- Email: [email protected]
Accuracy evaluation of the optical surface monitoring system on EDGE linear accelerator in a phantom study
Medical Dosimetry ] (2016) ]]]–]]]
Medical Dosimetry journal homepage: www.meddos.org
Accuracy evaluation of the optical surface monitoring system on EDGE linear accelerator in a phantom study Pietro Mancosu, M.Sc., Antonella Fogliata, M.Sc., Antonella Stravato, M.Sc., Stefano Tomatis, M.Sc., Luca Cozzi, Ph.D., and Marta Scorsetti, M.D. Radiotherapy and Radiosurgery Department, Humanitas Clinical and Research Center, Milan-Rozzano, Italy
A R T I C L E I N F O
A B S T R A C T
Article history: Received 7 January 2015 Received in revised form 30 November 2015 Accepted 29 December 2015
Frameless stereotactic radiosurgery (SRS) requires dedicated systems to monitor the patient position during the treatment to avoid target underdosage due to involuntary shift. The optical surface monitoring system (OSMS) is here evaluated in a phantom-based study. The new EDGE linear accelerator from Varian (Varian, Palo Alto, CA) integrates, for cranial lesions, the common cone beam computed tomography (CBCT) and kV-MV portal images to the optical surface monitoring system (OSMS), a device able to detect real-time patient's face movements in all 6 couch axes (vertical, longitudinal, lateral, rotation along the vertical axis, pitch, and roll). We have evaluated the OSMS imaging capability in checking the phantoms' position and monitoring its motion. With this aim, a home-made cranial phantom was developed to evaluate the OSMS accuracy in 4 different experiments: (1) comparison with CBCT in isocenter location, (2) capability to recognize predefined shifts up to 21 or 3 cm, (3) evaluation at different couch angles, (4) ability to properly reconstruct the surface when the linac gantry visually block one of the cameras. The OSMS system showed, with a phantom, to be accurate for positioning in respect to the CBCT imaging system with differences of 0.6 ⫾ 0.3 mm for linear vector displacement, with a maximum rotational inaccuracy of 0.31. OSMS presented an accuracy of 0.3 mm for displacement up to 1 cm and 11, and 0.5 mm for larger displacements. Different couch angles (451 and 901) induced a mean vector uncertainty o 0.4 mm. Coverage of 1 camera produced an uncertainty o 0.5 mm. Translations and rotations of a phantom can be accurately detect with the optical surface detector system. & 2016 American Association of Medical Dosimetrists.
Keywords: Optical surface monitoring system EDGE linear accelerator Patient positioning
Introduction Stereotactic radiosurgery (SRS) was originally developed with frame-based head fixations for a single high dose fraction. Such a methodology allowed a very high precision of target localization and a very rigid patient immobilization; however, it was an invasive approach, and was rarely adopted for fractionated stereotactic radiotherapy (SRT) for that reason. The advantages of using frameless methods include and are not limited to the strongly improved patient comfort, the possibility of fractionating the treatment, and an improved clinical workflow, as frame-based methods require imaging and treatment to occur on the same day.
Reprint requests to Antonella Fogliata, Radiotherapy and Radiosurgery Department, Humanitas Clinical and Research Center, Rozzano-Milan, Italy. E-mail: [email protected] http://dx.doi.org/10.1016/j.meddos.2015.12.003 0958-3947/Copyright Ó 2016 American Association of Medical Dosimetrists
The first approaches of frameless stereotactic treatment used specific thermoplastic masks for immobilization and optical fiducials mounted on a bite block,1 as the sole thermoplastic masks were not suitable to limit head motion, mostly for the potential rotational movement of the patient head. Moreover, bite blocks are not suitable for patients with poor dentition and still suffer from patient discomfort. With the advances in image guidance, the frameless, imageguided SRS and SRT have been clinically implemented with success in many institutions.2-4 The introduction of the cone beam computed tomography (CBCT) into the clinical setting provided a high quality 3D imaging of the real patient anatomy, improving the image guidance. Nevertheless, the CBCT, rotating the gantry around the isocenter during image acquisitions, is limited to the treatment couch position at 01, and cannot acquire images during the treatment.
2
P. Mancosu et al. / Medical Dosimetry ] (2016) ]]]–]]]
Recently, optical surface imaging for verifying position and monitoring the patient motion has been explored, allowing surface image detection at every couch position, and during the treatment. Agreements of the order of 1 mm were achieved.5 With such systems, the detected daily patient surface is registered to a previously recorded reference surface (generated in principle from the CT scan, or at the time of the first or a previous treatment) to calculate the patient displacement in terms of couch coordinates needed to best match the 2 surfaces in a 3D space. With this methodology is also possible to monitor in real time the patient during the treatment and possibly interrupt or correct the treatment when the position accuracy is out of a predefined threshold. Surface optical systems have been studied in phantom to characterize the system and evaluate its performances,5-7 using the AlignRT (VisionRT, London, UK) system. For example Bert et al.7 found a maximum standard deviation of 0.75 mm and 0.11 over the 61 of freedom (6 DOF) of the couch, with a root-mean-square error of the distance between the surfaces evaluated to be 0.65 mm over a wide variety of conditions. A very interesting characteristic of such systems is the almost maskless approach, where thermoplastic immobilization with large opening onto the patient face allow on one side to detect on real time the patient face movement (and not the mask or the movements of the fiducials), and on the other side to improve again the patient comfort. The clinical application of frameless and almost maskless approach in the brain stereotactic treatments presented clinically satisfactory results, as reported for example by Nath et al.,8 or Cerviño et al.9 Similar good results have been achieved also in the extracranial stereotactic treatments, as partial breast irradiation,10 thorax,11 or even for deep-inspiration breath hold gating treatments.12 The linac vendors are today approaching the stereotactic treatments in a more and more integrated flow. It is, e.g., the case of the EDGE linear accelerator from Varian (Varian, Palo Alto, CA), that is a machine specifically developed for such stereotactic treatments. The characteristics of this linac are a high level of precision and the possibility to include positioning systems for stereotactic treatments, as the Calypso system (Varian, Palo Alto, CA) for extracranial SRS/SRT, and the optical surface monitoring system (OSMS)
(Varian, Palo Alto, CA) for frameless, almost maskless brain SRS/ SRT. Aim of this work is to evaluate, on a pure phantom approach, the accuracy of the OSMS system for positioning purposes compared to the widely used CBCT, in the new frame determined by the association of the EDGE linear accelerator and the OSMS system. All the inaccuracies related to the real patient motion are out of the scope of the present study. Methods and Materials The EDGE linear accelerator recently installed at Humanitas Center has the OSMS mounted. The linac is also equipped with 120HD MLC, 3 energies (6 and 10 MV flattening ilter free [FFF], and 6 MV), PerfectPitch (Varian, Palo Alto, CA) 6 DOF couch with Calypso top, and the integrated kV and MV imaging system XI (Varian, Palo Alto, CA). This last system consists on the portal imager PV-aS1200 for MV portal and cine images, and the kV source-detector system, which enables users to acquire radiographic and fluoroscopic images, and CT projections that can be reconstructed as 3D cone beam CT, CBCT, images, 4D-CBCT. System description Optical surface monitoring system The OSMS is a video-based 3D surface imaging system used to detect and reconstruct the skin surface of a patient in 3D before and during the radiotherapy treatment. It consists of 3 ceiling camera units (camera in the following), positioned as shown in Fig. 1: 2 laterally to the treatment couch, and the third centrally located at the foot of the couch. A projector unit projects a red light speckle pattern onto the patient. Overall, 2 image sensors located on either side of the projector acquire the image of the patient and the speckle pattern. A close-range digital speckle photogrammetry13 reconstructs the 3D surface as also depicted in Fig. 1 onto the phantom. With the images from the 3 cameras, the system reconstructs the 3D surface for all the gantry positions, even in the cases where the linac head (rotated around ⫾ 451) interposes between one of the cameras and the isocenter, obscuring the image projection for that camera. A reference surface is generated by importing the body contour from a treatment planning system and based on a CT dataset. The OSMS can also directly capture the reference surface, using the patient imaged on the treatment couch, in cases where the body contour could not be used for surface matching for any reason. Before each treatment session, the matching inside a region of interest (ROI) defined by the user of the acquired and the reference patient surface is proposed. When aligning surfaces in the ROIs, the system calculates a rigid-body transformation, 3 shifts and 3 rotations; a minimization process involves an iterative least squares estimation that progressively updates the shifts and rotations as a 6D vector until the surface distance is minimal. Such distance is a measure of
Fig. 1. Optical surface monitoring system and EDGE linac. Upper picture: the linac and the OSMS setup; lower left: a camera; lower right: the head phantom and the same enlightened with the speckle pattern. (Color version of figure is available online.)
