British Journal of Radiology (2006) 79, S66-S78
© 2006 British Institute of Radiology
doi: 10.1259/bjr/68255935
Developments in and experience of kilovoltage X-ray cone beam image-guided radiotherapy
C J Moore, PhD
1
A Amer, PhD
1
T Marchant, PhD
1
J R Sykes, MSc
1
J Davies, DCR(T)
2
J Stratford, DCR(T)
2
C McCarthy, PgDip
2
C MacBain, MD
3
A Henry, MD
3
P Price, MD
3 and
P C Williams, PhD
1
1 North Western Medical Physics Department, 2 Radiotherapy Research Facility, 3 Academic Department of Radiation Oncology, Christie Hospital, Manchester M20 4BX, UK
 |
Abstract
|
|---|
This paper offers a realistic review of kilovoltage X-ray cone beam tomography integrated with the treatment machine for image-guided radiotherapy in the light of experience taking a commercial system from prototype development into clinical use. It shows that key practicalities cannot be ignored, in particular the regular characterization of mechanical flex during gantry rotation, the mapping of defects in flat panel image transducers and their response to X-ray exposure. The number of X-ray projections and the doses required for clinically useful cone beam reconstruction at different therapy sites are considered in the context of imaging that is fit for purpose. Three roles for cone beam tomography in radiotherapy are identified: patient setup in three dimensions (3D), where even low dose cone beam tissue detail is superior to megavoltage imaging; disease targeting where, despite wide field scatter and slow scanning, it is possible to generate images that are suitable for tumour delineation even at challenging sites; adaptive treatment planning, where calibrated cone beam images have been shown to provide sufficient target detail to support "plan of the day" selection and have the potential for planning with bulk corrections. With frequent use in mind, the need to limit patient dose during setup, yet maximize much needed image quality in the target zone, is considered. Finally, it is noted that the development of cone beam tomography for radiotherapy is far from complete, with X-ray source, image transducer, reconstruction algorithms and techniques for image profile collection still being researched.
|
Historical background and introduction
|
|---|
In 1984, industrial researchers Feldkamp, Davis and Kress (FDK) [1] reported a circular scanning algorithm for three-dimensional (3D) X-ray cone beam tomographic (CBT) reconstruction. However, even two-dimensional (2D) reconstructive imaging was slow to appear in the radiotherapy treatment room, although in 1992 Lewis et al [2] described 2D, megavoltage X-ray, computerized tomographic (CT) reconstruction. This delay can be understood in the context of the operating tolerances for many radiotherapy techniques, which remained relatively generous until the mid-1990s when the impact of the recommendations set out in ICRU report 50 [3] and the development of computerized multileaf collimation (MLC) for conformal radiotherapy began to be felt. Until then, it had been sufficient to base patient setup on skin marks and verification on the bony detail visible in wide field, megavoltage verification films. Megavoltage electronic portal imaging devices (EPIDs) appeared in the late 1980s [4], offering the advantage of integration with the gantry of the treatment machine and instant availability of a digital image without the need for film processing. Alignment and use with the treatment source provided beam's eye views (BEVs) at arbitrary gantry angles. Early EPIDs offered adequate spatial resolutions of a few millimetres in the imaging plane and repeated use for time lapse imaging of tumour motion, notably in the lung.
The obvious alternative to the EPID, a CT scanner introduced into the confined space of an existing treatment room, had none of these particular advantages for setup and monitoring. CT had been carefully optimized for temporally and spatially discontinuous, sequential transaxial scanning, with images accurately calibrated in Hounsfield units. These were ideally suited to treatment planning, but not to BEV production. Antonuk's work on X-ray image transducer panels [5], employing selenium and amorphous silicon (aSi) technologies, appeared to promise a significant improvement in the application of the EPID, despite the underlying physical limitations constraining megavoltage imaging to physical density mapping (see Chen et al in this issue). However, as well as improving portal imaging these efficient, high resolution flat panel transducers provided the basis for the practical implementation of wide-field CBT imaging for radiotherapy using kilovoltage X-rays.
