame based stereotactic treatment. Replacing the function of frame based systems provides greater comfort to the patient and allows for treatments of multiple fractions where indicated. Either bones or fiducials can be localized and registered to the perioperative images in real-time.
A table of DRRs samples the full range of motion of the target centre with six degrees of freedom. The registration process verifies that the real-time images represent an acceptable position within the range sampled by the reference images. It interpolates the actual position and orientation of the real image with respect to the reference. The correct treatment is obtained in two stages: i) prior to treatment, approximate alignment is obtained by automatic or manual repositioning of the table, and ii) final alignment during treatment is achieved by automatic adjustment of the robot’s position.
After the initial registration is performed, the software tracks the target centre, via the fiducials or bony landmarks of the skull and sends any positional correction data to the manipulator.
Surgical guidance: 6D skull tracking and 6D fiducial tracking
Tracking the full range of motion of the target centre is feasible with six degrees of freedom defined in the CT image reference frame. Skull or fiducial tracking is performed using an image processing algorithm which compares pairs of live orthogonal (LAO/RAO) images to sets of equivalent reference images generated from the preoperative CT scan. Both rotational and translational movements of the patient can be tracked for enhanced treatment accuracy.
During initial patient alignment, translational and rotational information is displayed on the patient alignment screen of the control monitor. The tracking information is then used to assist the operator in aligning the patient manually or may be used as inputs to the automatic alignment feature provided by the Axum, five-degree of freedom, patient table. During treatment, changes in alignment within certain limits (±10 mm in x, y and z; ±1° pitch and roll, ± 3° yaw) are compensated for automatically by the robot. If a movement exceeds these limits an error warning is triggered allowing the system to pause for patient realignment.
Fiducial tracking enables treatment in body regions other than skull. Radiopaque fiducial markers (small gold seeds or stainless steel screws) are implanted in close proximity to, or within, the lesion one to seven days prior to preoperative imaging. Based on these reference points, the fiducial tracking algorithm searches the pair of X-ray images, detects the fiducial markers, and calculates translations and rotations of the target. As with skull tracking, when the specified limits are exceeded the system pauses for manual realignment.
While only one fiducial is required to track anatomical translations, three or more are necessary to compute both translational and rotational motions. Fiducial locations are detected in the X-ray images and motion is computed by comparing their positions to the preoperative position of the fiducials as computed in the DRR (Figure 6)
Surgical guidance: with or without compensation for respiration
The CyberKnife system uses a six-axis manipulator for positioning and manoeuvring the compact X-band 6MV linear accelerator to up to 1200 treatment positions defined in the treatment planning phase. Interchangeable collimators, ranging from 5 mm to 60 mm in diameter, deliver radiation in corresponding circular fields at the treatment distance. High end multi-processor computers provide the computational power to control the device for treatment delivery according to the treatment plan.
The robotic manipulator (Figure 7) is capable of positioning the linear accelerator and pointing the beam at the treatment target within a treatment hemisphere at a source to axis distance (SAD) ranging from 100 cm to 65 cm. The robotic targeting precision is better than 0.2 mm. The overall precision of treatment delivery (CT, treatment planning, image guidance, robot and linear accelerator), is better than 0.95 mm (10) for cranial/ spinal lesions and 1.5 mm for moving targets tracked with the Synchrony™ respiratory tracking module. More recent multi-center phantom studies for spinal radiosurgery report a mean error of 0.7±0.3 mm, a treatment delivery precision of 0.3±0.1 mm, and a spine fiducial tracking error of ,0.3 mm for radial translations of up to 14 mm and ,0.7 mm for rotations of up to 4.5° (11) .
Early robot designs assumed fixed targets and environments. Likewise early surgical robots assumed that the anatomy was a rigid target. The ability to sense and react to soft tissue motion (12) which eluded surgical robot design in the past has been solved with Synchrony. Synchrony is an automated lesion tracking system which tracks both external and internal fiducials during the course of treatment and modifies the beam position, in real time, during that tracking process (Figure 5). Traditionally, compensating for respiratory motion of tumours involved including safety margins around the tumour of up to 2 cm or more in the treatment (13, 14). With Synchrony, these margins are significantly reduced, allowing treatment of smaller volumes and the associated sparing of adjacent normal tissues. This motion tracking subsystem provides the capability to maintain the system’s high dose conformality index (CI) while simultaneously treating targets that move with respiration. The manipulator arm is dynamically controlled, compensating for target motion due to respiration, while radiation is being delivered.
The Synchrony respiratory tracking subsystem is one of the most advanced subsystems within the CyberKnife system’s product configuration (15). With Synchrony, implanted fiducials measure internal tumour movement. This is accomplished as follows. Fiducials are imaged preoperatively during the CT scan. During perioperative procedures, fiducials are imaged, detected and located using the pair of orthogonal diagnostic X-ray images. The indirectly computed preoperative CT-based LAO and RAO DRR images are then co-registered to the continuously acquired perioperative LAO and RAO images.
Multiple light emitting diodes (LEDs) are fibre optically connected to optical lenses strategically positioned on the patient’s chest and/or abdomen. These light sources are monitored in real time (32 frames per second) by 3 Charged Coupled Device (CCD) cameras, housed in a single structure and mounted on the ceiling. Classic localization algorithms are used to accurately measure the location of each marker in space. Next a correspondence model based on the locations of both implanted fiducials and LED markers is created. This correspondence model between internal and external movements is updated and verified throughout the treatment by comparing estimated motions and current locations of the fiducials while delays are compensated for with a prediction model. Any displacement errors higher than a user prescribed value, based on an analysis of tumour motion (16), will trigger the system to stop so that adjustments can be made before continuing on with radiosurgery.