P. Mancosu et al. / Medical Dosimetry ] (2016) ]]]–]]] the misalignment of the 2 surfaces, minimized during the registration process, and is computed by taking a sample of points on the real-time surface and finding for each point the corresponding closest point inside the reference ROI. The distance between these points is then squared and summed. This transformation is finally expressed as a set of new couch coordinates at which the patient position best matches with the reference data, used to adjust the treatment couch position. Cone beam computed tomography The absolute positioning accuracy of the whole positioning unit system (kV source and kV detector) is specified as 1.5 mm in isocenter radius, where the absolute position is defined as the isocenter position of the extracted source/ detector, relative to the isocenter position of the arm movements. During CBCT scan, the treatment head fully rotates around the patient to collect the data and then compute the 3D reconstruction. Owing to its rotational movement, care should be taken to ensure that CBCT acquisition is initiated only where no possibility of the couch colliding with the accelerator or imaging arms, that is guaranteed by the couch positioned at 01. For this reason, CBCT can be acquired only with couch positioned at 01 rotation. Among other options, an automated 3D-3D matching between the acquired CBCT and the reference CT is available. This can be done on all 6 couch axes (vertical, longitudinal, lateral, rotation along the vertical axis [v-rotation in the following], pitch, and roll), choosing a range of HU where to optimize the matching. The here chosen range is defined by the built-in values for bone HU on a volume of interest including the entire head phantom.
3
acquired surface image, selecting a ROI that included the forehead, the eyes, the nose and the cheekbone. This positioning procedure consists on typing the actual couch coordinates in the OSMS “move couch” option. From this position, the system calculates the new couch coordinates according to its matching. The new 6 DOF coordinates of the couch positioning were recorded as “OSMS positioning” as according to the OSMS matching. Then a CBCT was acquired and matched with the planning CT on the bony structures, according to the automated algorithm; such a procedure proposed a shift to be applied on the 6 couch coordinates. Those new 6 DOF couch coordinates were named “CBCT positioning,” and the difference between the OSMS positioning and the CBCT positioning was evaluated as accuracy of the OSMS relative to the CBCT positioning. The procedure has been repeated 10 times (during 3 measurement sessions), with full reposition of the phantom at each repetition, applying small (o 1 cm and o 11) intentional inaccuracies in all the directions to force the repositioning. The couch and the gantry were set at 01. The reference image for OSMS in this step was the body contour structure from the planning CT. The differences between the two positioning systems were recorded as average ⫾ standard deviation, as well as with their maximum and minimum values. A second set of measurements was acquired in similar conditions as above, but with a more inaccurate initial positioning of the phantom, with at least one of the linear direction wrongly displaced by more than 1 cm, and at least one of the rotational directions by more than 11, to have an initial condition far from the optimal matching. The differences between CBCT and OSMS were then evaluated similarly to the first measurements set.
System calibration Optical surface monitoring system The system is calibrated to the linac coordinate system. The system calibration consists in a monthly and daily quality assurance (QA) program to ensure that the cameras are accurately calibrated to the treatment isocenter. The calibration and quality assurance procedures shall be performed using an ad-hoc designed plate, provided by the vendor, to position at isocenter. The calibration plate is a white slab with a matrix of 32 32 black spots of about 1 cm diameter, 2 cm distant (called blobs), 4 of them, numbered and larger than the others (1.7 cm diameter), are positioned on the corners of a square of a side of 10 blobs. In the monthly QA, the blobs have to be accurately visualized and recognized in all the images captured by the cameras once the four larger blobs are manually selected (or automatically recognized from a previous calibration). The daily QA process checks that the cameras are in the correct position (mutually and relative to the calibration plate), and automatically detects the blobs according to the monthly QA calibration files. Cone beam computed tomography The so-called IsoCal (Varian, Palo Alto, CA) calibration is the approach to calibrate the imager isocenters, finding the treatment isocenter (MV), and then to relate the kV isocenter to this location. It uses a special partial transmission plate (mounted on the linac head) and an isocenter calibration phantom consisting of 16 tungsten-bearing balls arranged in a special pattern and embedded in a plastic cylinder, mounted on the front end of the treatment couch through a special adapter. The IsoCal procedure calibrates the MV and kV detector arms to the radiation isocenter, and calculates the corrections for the longitudinal and lateral positions of both the MV and kV imagers for each gantry angle (typically 120 different positions). Phantom setting A home-made phantom was used to evaluate the OSMS accuracy for positioning procedures. It was composed by a cranial bony-like structure, filled with mold material to cover the orbital and nose cavities to mimic the face surface (Fig. 1). The phantom was positioned onto the treatment couch with no masks and no frame, and was scanned on a Philips Brilliance CT scanner with a slice spacing of 1 mm as clinically used in our institute for SRS plans. The CT dataset has been imported in the Eclipse treatment planning system (Varian, Palo Alto, CA), and a body contour covering the whole phantom head was precisely delineated as a structure. For such a contouring, the automatic body detection tool was used with a lower threshold of 350 HU; then, being in the current study the nasal region the most critical surfaces precisely delineate, a manual and careful correction was there applied when needed. A virtual plan was finally prepared to define an isocenter in the phantom.
Evaluation of OSMS response to predefined inaccuracies After the phantom repositioning according to CBCT matching with CT planning, the image read by the OSMS cameras in the new position was set as reference image, zeroing in this way the relative difference between CBCT and OSMS positioning. Predefined shifts from those reference coordinates have been applied to the couch for each of the 6 directions separately, 1 at a time as follows: the linear shifts, i.e., vertical, longitudinal, and lateral with magnitude 3, 2, 1, 0.2, 0.1, 0.1, 0.2, 1, 2, and 3 cm; the rotational shifts, i.e., v-rotation, pitch, and roll with magnitude 2, 1, 0.5, 0.3, 0.2, 0.1, 0.1, 0.2, 0.3, 0.5, 1, and 21. The suggested OSMS shifts in 6 DOF were recorded. The differences between the predefined applied shifts and the OSMS shifts reflect the accuracy of OSMS repositioning in a range of ⫾ 3 cm for the linear movements, and ⫾ 21 for the rotational movements. Finally, 30 randomly assigned shifts simultaneously in all the 6 directions were applied; the random displacement was chosen to be within 0.5 cm and 0.51 in one third of the cases, within 1 cm and 11 in a second third, and within 2 cm and 21 in the remaining. The differences between the predefined applied shifts and the OSMS shifts were recorded. All the above measurements were repeated 5 times with a complete reposition of the phantom before each measurement set for a total of 540 different measurements. The reference image for the above measurements was the OSMS captured image.
Evaluation of OSMS response at different couch angles The same measurements just described were acquired also with the couch pedestal rotated at 901 and 451, to evaluate the ability of OSMS to reconstruct the 6 coordinates and their directions properly. A total of 96 acquisitions were performed for each new couch angle. The reference image for this evaluation step is the OSMS captured image. The couch angle value is typed into the OSMS software to allow the system to reconstruct the image properly.
Evaluation of OSMS response with two of three images The same acquisitions were again repeated, but with couch positioned at 01 and gantry at 451, to evaluate the ability of OSMS to reconstruct 3D surface properly from only 2 images instead of three, as the linac head is interposed between one of the cameras and the isocenter. A total of 96 acquisitions were performed. The reference image for this evaluation step is the OSMS captured image.
Results OSMS evaluation Overall, 4 different phantom-based experiments were set up to evaluate the OSMS system. Evaluation of OSMS relative to CBCT patient positioning The phantom was positioned at the isocenter (simple visual positioning, allowing potentially large displacements), and subsequently prepositioned with the OSMS system, by matching the CT body contour imported from Eclipse and the
Systems calibrations Optical surface monitoring system A summary of the reports of the daily QA checks of the OSMS on the period from March to July 2014 is presented in Fig. 2 for the errors of the single cameras, and the relative error of the entire system.
4
P. Mancosu et al. / Medical Dosimetry ] (2016) ]]]–]]] Table 1 Differences between OSMS and CBCT positioning in the 6 directions, for small and large initial mispositioning. Small mispositioning (all o 1 cm and o 11) Direction Vertical (mm) Longitudinal (mm) Lateral (mm) V-rotation (deg.) Pitch (deg.) Roll (deg.)