Conformal radiotherapy, implemented with ICRU guidelines in mind [3], requires image grey level detail to be reduced to a set of explicit structural boundaries measured within the treatment room frame of reference. Indeed, spatial measurement from CT image sequences, effectively an image volume, has become a distinct activity in the modern radiotherapy treatment planning room. Clearly "image assisted" planning has the potential to develop still further as functional imaging, in the form of MRI and PET, becomes more widely available. In the last decade, this geometric conformation of irradiation fields to projected tumour shape has rapidly evolved into more precise 3D dose conformation, the most demanding application of this being intensity-modulated radiotherapy (IMRT). In this technique, MLC technology is used to sequentially superimpose differently sized and offset beam segments to form the equivalent of one intensity-modulated beam. In the treatment room, this technically complex procedure has propelled radiotherapy from 2D setup imaging into 3D image-guided radiotherapy (IGRT). Setup imaging has been supported by the assumption that the correlation between readily identifiable and stable features, usually rigid bony anatomy, and the tumour target is maintained within certain margins for each fraction of the treatment. Orthogonal portal images showing bony detail, if not the target itself, are then sufficient for setup against digitally reconstructed radiographs (DRRs). A further assumption is that patient rotations can be identified unambiguously from a pair of superposition images. IGRT aims to actively inform setup and disease targeting by imaging bone and soft tissues, showing tumour and organs at risk in 3D, as close to moment of therapeutic irradiation as practicable. The adequacy of targeting, and even the pre-treatment plan itself, can then be assessed quantitatively and corrective action taken as appropriate. This definition of IGRT indicates a measurement process in the treatment room on a scale that rivals image assisted pre-treatment planning. Consequently, IGRT has stimulated the rapid and continuing development of complementary, commercial CBT systems, the subject of this paper. This is a case of X-ray imaging development being driven by a need that is not diagnostic and it signals the first widespread use and commercial availability of low energy X-ray imaging in the treatment room.
|
Practicalities and the evolution of clinical cone beam systems for radiotherapy
|
|---|
When CT scanners became available in the 1970s, it became possible to gather 2D axial image sequences with enough soft-tissue contrast for tumour delineation and to use image processing to extract body contours and organ boundaries for dosimetric modelling in radiotherapy planning (RTP). In addition, pixel by pixel dose computation became an alternative to planning based on bulk inhomogeneities. These attributes have influenced the specifications underlying the recent commercial development of CBT 3D-imaging for radiotherapy by the major manufacturers. Despite ongoing research seeking to generate reliable sources of low energy X-rays from the linear accelerator (linac) itself, all the major treatment machine manufacturers have opted to offer kilovoltage X-ray tubes mounted on the treatment gantry for CBT. They are using flat panel transducers to acquire a rotation fluoroscopy image sequence for input to 3D filtered back-projection algorithms.
At least for now, the algorithms are mostly extensions of the approach originally described by FDK in 1984, who had in mind the needs of the industrial non-destructive testing communities. This is a world apart from radiotherapy where professionals might have unduly high expectations of CBT because of their acquaintance with diagnostic CT. FDK CBT is based on the circular rotation of a cone about a fixed isocentre. This has a central plane with projection profile that appears to be from a 2D object. Hence, in a pre-calculated manner, it is filtered prior to reverse projection along the paths of the original X-rays, so taking out back-projection blurring to produce a clear reconstruction [14]. The process is iterated for non-central planes, even though the projection data are insufficient for exact reconstruction. In order to complete the projection data, one would need to move the scanning cone of X-rays away from the purely circular trajectory in some manner, or to move the patient. The limitations of the FDK approach result in blurring in the axial direction wherever there are inhomogeneities [6]. High density bone can produce streaking in surrounding soft tissues, the magnitude of which, for the Elekta Synergy system (Elekta, Crawley, UK), can be judged by inspecting the figures appearing in this paper.