Average ⫾ 1 SD 0.22 ⫾ 0.24 0.17 ⫾ 0.13 0.06 ⫾ 0.28 0.06 ⫾ 0.11 0.06 ⫾ 0.08 0.00 ⫾ 0.07
Large mispositioning (at least one 4 1 cm and 4 11)
Range Average ⫾ 1 SD 0.3, 0.6 0.39 ⫾ 0.44 0.5, 0.1 0.00 ⫾ 0.21 0.5, 0.3 0.18 ⫾ 0.45 0.2, 0.1 0.11 ⫾ 0.11 0.2, 0.1 0.10 ⫾ 0.09 0.1, 0.1 0.02 ⫾ 0.11
Range 0.3, 1.0 0.5, 0.2 1.1, 0.4 0.3, 0.1 0.2, 0.1 0.2, 0.1
SD ¼ standard deviation.
the intrinsic accuracy of the two systems according to the calibration and check procedures. The second set of measurements, with coarser initial phantom positioning, presents a higher uncertainty for the linear direction. The average of the vector lengths from the linear direction distances were 0.45 ⫾ 0.13 and 0.68 ⫾ 0.35 mm for the small and large initial displacements, with an overall average of 0.57 ⫾ 0.28 mm.
Fig. 2. Errors from the daily QA procedure, from March to July, 2014. (Color version of figure is available online.)
The error of the single cameras is the distance from the current calibration plate position to the monthly calibration plate position. This error was 1.0 ⫾ 0.9 mm for each camera. The relative error is here the difference in 3D motion computed by a pair of cameras and is a measure of the inconsistency in movement detection between cameras. This error was o 0.5 mm in more than 98% of the tests, with an average value of 0.12 ⫾ 0.12 mm. Cone beam computed tomography The kV imaging system accuracy according to the IsoCal procedure presented a maximum shift of 0.11 mm and an imager rotation of 0.051. Such a shift has to be combined with the treatment isocenter radius evaluated with the Winston-Lutz test, that was measured as of 0.31 mm for gantry and collimator rotations, and 0.54 mm including the couch rotation. These bring to an accuracy in the positioning of 0.33 and 0.56 mm without and with couch inclusion, respectively. OSMS evaluation The positioning distance is defined as the difference between the OSMS and the expected shifts derived from the CBCT acquisition or from a predefined applied shift. Evaluation of OSMS relative to CBCT patient positioning The distances between OSMS and CBCT positioning in the 6 directions are summarized in Table 1, in the 2 groups of small and large initial shifts. Those values reflect the good agreement between the 2 systems, with a maximum linear distance of 0.6 mm and a maximum rotational distance of 0.21, and a mean that stays inside
Evaluation of OSMS response to predefined inaccuracies The distances from the predefined shifts are summarized in Fig. 3, where all the 6 directions are presented for the shifts applied in each single direction. Table 2 reports the average distances. The rotational directions present the largest distances when linear shifts are applied, mainly the roll, followed by the pitch; conversely, the rotational distances are smaller in situations where an intentional rotational shift was applied. The linear shifts are similar for displacements in all the directions, keeping the isocenter position within 0.5 mm, with the vertical direction showing in some conditions the worst case. The results of all the predefined random shifts in 6 directions at a time (150 acquisitions) were within 0.1 ⫾ 0.2 mm for the linear, and within 0.1 ⫾ 0.11 for the rotational directions. The average vector length determined by the linear direction displacements was 0.4 ⫾ 0.2 mm. In Fig. 4(A), the distances are stratified for the 0.5, 1, and 2 cm and degree maximum shifts, and summarized for all the directions, as average distances and standard deviations in each of the three sets. Such results confirm what analyzed in the single direction shifts: (1) lower accuracy in determining the roll misplacement, which could worsen when increasing the displacement, (2) linear displacements within 0.5 mm. Moreover, there is no clear trend in the distance when larger displacements are applied. The average of the vector length determined from the linear direction distances were 0.30 ⫾ 0.14, 0.33 ⫾ 0.12, 0.50 ⫾ 0.28 mm for the 3 sets of 0.5, 1, and 2 cm and degree maximum shifts, respectively. Evaluation of OSMS response at different couch angles The distances from the applied intentional shifts on single directions up to 3 cm or 21 (on 66 acquisitions for each couch angle) were in average 0.0 ⫾ 0.3 mm and 0.00 ⫾ 0.051 for 901 couch angle, and 0.0 ⫾ 0.2 mm and 0.01 ⫾ 0.071 for 451 couch angle, proving the ability of the OSMS to properly manage the phantom and the coordinate system rotation. The shifts applied in all directions for the 3 sets o 0.5, o 1, o 2 cm and degrees, respectively (10 acquisitions each set) are reported in Figs. 4(B) and (D), presenting an average of the vector length determined from the linear direction distances of 0.20 ⫾ 0.07, 0.31 ⫾ 0.16, 0.39 ⫾ 0.22 mm for the 3 sets of shifts and 901 couch, and 0.11 ⫾ 0.09, 0.24 ⫾ 0.08, 0.21 ⫾ 0.08 mm with 451 couch.
P. Mancosu et al. / Medical Dosimetry ] (2016) ]]]–]]]
5
Fig. 3. Distances in all the 6 directions, from intentional shifts in one direction only: linear shifts in the upper row, rotational in the lower row. (Color version of figure is available online.)
Table 2 Distances from predefined 1 direction shifts. Data represent the average distances ⫾ 1SD in all the 6 directions, for each predefined shift direction VERT shift Residual Residual Residual Residual Residual Residual
error error error error error error
on on on on on on
VERT (cm) LONG (cm) LAT (cm) V-ROT (deg.) pitch (deg.) roll (deg.)
0.00 0.00 0.01 0.03 0.03 0.04
⫾ ⫾ ⫾ ⫾ ⫾ ⫾
0.01 0.01 0.01 0.02 0.02 0.07
LONG shift
LAT shift
0.01 0.00 0.01 0.02 0.02 0.03
0.01 0.00 0.01 0.03 0.00 0.05
⫾ ⫾ ⫾ ⫾ ⫾ ⫾
0.01 0.01 0.01 0.06 0.04 0.07
⫾ ⫾ ⫾ ⫾ ⫾ ⫾
ROT shift 0.01 0.01 0.01 0.05 0.05 0.07
0.01 0.00 0.01 0.03 0.01 0.02
⫾ ⫾ ⫾ ⫾ ⫾ ⫾
0.02 0.01 0.01 0.02 0.02 0.04
PITCH shift
ROLL shift
0.01 0.00 0.01 0.02 0.00 0.03
0.00 0.00 0.01 0.03 0.01 0.01
⫾ ⫾ ⫾ ⫾ ⫾ ⫾
0.01 0.01 0.02 0.03 0.03 0.03
⫾ ⫾ ⫾ ⫾ ⫾ ⫾
0.01 0.01 0.01 0.02 0.02 0.04
VERT ¼ vertical; LONG ¼ longitudinal; LAT ¼ lateral, V-ROT ¼ rotation along the vertical axis.
Fig. 4. Distances of the predefined randomly chosen shifts in all the 6 directions. Results are stratified for shifts o 0.5, o 1, o 2 cm or degree, respectively. Error bars refer to 1 SD over all measurements. (A) Gantry and couch set at 01; (B) gantry set at 01, couch at 901; (C) gantry set at 451, couch at 01; and (D) gantry set at 01, couch at 451.
6
P. Mancosu et al. / Medical Dosimetry ] (2016) ]]]–]]]
Evaluation of OSMS response with 2 of 3 images The distances from applied intentional shifts on single directions up to 3 cm or 21 (on 66 acquisitions) with gantry set at 451 and couch 01 (image reconstructed with 2 of 3 cameras) were in average 0.1 ⫾ 0.8 mm and 0.01 ⫾ 0.101, similar to the acquisitions with 3 cameras, but with a higher standard deviation. The shifts applied in all directions for the 3 sets o 0.5, o 1, o 2 cm and degrees, respectively (10 acquisitions each set) are reported in Fig. 4(C), presenting an average of the vector length determined from the linear direction distances of 0.46 ⫾ 0.19, 0.34 ⫾ 0.19, 0.36 ⫾ 0.25 mm for the three sets of shifts.
Discussion This work evaluated, on a phantom base, the accuracy of using the surface imaging system OSMS as implemented on the recently installed EDGE linac system. The results of the OSMS relative to the CBCT positioning presented an average difference of 0.6 ⫾ 0.3 mm and maximum rotational inaccuracy of 0.31, including small and large initial positioning. Such results should be compared to what obtained by Peng et al.,6 where the mean vector difference was of 0.2 ⫾ 0.3 mm and the maximum rotational difference was 1.31. The results from that group were obtained with the AlignRT system compared to an Elekta kV CBCT. The data shown in the present study refer to a new frame composed by the OSMS system installed on an EDGE linear accelerator. This compares the optical system with the CBCT imaging system of a linear accelerator specifically designed as a comprehensive radiosurgery system, focused on very high precision. The accuracy of the OSMS skin surface registration relative to the CBCT internal anatomy match showed to be of the same order. However, a robust flow and procedure should be set to use the OSMS system routinely. For fractionated stereotactic treatments, a patient prepositioning via OSMS by matching the skin surface with the body contour from the planning CT could be suggested, followed by a verification and evaluation with a CBCT of the internal anatomy. After having applied eventual shifts according to the CBCT matching, a new reference image for further OSMS monitoring and positioning should be captured, being the correlation between the actual patient skin surface and the anatomical match guaranteed by the CBCT imaging. For patient surface detection, the fixation mask should be open enough to allow skin detection, while decreasing the mask detection. In this way the positioning and monitoring is relative to the patient movement instead of the mask (or external markers) displacement, enforcing the direct patient positioning check while minimizing those of indirect external surrogates (mask and markers). The usage of open masks is on one side a technical need (the actual check is of the patient surface), and on the other side it improves the patient comfort, allowing better patient compliance. Of course, an optimal balance of the open and closed surface of the mask has to be evaluated and found, as too open masks could increase patient ability to move, reducing hence position accuracy. The acquisition and surface reconstruction time is a very fast procedure. As soon as the OSMS is fully integrated in the EDGE unit, with the possibility to directly apply the shifts to the couch movements, as it is today with the CBCT imaging, the OSMS usage require very short time, requiring no extra time for the treatment session. In summary, the efficient surface detection and matching is achievable with improved patient comfort, and without detriment to the accuracy of treatment delivery, treatment times, or clinical outcomes.2-4,8,9 More investigations are needed to determine the correlation and accuracy between internal patient anatomy and superficial positioning in real clinical cases, where the possible
patient movement and anatomy changes would play a role independent from the pure phantom-based accuracy of the system, as the current study. An interesting point related to the surface matching is the choice of the ROI where to register the acquired and the reference 3D surfaces. In the present study, the ROI included entirely the nose, the cheeks, the eyes and the forehead, and was chosen to have a quite large surface with the maximum variation of heights and distances. A matching of smooth surfaces could easily be affected by positional errors that the mathematical registration cannot solve. The choice of a good ROI should in principle improve the matching results, being a good ROI a region that includes suitable surface features not invariant for rotations or translations. The facial surface is in this respect a good candidate, whereas, e.g., large flat surfaces, like the abdomen, should be less suitable for surface matching. Anyway, care should also be taken in regions where possible movement can happen, as it could be the case of the eyelids involuntary movement. To consider is the fact that the matching is between the OSMS acquisition and the patient surface as reconstructed from the body contouring delineated in the treatment planning system. Clearly, the accuracy of such a contouring in the ROI region is crucial for a proper match, and for that reason care has to be taken in drawing or finding the most appropriate tool for delineation, depending also on the patient. The body contour can also be considered the baseline of the patient, from which evidentiate possible changes in his/her anatomy, especially for fractionated treatments. Peng et al.6 found that the smallest registration variance for isocenter displacement with optical surface reconstruction was shown for the upper frontal only part of the face, or on one side only of the face, that are indeed the facial surfaces presenting the least height differences. Moreover, Wiersma et al.14 discussed that the time needed for surface reconstruction scales with the ROI size, hence the ROI has to be cautiously chosen, following the surface features more than simply the ROI size. A further step of the evaluation of OSMS in the EDGE frame is the use of surface image reconstruction for the patient monitoring during the treatment, including the possibility to hold the beam when the matching is out of a predefined threshold. Such experiments are under investigation in our group, but are out of the scope of the current study, that is phantom based. To consider is also the high dose per fraction of the stereotactic treatments. EDGE is equipped with FFF beams, i.e., with high dose rate (1400 and 2400 MU/min for 6 and 10 MV FFF, respectively). The treatment time, in terms of beam-on time, is hence greatly reduced up to a factor 4 for 10 MV FFF. Shorter treatment time could also imply lesser patient movement during the delivery. The major limitation of the current work is its focus only on phantom measurements. In this preliminary phase of studying the accuracy of the OSMS installed on the EDGE linac, the patient was not included in the flow, to evaluate the system per se, with no additional factors. However, the final use of the system is to detect patient positioning inaccuracies, where many additional factors concur to the results obtainable when a patient, and not a phantom, is positioned on the treatment couch. Those factors can refer to the surface detection and reconstruction system in following the continuous deformations that the same surface is undergoing due to the unavoidable small changes coming from unwitting contractures of the facial muscles, or eye movements. This could potentially adversely affect the real-time matching between the reference and the reconstructed surface, as the matching could be inadvertently be anchored to some anatomical points affecting the registration quality. Another important factor related to patient positioning is the surface changes that could occur from session to session, or from CT planning acquisition to the treatment session, changes due to anatomy, coming for
P. Mancosu et al. / Medical Dosimetry ] (2016) ]]]–]]]
example to a different contracture or relaxation of the facial muscles, even if stable throughout the entire treatment session. Studies on the usage of the OSMS with patients are under processing in our group to evaluate the accuracy of the system when used in its primary scope, distinguishing at that stage which is the accuracy of the system per se, and quantifying the uncertainty coming from the critical points related to patient positioning.
Conclusions The OSMS system under investigation showed to be accurate to determine phantom positioning respect to the CBCT imaging system: translations and rotations of a rigid and fixed phantom can be detected accurately with the optical surface detection system.
Acknowledgments L. Cozzi acts as Scientific Advisor to Varian Medical Systems and is Clinical Research Scientist at Humanitas Cancer Center. References 1. Bova, F.J.; Meeks, S.L.; Friedman, W.A.; et al. Optic-guided stereotactic radiotherapy. Med. Dosim. 23:221–8; 1998. 2. Breneman, J.C.; Steinmetz, R.; Smith, A.; et al. Frameless image-guided intracranial stereotactic radiosurgery: clinical outcomes for brain metastases. Int. J. Radiat. Oncol. Biol. Phys. 74:702–6; 2009.
7
3. Ramakrishna, N.; Rosca, F.; Friesen, S.; et al. A clinical comparison of patient setup and intra-fraction motion using frame-based radiosurgery versus a frameless image-guided radiosurgery system for intracranial lesions. Radiother. Oncol. 95:109–15; 2010. 4. Ohtakara, K.; Hayashi, S.; Tanaka, H.; et al. Clinical comparison of positional accuracy and stability between dedicated versus conventional masks for immobilization in cranial stereotactic radiotherapy using 6-degree-offreedom image guidance system-integrated platform. Radiother. Oncol. 102:198–205; 2012. 5. Cerviño, L.I.; Pawlicki, T.; Lawson, J.D.; et al. Frame-less and mask-less cranial stereotactic radiosurgery: a feasibility study. Phys. Med. Biol. 55:1863–73; 2010. 6. Peng, J.L.; Kahler, D.; Li, J.G.; et al. Characterization of a real-time suface imageguided stereotactic positioning system. Med. Phys. 37:5421–33; 2010. 7. Bert, C.; Metheany, K.G.; Doppke, K.; et al. A phantom evaluation of a stereovision surface imaging system for radiotherapy patient setup. Med. Phys. 32:2753–62; 2005. 8. Nath, S.K.; Lawson, J.D.; Wang, J.Z.; et al. Optically-guided frameless linac-based radiosurgery for brain metastases: clinical experience. J. Neurooncol. 97:67–72; 2010. 9. Cerviño, L.I.; Detorie, N.; Taylor, M.; et al. Initial clinical experience with a frameless and maskless stereotactic radiosurgery treatment. Pract. Radiat. Oncol 2:54–62; 2012. 10. Bert, C.; Metheany, K.G.; Doppke, K.P.; et al. Clinical experience with a 3D surface patient setup system for alignment of partial-breast irradiation patients. Int. J. Radiat. Oncol. Biol. Phys. 64:1265–74; 2006. 11. Schöffel, P.J.; Harms, W.; Sroka-Perez, G.; et al. Accuracy of a commercial optical 3D surface imaging system for realignment of patients for radiotherapy of the thorax. Phys. Med. Biol. 52:3949–63; 2007. 12. Cerviño, L.I.; Gupta, S.; Rose, M.A.; et al. Using surface imaging and visual coaching to improve the reproducibility and stability of deep-inspiration breath hold for left-breast-cancer radiotherapy. Phys. Med. Biol. 54:6853–65; 2009. 13. Rogus, R.D.; Stern, R.L.; Kubo, H.D. Accuracy of a photogrammetry-based patient positioning and monitoring system for radiation therapy. Med. Phys. 26:721–8; 1999. 14. Wiersma, R.D.; Tomarken, S.L.; Grelewicz, Z.; et al. Spatial and temporal performance of 3D optical surface imaging for real-time head position tracking. Med. Phys. 40(111712-1):8; 2013.