In so far as FDK artefacts alter grey-level gradients, this adds some uncertainty to boundary identification and contouring. However, this has not been found to affect cross modality matching for 3D setup purposes, even in the head and neck regions. By comparison, the partial volume effects seen in conventional CT can be large with slice widths of a few millimetres, although this is often not fully appreciated. Blurred, low contrast reconstruction due to anatomical motion during the slow process of profile acquisition is a significant problem facing CBT in radiotherapy, especially when that motion deviates from periodic behaviour. Add to this the effects of cone beam scatter and it can be seen that the practical limitations to CBT are at least as important as those arising from inexact reconstruction.
FDK CBT is a practical proposition in external beam radiotherapy where rotating gantries suitable for circular scanning are the norm. The integration of CBT and medical linac technology has clearly been influenced by the work of Mosleh-Shirazi et al [7] and Jaffray et al [8]. Elekta's Synergy clearly reflects the latter and has the cone beam X-ray volume imaging (XVI) system mounted on a linac gantry at 90° to the treatment beam with extending, twin pillar supports for the X-ray tube and an extending fork mount for the folding panel imager. Varian's On Board Imager (OBI; Varian, Palo Alto, CA) also has its cone beam system perpendicular to the treatment beam, but uses folding robotic arms for both the X-ray tube and imager. In contrast, Siemens' Artiste (Siemens, Erlangen, Germany) comes closest to providing a BEV by using a 180° design based on a kilovoltage source firing directly towards the treatment head in front of which the panel imager is deployed. Given the similarities of the integrated systems, the practicalities of CBT in the treatment room inevitably revolve around common factors; mechanical aspects, image processing and doses.
The mechanics of CBT
The lessons learned implementing CT remain invaluable, even though at first sight the refined, purpose built machinery in the form of the CT ring mounting bares no comparison with a bulky linac gantry to which a radiographic unit has now been added for CBT. The IEC requirements for safety limit full linac gantry rotation are asked to be a minimum of 60 s [9], which makes CBT in the treatment room at least as slow as the first clinical CT scanners. Flex in the combined structures of linac and radiographic imaging system means that the central rays from the kilovoltgage source do not cross at a single point in space during the rotation fluoroscopy that feeds CBT reconstruction. X-ray projections of an optimally centred ball-bearing test object trace out a track on the aSi transducer plane that can be resolved into orthogonal curves termed "flex maps" that are quite specific to the machine being investigated [10]. The flex map pairs allow kilovoltage X-ray image projections to be re-aligned prior to filtered back-projection in CBT.
For the Synergy system installed at the Christie Hospital, Manchester, extended monitoring has shown that the pattern and range of mechanical flexing in a large IGRT system is reproducible and stable over months of clinical use. Flex is periodic with peak-to-trough amplitudes that are usually less than two pixels (
1.5 mm) in the plane of the flat panel imaging transducer. The kV and MV isocentres and their systematic difference can be accurately set to < 0.5 mm, which is now an action level for re-evaluating flex maps. Maintenance operations also trigger re-evaluation. The reason why is illustrated by Figure 1a
, which shows a flex map pair produced in July 2004 and then again in September 2004 following work on the positioning micro-switches of the imaging panel. Although the flex maps measured along the panel's V-axis (gun target direction) are the same shape, they are clearly shifted by nearly a pixel. After re-evaluation, flex maps are much more reproducible and repeatable than shown in this example. Figure 1b shows the isocentre stability, which is assessed weekly. Over a period of approximately 1 year the isocentre reached the 0.5 mm action threshold only twice. These and two mechanical maintenance events prompted flex map re-evaluations.