Medical Dosimetry journal homepage: www.meddos.org
Accuracy evaluation of the optical surface monitoring system on EDGE linear accelerator in a phantom study Pietro Mancosu, M.Sc., Antonella Fogliata, M.Sc., Antonella Stravato, M.Sc., Stefano Tomatis, M.Sc., Luca Cozzi, Ph.D., and Marta Scorsetti, M.D. Radiotherapy and Radiosurgery Department, Humanitas Clinical and Research Center, Milan-Rozzano, Italy
A R T I C L E I N F O
A B S T R A C T
Article history: Received 7 January 2015 Received in revised form 30 November 2015 Accepted 29 December 2015
Frameless stereotactic radiosurgery (SRS) requires dedicated systems to monitor the patient position during the treatment to avoid target underdosage due to involuntary shift. The optical surface monitoring system (OSMS) is here evaluated in a phantom-based study. The new EDGE linear accelerator from Varian (Varian, Palo Alto, CA) integrates, for cranial lesions, the common cone beam computed tomography (CBCT) and kV-MV portal images to the optical surface monitoring system (OSMS), a device able to detect real-time patient's face movements in all 6 couch axes (vertical, longitudinal, lateral, rotation along the vertical axis, pitch, and roll). We have evaluated the OSMS imaging capability in checking the phantoms' position and monitoring its motion. With this aim, a home-made cranial phantom was developed to evaluate the OSMS accuracy in 4 different experiments: (1) comparison with CBCT in isocenter location, (2) capability to recognize predefined shifts up to 21 or 3 cm, (3) evaluation at different couch angles, (4) ability to properly reconstruct the surface when the linac gantry visually block one of the cameras. The OSMS system showed, with a phantom, to be accurate for positioning in respect to the CBCT imaging system with differences of 0.6 ⫾ 0.3 mm for linear vector displacement, with a maximum rotational inaccuracy of 0.31. OSMS presented an accuracy of 0.3 mm for displacement up to 1 cm and 11, and 0.5 mm for larger displacements. Different couch angles (451 and 901) induced a mean vector uncertainty o 0.4 mm. Coverage of 1 camera produced an uncertainty o 0.5 mm. Translations and rotations of a phantom can be accurately detect with the optical surface detector system. & 2016 American Association of Medical Dosimetrists.
Keywords: Optical surface monitoring system EDGE linear accelerator Patient positioning
Introduction Stereotactic radiosurgery (SRS) was originally developed with frame-based head fixations for a single high dose fraction. Such a methodology allowed a very high precision of target localization and a very rigid patient immobilization; however, it was an invasive approach, and was rarely adopted for fractionated stereotactic radiotherapy (SRT) for that reason. The advantages of using frameless methods include and are not limited to the strongly improved patient comfort, the possibility of fractionating the treatment, and an improved clinical workflow, as frame-based methods require imaging and treatment to occur on the same day.
Reprint requests to Antonella Fogliata, Radiotherapy and Radiosurgery Department, Humanitas Clinical and Research Center, Rozzano-Milan, Italy. E-mail: [email protected] http://dx.doi.org/10.1016/j.meddos.2015.12.003 0958-3947/Copyright Ó 2016 American Association of Medical Dosimetrists
The first approaches of frameless stereotactic treatment used specific thermoplastic masks for immobilization and optical fiducials mounted on a bite block,1 as the sole thermoplastic masks were not suitable to limit head motion, mostly for the potential rotational movement of the patient head. Moreover, bite blocks are not suitable for patients with poor dentition and still suffer from patient discomfort. With the advances in image guidance, the frameless, imageguided SRS and SRT have been clinically implemented with success in many institutions.2-4 The introduction of the cone beam computed tomography (CBCT) into the clinical setting provided a high quality 3D imaging of the real patient anatomy, improving the image guidance. Nevertheless, the CBCT, rotating the gantry around the isocenter during image acquisitions, is limited to the treatment couch position at 01, and cannot acquire images during the treatment.
2
P. Mancosu et al. / Medical Dosimetry ] (2016) ]]]–]]]
Recently, optical surface imaging for verifying position and monitoring the patient motion has been explored, allowing surface image detection at every couch position, and during the treatment. Agreements of the order of 1 mm were achieved.5 With such systems, the detected daily patient surface is registered to a previously recorded reference surface (generated in principle from the CT scan, or at the time of the first or a previous treatment) to calculate the patient displacement in terms of couch coordinates needed to best match the 2 surfaces in a 3D space. With this methodology is also possible to monitor in real time the patient during the treatment and possibly interrupt or correct the treatment when the position accuracy is out of a predefined threshold. Surface optical systems have been studied in phantom to characterize the system and evaluate its performances,5-7 using the AlignRT (VisionRT, London, UK) system. For example Bert et al.7 found a maximum standard deviation of 0.75 mm and 0.11 over the 61 of freedom (6 DOF) of the couch, with a root-mean-square error of the distance between the surfaces evaluated to be 0.65 mm over a wide variety of conditions. A very interesting characteristic of such systems is the almost maskless approach, where thermoplastic immobilization with large opening onto the patient face allow on one side to detect on real time the patient face movement (and not the mask or the movements of the fiducials), and on the other side to improve again the patient comfort. The clinical application of frameless and almost maskless approach in the brain stereotactic treatments presented clinically satisfactory results, as reported for example by Nath et al.,8 or Cerviño et al.9 Similar good results have been achieved also in the extracranial stereotactic treatments, as partial breast irradiation,10 thorax,11 or even for deep-inspiration breath hold gating treatments.12 The linac vendors are today approaching the stereotactic treatments in a more and more integrated flow. It is, e.g., the case of the EDGE linear accelerator from Varian (Varian, Palo Alto, CA), that is a machine specifically developed for such stereotactic treatments. The characteristics of this linac are a high level of precision and the possibility to include positioning systems for stereotactic treatments, as the Calypso system (Varian, Palo Alto, CA) for extracranial SRS/SRT, and the optical surface monitoring system (OSMS)
(Varian, Palo Alto, CA) for frameless, almost maskless brain SRS/ SRT. Aim of this work is to evaluate, on a pure phantom approach, the accuracy of the OSMS system for positioning purposes compared to the widely used CBCT, in the new frame determined by the association of the EDGE linear accelerator and the OSMS system. All the inaccuracies related to the real patient motion are out of the scope of the present study. Methods and Materials The EDGE linear accelerator recently installed at Humanitas Center has the OSMS mounted. The linac is also equipped with 120HD MLC, 3 energies (6 and 10 MV flattening ilter free [FFF], and 6 MV), PerfectPitch (Varian, Palo Alto, CA) 6 DOF couch with Calypso top, and the integrated kV and MV imaging system XI (Varian, Palo Alto, CA). This last system consists on the portal imager PV-aS1200 for MV portal and cine images, and the kV source-detector system, which enables users to acquire radiographic and fluoroscopic images, and CT projections that can be reconstructed as 3D cone beam CT, CBCT, images, 4D-CBCT. System description Optical surface monitoring system The OSMS is a video-based 3D surface imaging system used to detect and reconstruct the skin surface of a patient in 3D before and during the radiotherapy treatment. It consists of 3 ceiling camera units (camera in the following), positioned as shown in Fig. 1: 2 laterally to the treatment couch, and the third centrally located at the foot of the couch. A projector unit projects a red light speckle pattern onto the patient. Overall, 2 image sensors located on either side of the projector acquire the image of the patient and the speckle pattern. A close-range digital speckle photogrammetry13 reconstructs the 3D surface as also depicted in Fig. 1 onto the phantom. With the images from the 3 cameras, the system reconstructs the 3D surface for all the gantry positions, even in the cases where the linac head (rotated around ⫾ 451) interposes between one of the cameras and the isocenter, obscuring the image projection for that camera. A reference surface is generated by importing the body contour from a treatment planning system and based on a CT dataset. The OSMS can also directly capture the reference surface, using the patient imaged on the treatment couch, in cases where the body contour could not be used for surface matching for any reason. Before each treatment session, the matching inside a region of interest (ROI) defined by the user of the acquired and the reference patient surface is proposed. When aligning surfaces in the ROIs, the system calculates a rigid-body transformation, 3 shifts and 3 rotations; a minimization process involves an iterative least squares estimation that progressively updates the shifts and rotations as a 6D vector until the surface distance is minimal. Such distance is a measure of
Fig. 1. Optical surface monitoring system and EDGE linac. Upper picture: the linac and the OSMS setup; lower left: a camera; lower right: the head phantom and the same enlightened with the speckle pattern. (Color version of figure is available online.)
P. Mancosu et al. / Medical Dosimetry ] (2016) ]]]–]]] the misalignment of the 2 surfaces, minimized during the registration process, and is computed by taking a sample of points on the real-time surface and finding for each point the corresponding closest point inside the reference ROI. The distance between these points is then squared and summed. This transformation is finally expressed as a set of new couch coordinates at which the patient position best matches with the reference data, used to adjust the treatment couch position. Cone beam computed tomography The absolute positioning accuracy of the whole positioning unit system (kV source and kV detector) is specified as 1.5 mm in isocenter radius, where the absolute position is defined as the isocenter position of the extracted source/ detector, relative to the isocenter position of the arm movements. During CBCT scan, the treatment head fully rotates around the patient to collect the data and then compute the 3D reconstruction. Owing to its rotational movement, care should be taken to ensure that CBCT acquisition is initiated only where no possibility of the couch colliding with the accelerator or imaging arms, that is guaranteed by the couch positioned at 01. For this reason, CBCT can be acquired only with couch positioned at 01 rotation. Among other options, an automated 3D-3D matching between the acquired CBCT and the reference CT is available. This can be done on all 6 couch axes (vertical, longitudinal, lateral, rotation along the vertical axis [v-rotation in the following], pitch, and roll), choosing a range of HU where to optimize the matching. The here chosen range is defined by the built-in values for bone HU on a volume of interest including the entire head phantom.
3
acquired surface image, selecting a ROI that included the forehead, the eyes, the nose and the cheekbone. This positioning procedure consists on typing the actual couch coordinates in the OSMS “move couch” option. From this position, the system calculates the new couch coordinates according to its matching. The new 6 DOF coordinates of the couch positioning were recorded as “OSMS positioning” as according to the OSMS matching. Then a CBCT was acquired and matched with the planning CT on the bony structures, according to the automated algorithm; such a procedure proposed a shift to be applied on the 6 couch coordinates. Those new 6 DOF couch coordinates were named “CBCT positioning,” and the difference between the OSMS positioning and the CBCT positioning was evaluated as accuracy of the OSMS relative to the CBCT positioning. The procedure has been repeated 10 times (during 3 measurement sessions), with full reposition of the phantom at each repetition, applying small (o 1 cm and o 11) intentional inaccuracies in all the directions to force the repositioning. The couch and the gantry were set at 01. The reference image for OSMS in this step was the body contour structure from the planning CT. The differences between the two positioning systems were recorded as average ⫾ standard deviation, as well as with their maximum and minimum values. A second set of measurements was acquired in similar conditions as above, but with a more inaccurate initial positioning of the phantom, with at least one of the linear direction wrongly displaced by more than 1 cm, and at least one of the rotational directions by more than 11, to have an initial condition far from the optimal matching. The differences between CBCT and OSMS were then evaluated similarly to the first measurements set.
System calibration Optical surface monitoring system The system is calibrated to the linac coordinate system. The system calibration consists in a monthly and daily quality assurance (QA) program to ensure that the cameras are accurately calibrated to the treatment isocenter. The calibration and quality assurance procedures shall be performed using an ad-hoc designed plate, provided by the vendor, to position at isocenter. The calibration plate is a white slab with a matrix of 32 32 black spots of about 1 cm diameter, 2 cm distant (called blobs), 4 of them, numbered and larger than the others (1.7 cm diameter), are positioned on the corners of a square of a side of 10 blobs. In the monthly QA, the blobs have to be accurately visualized and recognized in all the images captured by the cameras once the four larger blobs are manually selected (or automatically recognized from a previous calibration). The daily QA process checks that the cameras are in the correct position (mutually and relative to the calibration plate), and automatically detects the blobs according to the monthly QA calibration files. Cone beam computed tomography The so-called IsoCal (Varian, Palo Alto, CA) calibration is the approach to calibrate the imager isocenters, finding the treatment isocenter (MV), and then to relate the kV isocenter to this location. It uses a special partial transmission plate (mounted on the linac head) and an isocenter calibration phantom consisting of 16 tungsten-bearing balls arranged in a special pattern and embedded in a plastic cylinder, mounted on the front end of the treatment couch through a special adapter. The IsoCal procedure calibrates the MV and kV detector arms to the radiation isocenter, and calculates the corrections for the longitudinal and lateral positions of both the MV and kV imagers for each gantry angle (typically 120 different positions). Phantom setting A home-made phantom was used to evaluate the OSMS accuracy for positioning procedures. It was composed by a cranial bony-like structure, filled with mold material to cover the orbital and nose cavities to mimic the face surface (Fig. 1). The phantom was positioned onto the treatment couch with no masks and no frame, and was scanned on a Philips Brilliance CT scanner with a slice spacing of 1 mm as clinically used in our institute for SRS plans. The CT dataset has been imported in the Eclipse treatment planning system (Varian, Palo Alto, CA), and a body contour covering the whole phantom head was precisely delineated as a structure. For such a contouring, the automatic body detection tool was used with a lower threshold of 350 HU; then, being in the current study the nasal region the most critical surfaces precisely delineate, a manual and careful correction was there applied when needed. A virtual plan was finally prepared to define an isocenter in the phantom.
Evaluation of OSMS response to predefined inaccuracies After the phantom repositioning according to CBCT matching with CT planning, the image read by the OSMS cameras in the new position was set as reference image, zeroing in this way the relative difference between CBCT and OSMS positioning. Predefined shifts from those reference coordinates have been applied to the couch for each of the 6 directions separately, 1 at a time as follows: the linear shifts, i.e., vertical, longitudinal, and lateral with magnitude 3, 2, 1, 0.2, 0.1, 0.1, 0.2, 1, 2, and 3 cm; the rotational shifts, i.e., v-rotation, pitch, and roll with magnitude 2, 1, 0.5, 0.3, 0.2, 0.1, 0.1, 0.2, 0.3, 0.5, 1, and 21. The suggested OSMS shifts in 6 DOF were recorded. The differences between the predefined applied shifts and the OSMS shifts reflect the accuracy of OSMS repositioning in a range of ⫾ 3 cm for the linear movements, and ⫾ 21 for the rotational movements. Finally, 30 randomly assigned shifts simultaneously in all the 6 directions were applied; the random displacement was chosen to be within 0.5 cm and 0.51 in one third of the cases, within 1 cm and 11 in a second third, and within 2 cm and 21 in the remaining. The differences between the predefined applied shifts and the OSMS shifts were recorded. All the above measurements were repeated 5 times with a complete reposition of the phantom before each measurement set for a total of 540 different measurements. The reference image for the above measurements was the OSMS captured image.
Evaluation of OSMS response at different couch angles The same measurements just described were acquired also with the couch pedestal rotated at 901 and 451, to evaluate the ability of OSMS to reconstruct the 6 coordinates and their directions properly. A total of 96 acquisitions were performed for each new couch angle. The reference image for this evaluation step is the OSMS captured image. The couch angle value is typed into the OSMS software to allow the system to reconstruct the image properly.
Evaluation of OSMS response with two of three images The same acquisitions were again repeated, but with couch positioned at 01 and gantry at 451, to evaluate the ability of OSMS to reconstruct 3D surface properly from only 2 images instead of three, as the linac head is interposed between one of the cameras and the isocenter. A total of 96 acquisitions were performed. The reference image for this evaluation step is the OSMS captured image.
Results OSMS evaluation Overall, 4 different phantom-based experiments were set up to evaluate the OSMS system. Evaluation of OSMS relative to CBCT patient positioning The phantom was positioned at the isocenter (simple visual positioning, allowing potentially large displacements), and subsequently prepositioned with the OSMS system, by matching the CT body contour imported from Eclipse and the
Systems calibrations Optical surface monitoring system A summary of the reports of the daily QA checks of the OSMS on the period from March to July 2014 is presented in Fig. 2 for the errors of the single cameras, and the relative error of the entire system.
4
P. Mancosu et al. / Medical Dosimetry ] (2016) ]]]–]]] Table 1 Differences between OSMS and CBCT positioning in the 6 directions, for small and large initial mispositioning. Small mispositioning (all o 1 cm and o 11) Direction Vertical (mm) Longitudinal (mm) Lateral (mm) V-rotation (deg.) Pitch (deg.) Roll (deg.)