View larger version (34K):
[in this window]
[in a new window]
|
Figure 1. (a) Flex map pairs taken 2 months apart. U and V describe the projection of a centred ball bearing resolved in the directions of the major orthogonal axes of the imaging panel. Deviations are shown in pixels, at a scale of 0.5 mm pixel–1. Flex maps peaking with the two highest offsets (0.0 and 0.7 pixels) were measured on 19 July 2004 and the remaining two were measured on 15 September 2004 (lower right hand key). (b) Isocentre calibration results for a 1 year period. Variation to the threshold level occurred twice (right hand vertical lines) triggering flex map re-evaluation. Mechanical maintenance triggered precautionary flex map re-evaluation on a further two occasions (left hand vertical lines).
|
|
For 90° mounts, the retraction of the kilovoltage imaging system is necessary for easy access to the patient on the treatment couch. Clinical use requires repeated deployment of the X-ray tube to a single, fixed position. On our system, the unfolding and offsetting of the transducer panel for CBT imaging is to one of three fixed positions, determined by the choice of a nominal 26 cm, 41 cm or 51 cm reconstruction field. Isocentre tests allow the reproducibility of deployment to be monitored, since any drift in panel translation will be revealed as a shift in the isocentre position on the imaging transducer array. Ideally, the normal to the plane of the imaging panel also remains normal to the gantry rotation axis. In addition the rows of transducer elements in the aSi array should remain parallel to the scanning axis throughout gantry rotation. Any deviations from this ideal need to be measured and taken into account during CBT reconstruction. Figure 2 shows the effect of 0.6° panel skew in the plane of the imaging panel. This was discovered by simply imaging a suspended engineering-standard steel ruler and noting that the projected edges were stepped rather than smooth, due to diagonal crossing of the rows of discrete transducer elements [11]. Comprehensive geometric calibration of cone beam systems and the development of test phantoms have been described by Smekal et al [12] and Cho et al [13]. Skew produces an obvious loss in apparent bone density, the blurring of otherwise fine detail and the possibility of double edging in bony anatomy. The principle of using image processing algorithms to automatically identify the degree of skew and optimally correct for its effects has been described by Amer et al [11]. However, skew has only been encountered once in the 3 years that Synergy has been installed at the Christie Hospital.
View larger version (103K):
[in this window]
[in a new window]
|
Figure 2. (a) The result of 0.6° in-plane, image transducer skew on cone beam tomography (CBT). (b) Results after skew correction. Note the improved contrast and fine detail.
|
|
Image data processing in CBT
The aSi panel detectors in commercial CBT are indirect detection systems. Light produced in a thin layer of luminescent material is captured efficiently by contact interfacing to a silicon diode array. When deployed the arrays must be sufficiently large to image typical treatment volumes. Hence, it is not uncommon to find compound aSi panels that are made from two transducer arrays. In conventional CT the behaviour of a few thousand, high efficiency, solid state crystal detectors is charted. In CBT, flat panels have 2D arrays typically providing 1024x1024 transducer elements that must be calibrated for detector flaws, offset and gain.
The manufacture of the imaging panel is an imperfect process and so redundancy is inbuilt. Nevertheless, unresponsive and unreliable, or "bad", transducer elements remain that must be identified and entered into a bad-pixel map in order to avoid their use in CBT back projection, where they would produce ring artefacts. The aSi panels used at the Christie are Perkin Elmer RID1640-1,048,576 gadolinium oxysulphide transducers (Perkin Elmer, Santa Clara, USA). They provide a compound, rectangular transducer area of 41 cmx41 cm operating at a sampling rate of 3 Hz. The gadolinium oxysulphide panel has a measured bad pixel population of several hundred unresponsive pixels and several thousand unreliable pixels, locally termed "transients". Figure 3 illustrates their distribution and instability. More efficient caesium iodide panels, working at 6 Hz, are available.