Average ⫾ 1 SD 0.22 ⫾ 0.24 0.17 ⫾ 0.13 0.06 ⫾ 0.28 0.06 ⫾ 0.11 0.06 ⫾ 0.08 0.00 ⫾ 0.07
Large mispositioning (at least one 4 1 cm and 4 11)
Range Average ⫾ 1 SD 0.3, 0.6 0.39 ⫾ 0.44 0.5, 0.1 0.00 ⫾ 0.21 0.5, 0.3 0.18 ⫾ 0.45 0.2, 0.1 0.11 ⫾ 0.11 0.2, 0.1 0.10 ⫾ 0.09 0.1, 0.1 0.02 ⫾ 0.11
Range 0.3, 1.0 0.5, 0.2 1.1, 0.4 0.3, 0.1 0.2, 0.1 0.2, 0.1
SD ¼ standard deviation.
the intrinsic accuracy of the two systems according to the calibration and check procedures. The second set of measurements, with coarser initial phantom positioning, presents a higher uncertainty for the linear direction. The average of the vector lengths from the linear direction distances were 0.45 ⫾ 0.13 and 0.68 ⫾ 0.35 mm for the small and large initial displacements, with an overall average of 0.57 ⫾ 0.28 mm.
Fig. 2. Errors from the daily QA procedure, from March to July, 2014. (Color version of figure is available online.)
The error of the single cameras is the distance from the current calibration plate position to the monthly calibration plate position. This error was 1.0 ⫾ 0.9 mm for each camera. The relative error is here the difference in 3D motion computed by a pair of cameras and is a measure of the inconsistency in movement detection between cameras. This error was o 0.5 mm in more than 98% of the tests, with an average value of 0.12 ⫾ 0.12 mm. Cone beam computed tomography The kV imaging system accuracy according to the IsoCal procedure presented a maximum shift of 0.11 mm and an imager rotation of 0.051. Such a shift has to be combined with the treatment isocenter radius evaluated with the Winston-Lutz test, that was measured as of 0.31 mm for gantry and collimator rotations, and 0.54 mm including the couch rotation. These bring to an accuracy in the positioning of 0.33 and 0.56 mm without and with couch inclusion, respectively. OSMS evaluation The positioning distance is defined as the difference between the OSMS and the expected shifts derived from the CBCT acquisition or from a predefined applied shift. Evaluation of OSMS relative to CBCT patient positioning The distances between OSMS and CBCT positioning in the 6 directions are summarized in Table 1, in the 2 groups of small and large initial shifts. Those values reflect the good agreement between the 2 systems, with a maximum linear distance of 0.6 mm and a maximum rotational distance of 0.21, and a mean that stays inside
Evaluation of OSMS response to predefined inaccuracies The distances from the predefined shifts are summarized in Fig. 3, where all the 6 directions are presented for the shifts applied in each single direction. Table 2 reports the average distances. The rotational directions present the largest distances when linear shifts are applied, mainly the roll, followed by the pitch; conversely, the rotational distances are smaller in situations where an intentional rotational shift was applied. The linear shifts are similar for displacements in all the directions, keeping the isocenter position within 0.5 mm, with the vertical direction showing in some conditions the worst case. The results of all the predefined random shifts in 6 directions at a time (150 acquisitions) were within 0.1 ⫾ 0.2 mm for the linear, and within 0.1 ⫾ 0.11 for the rotational directions. The average vector length determined by the linear direction displacements was 0.4 ⫾ 0.2 mm. In Fig. 4(A), the distances are stratified for the 0.5, 1, and 2 cm and degree maximum shifts, and summarized for all the directions, as average distances and standard deviations in each of the three sets. Such results confirm what analyzed in the single direction shifts: (1) lower accuracy in determining the roll misplacement, which could worsen when increasing the displacement, (2) linear displacements within 0.5 mm. Moreover, there is no clear trend in the distance when larger displacements are applied. The average of the vector length determined from the linear direction distances were 0.30 ⫾ 0.14, 0.33 ⫾ 0.12, 0.50 ⫾ 0.28 mm for the 3 sets of 0.5, 1, and 2 cm and degree maximum shifts, respectively. Evaluation of OSMS response at different couch angles The distances from the applied intentional shifts on single directions up to 3 cm or 21 (on 66 acquisitions for each couch angle) were in average 0.0 ⫾ 0.3 mm and 0.00 ⫾ 0.051 for 901 couch angle, and 0.0 ⫾ 0.2 mm and 0.01 ⫾ 0.071 for 451 couch angle, proving the ability of the OSMS to properly manage the phantom and the coordinate system rotation. The shifts applied in all directions for the 3 sets o 0.5, o 1, o 2 cm and degrees, respectively (10 acquisitions each set) are reported in Figs. 4(B) and (D), presenting an average of the vector length determined from the linear direction distances of 0.20 ⫾ 0.07, 0.31 ⫾ 0.16, 0.39 ⫾ 0.22 mm for the 3 sets of shifts and 901 couch, and 0.11 ⫾ 0.09, 0.24 ⫾ 0.08, 0.21 ⫾ 0.08 mm with 451 couch.
P. Mancosu et al. / Medical Dosimetry ] (2016) ]]]–]]]
5
Fig. 3. Distances in all the 6 directions, from intentional shifts in one direction only: linear shifts in the upper row, rotational in the lower row. (Color version of figure is available online.)
Table 2 Distances from predefined 1 direction shifts. Data represent the average distances ⫾ 1SD in all the 6 directions, for each predefined shift direction VERT shift Residual Residual Residual Residual Residual Residual
error error error error error error
on on on on on on
VERT (cm) LONG (cm) LAT (cm) V-ROT (deg.) pitch (deg.) roll (deg.)
0.00 0.00 0.01 0.03 0.03 0.04
⫾ ⫾ ⫾ ⫾ ⫾ ⫾
0.01 0.01 0.01 0.02 0.02 0.07
LONG shift
LAT shift
0.01 0.00 0.01 0.02 0.02 0.03
0.01 0.00 0.01 0.03 0.00 0.05
⫾ ⫾ ⫾ ⫾ ⫾ ⫾
0.01 0.01 0.01 0.06 0.04 0.07
⫾ ⫾ ⫾ ⫾ ⫾ ⫾
ROT shift 0.01 0.01 0.01 0.05 0.05 0.07
0.01 0.00 0.01 0.03 0.01 0.02
⫾ ⫾ ⫾ ⫾ ⫾ ⫾
0.02 0.01 0.01 0.02 0.02 0.04
PITCH shift
ROLL shift
0.01 0.00 0.01 0.02 0.00 0.03
0.00 0.00 0.01 0.03 0.01 0.01
⫾ ⫾ ⫾ ⫾ ⫾ ⫾
0.01 0.01 0.02 0.03 0.03 0.03
⫾ ⫾ ⫾ ⫾ ⫾ ⫾
0.01 0.01 0.01 0.02 0.02 0.04
VERT ¼ vertical; LONG ¼ longitudinal; LAT ¼ lateral, V-ROT ¼ rotation along the vertical axis.
Fig. 4. Distances of the predefined randomly chosen shifts in all the 6 directions. Results are stratified for shifts o 0.5, o 1, o 2 cm or degree, respectively. Error bars refer to 1 SD over all measurements. (A) Gantry and couch set at 01; (B) gantry set at 01, couch at 901; (C) gantry set at 451, couch at 01; and (D) gantry set at 01, couch at 451.
6
P. Mancosu et al. / Medical Dosimetry ] (2016) ]]]–]]]
Evaluation of OSMS response with 2 of 3 images The distances from applied intentional shifts on single directions up to 3 cm or 21 (on 66 acquisitions) with gantry set at 451 and couch 01 (image reconstructed with 2 of 3 cameras) were in average 0.1 ⫾ 0.8 mm and 0.01 ⫾ 0.101, similar to the acquisitions with 3 cameras, but with a higher standard deviation. The shifts applied in all directions for the 3 sets o 0.5, o 1, o 2 cm and degrees, respectively (10 acquisitions each set) are reported in Fig. 4(C), presenting an average of the vector length determined from the linear direction distances of 0.46 ⫾ 0.19, 0.34 ⫾ 0.19, 0.36 ⫾ 0.25 mm for the three sets of shifts.