View larger version (42K):
[in this window]
[in a new window]
|
Figure 3. Bad pixels, based on their variability, on an amorphous silicon imaging panel.(a) 7000 entries with standard deviation>3 on 1 day. (b) 2000 entries with standard deviation>3 over 5 days.
|
|
Not surprisingly reconstruction by filtered back-projection is sensitive to small variations in the signals emerging from the flat panel transducer array, whether they are due to real variations in exposure in an image profile or defects. A 1% local variation is enough to produce obvious effects in CBT reconstruction. Hence, in addition to bad pixels, it is necessary to correct for the inherent structure generated by the panel imager whilst acquiring X-ray image profiles. In this context, the panel dark-field, an offset image seen in the absence of X-rays, introduces a great deal of unwanted irregular structure. Figure 4a shows an example. In practice, it is now known that sampling the offset image close to the time of clinical profile acquisition greatly improves CBT image quality. Flood field (gain) calibration imaging is the other side of profile correction. It allows the anisotropic, non-linear exposure responses of pixels across the flat panel to be taken into account. Figure 4b is an example of the unwanted, regular panel structure seen in high exposure gain images. Rather reassuringly, the authors have found offset and gain calibration of X-ray image profiles to be eminently practical and effective. Bow-tie filters can be used to physically equalize exposure across the volume of view, though these are not used in the examples presented in this paper. A useful side-effect is the reduction of low energy X-rays and reduction in skin dose.
View larger version (69K):
[in this window]
[in a new window]
|
Figure 4. (a) Mean dark field (offset image) computed from 50 sample frames for a gadolinium oxysulphide aSi imaging panel. The axes indicate that the panel can provide up to 1000x1000 pixels per image frame. (b) Mean flood field (gain image) computed from 50 sample frames for a gadolinium oxysulphide aSi imaging panel and high exposure.
|
|
CBT is essentially a process of carefully calibrated rotation fluoroscopy followed by reconstruction to produce an X-ray image volume, most commonly using the FDK filtered back projection algorithm. The image volume must be viewed with at least two very basic, but often ignored, rules from diagnostic radiology in mind. The first is to interrogate the CBT volume image in the most appropriate plane, which may not be the transaxial plane that most radiotherapy observers are conditioned to select by virtue of their experience with axial RTP-CT images in treatment planning systems. Coronal and sagittal views of RTP-CT image data can be highly discontinuous because axial slice thicknesses of several millimetres have been the de-facto standard for decades. CBT image volumes are not inherently slice based and the spatial resolution is the same along all three Cartesian axes. Hence, CBT coronal and sagittal sections can provide exquisitely resolved detail that is ideal for structure delineation almost free from significant partial volume effects. The top two rows of Figure 5 amply illustrate the point. The second rule is to compare like with like when matching features visible in RTP-CT and CBT in IGRT. A natural tendency is to compare axial CBT images, in their raw state, perhaps equivalent to 1 mm slice thickness, with RTP-CT axial slices that might be several times thicker. Averaging CBT data to produce an equivalent slice width to that found in the RTP-CT scan will increase soft tissue contrast–noise ratios and produce equivalent partial volume effects. Figure 6 illustrates the process for pelvic scans.
View larger version (75K):
[in this window]
[in a new window]
|
Figure 5. (a) Radiotherapy planning (RTP)-CT scan, 5 mm slice thickness. (b) Cone beam tomography (CBT) using 383 image profiles, 1 mm slice thickness. The reconstruction is consistent with the theoretical minimum number required for exact reconstruction. (c) CBT using 148 image profiles, 1 mm slice thickness. (d) CBT using 77 image profiles, 1 mm slice thickness. Note the profile aliasing crossing the CBT axial.
|
|
View larger version (102K):
[in this window]
[in a new window]
|
Figure 6. (a) Radiotherapy planning (RTP)-CT images, 5 mm thick should be compared with (b) the equivalent cone beam tomography (CBT) images, 5 mm equivalent slice thickness by processing (c) the raw CBT data, 1 mm thick gathered at treatment time. In this example, note the changes in rectal and bladder fill.
|
|
CBT reconstruction is underpinned by rotation fluoroscopy, which when replayed offers useful insight into patient dynamics, much like the conventional simulator. Although the fluoroscopy trajectory is prescribed for CBT, there is a particular advantage over conventional simulation. Where the gross tumour volume (GTV) or organ at risk (OAR) structure, surgical clips or other suitable landmarks are clear, then the fluoroscopy sequence can be interrogated to assess the amplitude and direction of movements, since the average spatial location (and hence depth), from any particular gantry angle, is known from the CBT reconstruction itself.
Since the rotation fluoroscopic sequences and CBT reconstructions are both clinically useful, their storage, communication and archiving is an issue that must be adequately addressed. In Manchester, this has been done by installing additional dual storage PCs on the clinical network. Both have terabyte disk capacity and gigabyte memory for multiple image volume display and manipulation. A multi-tape archiving system with matching capacity has also been purchased.
Doses in CBT
With high patient exposures and a large number of X-ray image profiles for reconstruction, it is possible to generate high quality CBT images. However, a principle that supports the practice of diagnostic radiology in the UK is that images produced using ionizing radiation need only be fit for purpose. In particular, the IRMER legislation [15] embodies the principle of doses being "as low as reasonably practical" (ALARP) in the context of risks and benefits to the patient. For radiotherapy there are two key applications of serial CBT; patient setup, which can be achieved using low dose imaging, and GTV/OAR imaging, which at first sight demands higher doses than those for patient setup in order to provide the necessary contrast detail. RTP-CT scanning outside the treatment room sets a benchmark weighted dose (two-thirds of the surface dose plus one third central dose in a cross section, used throughout this paper unless otherwise stated) of approximately 1–2 cGy, which is in line with megavoltage portal imaging in the treatment room at 1–2 cGy per image (maximum dose). Wide field imaging at these dose levels has been acceptable in radiotherapy because its use has been restricted to the early stages of the treatment planning and delivery process, where the correction of systematic setup errors, at least in terms of bony anatomy if not the tumour itself, is likely to have maximum benefit assuming the absence of time trends. This message is reinforced by radiobiological modelling studies [16]. Nevertheless, setup based on wide field orthogonal portal images for only 5 treatment days can lead to a cumulative dose of 10–20 cGy. With the appearance of CBT in the treatment room it is practical to acquire images for bony setup at every fraction with the added potential benefit of GTV/OAR visualization (see the next section). Serial use of cone beam reconstruction techniques delivering approximately 5 cGy per image can result in patient doses that approach the 1 Gy level [17, 18]. In the context of IRMER, these are approaching cautionary levels where deterministic effects of low energy kilovoltage X-ray on the skin might become visible across large areas. They have already added fuel to the debate on induced secondary cancers [19, 20].
The distinction between CBT and conventional CT is purely technical; patient dose is due to kilovoltage X-ray interactions with tissue, and so it is appropriate to make use of the existing framework for describing dose levels that is already established in diagnostic radiology. The CT dose index (CTDI) is the 2:1 weighted measure of skin and central dose that forms the basis of the European dose reference levels (EDRL) for different anatomical sites [21]. EDRL values range from 30 mGy to 60 mGy moving from the head to the pelvis. Amer et al [22] have reported typical CBT dose levels for head and neck (1.6 mGy using 100 kVp, 25 mA, 4 ms profiles), lung (6.0 mGy using 120 kVp, 40 mA, 10 ms profiles), breast (3 mGy using 120 kVp, 40 mA, 5 ms profiles) and pelvis (25 mGy using 130 kVp, 40 mA, 30 ms profiles). These are based on the Christie Hospital's approach of using a minimum set of approximately 400 profiles for exact CBT reconstruction to a 512 cubic image matrix. Serial CBT over 20 fractions would then result in total doses of between 2 cGy and 50 cGy, depending on the anatomical site. Given the benefits of target visualization these might be justifiable. However, the technique of zonal CBT reported by Moore et al [23] has the potential to simultaneously improve image contrast in the target zone whilst halving doses outside the target zone, making serial CBT throughout treatment no more expensive in terms of dose than short term EPID use.