Discussion This work evaluated, on a phantom base, the accuracy of using the surface imaging system OSMS as implemented on the recently installed EDGE linac system. The results of the OSMS relative to the CBCT positioning presented an average difference of 0.6 ⫾ 0.3 mm and maximum rotational inaccuracy of 0.31, including small and large initial positioning. Such results should be compared to what obtained by Peng et al.,6 where the mean vector difference was of 0.2 ⫾ 0.3 mm and the maximum rotational difference was 1.31. The results from that group were obtained with the AlignRT system compared to an Elekta kV CBCT. The data shown in the present study refer to a new frame composed by the OSMS system installed on an EDGE linear accelerator. This compares the optical system with the CBCT imaging system of a linear accelerator specifically designed as a comprehensive radiosurgery system, focused on very high precision. The accuracy of the OSMS skin surface registration relative to the CBCT internal anatomy match showed to be of the same order. However, a robust flow and procedure should be set to use the OSMS system routinely. For fractionated stereotactic treatments, a patient prepositioning via OSMS by matching the skin surface with the body contour from the planning CT could be suggested, followed by a verification and evaluation with a CBCT of the internal anatomy. After having applied eventual shifts according to the CBCT matching, a new reference image for further OSMS monitoring and positioning should be captured, being the correlation between the actual patient skin surface and the anatomical match guaranteed by the CBCT imaging. For patient surface detection, the fixation mask should be open enough to allow skin detection, while decreasing the mask detection. In this way the positioning and monitoring is relative to the patient movement instead of the mask (or external markers) displacement, enforcing the direct patient positioning check while minimizing those of indirect external surrogates (mask and markers). The usage of open masks is on one side a technical need (the actual check is of the patient surface), and on the other side it improves the patient comfort, allowing better patient compliance. Of course, an optimal balance of the open and closed surface of the mask has to be evaluated and found, as too open masks could increase patient ability to move, reducing hence position accuracy. The acquisition and surface reconstruction time is a very fast procedure. As soon as the OSMS is fully integrated in the EDGE unit, with the possibility to directly apply the shifts to the couch movements, as it is today with the CBCT imaging, the OSMS usage require very short time, requiring no extra time for the treatment session. In summary, the efficient surface detection and matching is achievable with improved patient comfort, and without detriment to the accuracy of treatment delivery, treatment times, or clinical outcomes.2-4,8,9 More investigations are needed to determine the correlation and accuracy between internal patient anatomy and superficial positioning in real clinical cases, where the possible
patient movement and anatomy changes would play a role independent from the pure phantom-based accuracy of the system, as the current study. An interesting point related to the surface matching is the choice of the ROI where to register the acquired and the reference 3D surfaces. In the present study, the ROI included entirely the nose, the cheeks, the eyes and the forehead, and was chosen to have a quite large surface with the maximum variation of heights and distances. A matching of smooth surfaces could easily be affected by positional errors that the mathematical registration cannot solve. The choice of a good ROI should in principle improve the matching results, being a good ROI a region that includes suitable surface features not invariant for rotations or translations. The facial surface is in this respect a good candidate, whereas, e.g., large flat surfaces, like the abdomen, should be less suitable for surface matching. Anyway, care should also be taken in regions where possible movement can happen, as it could be the case of the eyelids involuntary movement. To consider is the fact that the matching is between the OSMS acquisition and the patient surface as reconstructed from the body contouring delineated in the treatment planning system. Clearly, the accuracy of such a contouring in the ROI region is crucial for a proper match, and for that reason care has to be taken in drawing or finding the most appropriate tool for delineation, depending also on the patient. The body contour can also be considered the baseline of the patient, from which evidentiate possible changes in his/her anatomy, especially for fractionated treatments. Peng et al.6 found that the smallest registration variance for isocenter displacement with optical surface reconstruction was shown for the upper frontal only part of the face, or on one side only of the face, that are indeed the facial surfaces presenting the least height differences. Moreover, Wiersma et al.14 discussed that the time needed for surface reconstruction scales with the ROI size, hence the ROI has to be cautiously chosen, following the surface features more than simply the ROI size. A further step of the evaluation of OSMS in the EDGE frame is the use of surface image reconstruction for the patient monitoring during the treatment, including the possibility to hold the beam when the matching is out of a predefined threshold. Such experiments are under investigation in our group, but are out of the scope of the current study, that is phantom based. To consider is also the high dose per fraction of the stereotactic treatments. EDGE is equipped with FFF beams, i.e., with high dose rate (1400 and 2400 MU/min for 6 and 10 MV FFF, respectively). The treatment time, in terms of beam-on time, is hence greatly reduced up to a factor 4 for 10 MV FFF. Shorter treatment time could also imply lesser patient movement during the delivery. The major limitation of the current work is its focus only on phantom measurements. In this preliminary phase of studying the accuracy of the OSMS installed on the EDGE linac, the patient was not included in the flow, to evaluate the system per se, with no additional factors. However, the final use of the system is to detect patient positioning inaccuracies, where many additional factors concur to the results obtainable when a patient, and not a phantom, is positioned on the treatment couch. Those factors can refer to the surface detection and reconstruction system in following the continuous deformations that the same surface is undergoing due to the unavoidable small changes coming from unwitting contractures of the facial muscles, or eye movements. This could potentially adversely affect the real-time matching between the reference and the reconstructed surface, as the matching could be inadvertently be anchored to some anatomical points affecting the registration quality. Another important factor related to patient positioning is the surface changes that could occur from session to session, or from CT planning acquisition to the treatment session, changes due to anatomy, coming for
P. Mancosu et al. / Medical Dosimetry ] (2016) ]]]–]]]
example to a different contracture or relaxation of the facial muscles, even if stable throughout the entire treatment session. Studies on the usage of the OSMS with patients are under processing in our group to evaluate the accuracy of the system when used in its primary scope, distinguishing at that stage which is the accuracy of the system per se, and quantifying the uncertainty coming from the critical points related to patient positioning.
Conclusions The OSMS system under investigation showed to be accurate to determine phantom positioning respect to the CBCT imaging system: translations and rotations of a rigid and fixed phantom can be detected accurately with the optical surface detection system.
Acknowledgments L. Cozzi acts as Scientific Advisor to Varian Medical Systems and is Clinical Research Scientist at Humanitas Cancer Center. References 1. Bova, F.J.; Meeks, S.L.; Friedman, W.A.; et al. Optic-guided stereotactic radiotherapy. Med. Dosim. 23:221–8; 1998. 2. Breneman, J.C.; Steinmetz, R.; Smith, A.; et al. Frameless image-guided intracranial stereotactic radiosurgery: clinical outcomes for brain metastases. Int. J. Radiat. Oncol. Biol. Phys. 74:702–6; 2009.
7
3. Ramakrishna, N.; Rosca, F.; Friesen, S.; et al. A clinical comparison of patient setup and intra-fraction motion using frame-based radiosurgery versus a frameless image-guided radiosurgery system for intracranial lesions. Radiother. Oncol. 95:109–15; 2010. 4. Ohtakara, K.; Hayashi, S.; Tanaka, H.; et al. Clinical comparison of positional accuracy and stability between dedicated versus conventional masks for immobilization in cranial stereotactic radiotherapy using 6-degree-offreedom image guidance system-integrated platform. Radiother. Oncol. 102:198–205; 2012. 5. Cerviño, L.I.; Pawlicki, T.; Lawson, J.D.; et al. Frame-less and mask-less cranial stereotactic radiosurgery: a feasibility study. Phys. Med. Biol. 55:1863–73; 2010. 6. Peng, J.L.; Kahler, D.; Li, J.G.; et al. Characterization of a real-time suface imageguided stereotactic positioning system. Med. Phys. 37:5421–33; 2010. 7. Bert, C.; Metheany, K.G.; Doppke, K.; et al. A phantom evaluation of a stereovision surface imaging system for radiotherapy patient setup. Med. Phys. 32:2753–62; 2005. 8. Nath, S.K.; Lawson, J.D.; Wang, J.Z.; et al. Optically-guided frameless linac-based radiosurgery for brain metastases: clinical experience. J. Neurooncol. 97:67–72; 2010. 9. Cerviño, L.I.; Detorie, N.; Taylor, M.; et al. Initial clinical experience with a frameless and maskless stereotactic radiosurgery treatment. Pract. Radiat. Oncol 2:54–62; 2012. 10. Bert, C.; Metheany, K.G.; Doppke, K.P.; et al. Clinical experience with a 3D surface patient setup system for alignment of partial-breast irradiation patients. Int. J. Radiat. Oncol. Biol. Phys. 64:1265–74; 2006. 11. Schöffel, P.J.; Harms, W.; Sroka-Perez, G.; et al. Accuracy of a commercial optical 3D surface imaging system for realignment of patients for radiotherapy of the thorax. Phys. Med. Biol. 52:3949–63; 2007. 12. Cerviño, L.I.; Gupta, S.; Rose, M.A.; et al. Using surface imaging and visual coaching to improve the reproducibility and stability of deep-inspiration breath hold for left-breast-cancer radiotherapy. Phys. Med. Biol. 54:6853–65; 2009. 13. Rogus, R.D.; Stern, R.L.; Kubo, H.D. Accuracy of a photogrammetry-based patient positioning and monitoring system for radiation therapy. Med. Phys. 26:721–8; 1999. 14. Wiersma, R.D.; Tomarken, S.L.; Grelewicz, Z.; et al. Spatial and temporal performance of 3D optical surface imaging for real-time head position tracking. Med. Phys. 40(111712-1):8; 2013.