|
The clinical role of cone beam in radiotherapy
|
|---|
The parallels between CBT and conventional CT are by now very clear. However, there are some differences that reflect the intended application of CBT in the radiotherapy treatment room. For example, the Christie Synergy is an IGRT facility intended for patient setup and tumour targeting rather than the production of images with grey scale values calibrated for use in treatment planning. Nevertheless, there is clear potential for adaptive planning with bulk inhomogeneity corrections and the long scanning periods are reminiscent of the scanning protocols used in the early days of RTP, when averaged patient motion was viewed positively. Varian's cone beam promotional material (http://www.varian.com/orad/pdf/radonc.pdf) illustrates how adaptive planning of patients might be performed using cone beam volume images taken on the day of treatment. The clinical applications of CBT in radiotherapy deserve closer scrutiny since the expectation is that patient setup, disease targeting and possibly treatment planning can be used frequently, if not daily, over the course of a treatment. The liberal use of the wide field CBT exposures could drive concomitant patient doses to levels associated with North American medical practice. This will be avoided by advancing clinical practice in the context of the European legislation that protects the interests of the patient and embodied as IRMER in the UK [15].
Patient setup
Kilovoltage CBT is an obvious replacement for megavoltage EPIDs in patient setup, where 2D portal images are matched against digitally reconstructed radiographs (DRRs) derived from RTP-CT scans. The latter amounts to the matching of spatially continuous, low contrast physical density images to superposition images derived from sequential axial sections, which have high contrast by virtue of their photoelectric origins. Since CBT produces high contrast 3D image volumes with equally high spatial resolution along all axes, it can support translational and, given appropriate couch technology, full rotational correction of patient setup by direct comparison with RTP-CT scan volumes. RTP-CT protocols have evolved to match the resolutions of MLCs in wide field conformal therapy, which started with leaf projections of order 10 mm at isocentre. Hence, they focused on the production of matched CT sections spaced at intervals of 10 mm. More recently this has become 3–5 mm along the key superior–inferior (SI) axis parallel to the portal image plane. In turn, the crude DRRs generated from these sections for setup verification purposes were consistent with the relatively poor performance of early EPIDs in the treatment room, some of which could only offer 4 mm resolution. CBT in the treatment room can now offer better than 1 mm spatial resolution in the SI direction. Hence one would expect this to stimulate the increased use of matching, 1 mm resolution multi-slice and spiral CT in conformal treatment planning. This will improve not only field shaping but also setup verification.
CBT ushers in an era where the manual matching of selected features throughout two almost contiguous image volumes is far from easy. In addition, unambiguous 3D visualization through solid data is beyond current display technologies. However, computer assisted volume matching has already reached a degree of maturity. Using the Philips Syntegra package running on a Pinnacle treatment planning system manual (Philips Pinnacle, Milpitas, USA) and automatic matching of CBT to RTP-CT have been compared by Marchant et al [24]. This work concentrated on the pelvis, showing the equivalence of manual matching using bony detail and automated local correlation matching based on all the available CBT grey scale data. This result is not entirely unexpected, given the qualitative similarities of the human visual response function and the mathematics of correlation, which are greatly influenced by discontinuities in image data. However, it does highlight the fact that, so far as radiotherapy setup is concerned, high integrity grey scale reproduction in CBT may not be required and the need for many hundreds of X-ray image profiles to support reconstruction is questionable.
An indication of the minimum number of image profiles needed for optimum CBT reconstruction is readily established by considering the circular scanning trajectory and approximately cylindrical reconstruction region of interest (RROI) in commercial systems. For an RROI with base diameter and height N pixels, there are 
N3/4 volume elements to be characterized from a set of square X-ray image profiles, each representing N2 measurements. Simple division shows that N/4 (0.785N) image profiles are required to solve exactly for the attenuation at every point in the RROI. For N = 512, this suggests the acquisition of at least 402 X-ray image profiles or roughly one per degree for a full gantry rotation. To achieve this in a minimum 60 s scan requires the image transducer to sample at approximately 7 Hz, which is well within the performance of systems now offered by manufacturers. More profiles and more patient dose will improve image quality. However, guided by the ALARP principle, Sykes et al [25] at the Christie Hospital reported head phantom studies in which CBT reconstruction based on fewer than 100 image profiles was sufficient for matching to a high quality RTP-CT scan with sub-degree, sub-millimetre accuracy. Figure 5
illustrates the effect of reduced profile CBT reconstruction for a patient. Soft-tissue detail remains visible down to very low profile counts, despite the appearance of aliasing streaks crossing the RROI. Reduced profile CBT reconstruction for setup offers the prospect of containing or even reducing patient dose over wide fields of view. Reducing profile collection in turn suggests that faster scanning is a possibility and revisiting gantry design with a view to influencing IEC rotation limits could be beneficial.
Disease targeting
Targeting requires sufficient soft-tissue contrast detail at the disease site in order to define the position and shape of what can be termed the GTV of the day. The same is true for adjacent OAR, which may be part of a locally interacting configuration of diseased and healthy anatomy. This suggests high CBT exposure levels and large numbers of image profiles must be collected to improve the signal to noise ratios in reconstructions. Where there is a moving target, as in the lung, the possibility of motion reduction through phase selection and reconstruction using 4D CT techniques, as demonstrated by Sonke et al [26] for CBT on a Synergy system, can increase the need for profiles still further. The result is significantly increased X-ray tube loading. Such considerations have influenced the production versions of commercial IGRT systems, including Elekta's Synergy, which has an increased capacity X-ray tube compared with the prototype.
Improving soft tissue contrast in CBT is complicated by the appreciable scatter field produced by a wide field cone beam as it passes through the bulk of the patient. Anti-scatter grids at the image transducer plane have had some success. However, "coning down" potentially restricts the use of CBT in patient setup. This has prompted the use of the EPID dual exposure approach by Letourneau et al [17] for CBT, where a wide RROI CBT for setup purposes and a coned down "local tomography" RROI are superimposed to improve target visualization.
Technique development is clearly influencing the time needed for CBT investigations and increasing the dose to the patient. However, having already noted that reduced profile, reduced dose CBT can support accurate setup, Moore et al [27] have reported a simple "zonal CBT" technique to reduce scatter at source in the patient, improve image contrast–noise ratios in the target zone and simultaneously reduce patient dose compared with a full field technique. They use machined aluminium transmission diaphragms attached to the kilovoltage X-ray tube of an Elekta Synergy. These operate independently of the usual lead diaphragms, which are left fully open. During CBT, dose to the tissues lying beneath the transmission diaphragms is reduced by a factor of two or more, according to the thickness and composition of the diaphragms. In turn, the scatter from these tissues is correspondingly reduced. Between the transmission diaphragms, an unattenuated X-ray beam is directed towards the target zone, which now receives reduced scatter and hence lower dose from the extrazonal tissues. The early indications are that isocentre dose levels are readily halved by this technique. The result is improved soft-tissue contrast where it is needed, at the target, and sufficient edge detail elsewhere for wide field setup using Syntegra. Figure 7 shows a bladder patient scanned using zonal CBT with contrast enhanced target detail and useful soft tissue contrast well outside the enhanced target zone.
View larger version (53K):
[in this window]
[in a new window]
|
Figure 7. Zonal cone beam tomography(CBT) of a bladder patient reduces scatter from tissues outside the target zone, thus improving target contrast whilst simultaneously preserving edge structures for setup and reducing patient dose. (a) CBT coronal section showing improved target zone contrast (arrow). Note the preserved soft tissue detail above and below. (b) 1 mm axial image detail within the target zone. (c) 1 mm axial image detail outside the target zone.
|
|
Treatment planning
EPID targeting is based on bony setup comparison with the RTP-CT scan, where the systematic errors are likely to be trend free over several days, if not weeks. With CBT it is at once obvious that the GTV, OARs and impacting soft-tissues, such as bowel, execute movements that are not necessarily random or small scale. In addition to shape change, for example in bladder filling, the target can show positional time trends, as in rectal cases. Unexpectedly large internal displacements have been observed using CBT in cervix cases. Given CBT volumes of sufficient quality, a method of rapidly delineating targets and computer assisted movement of the treatment couch, there is the potential to perform fast adaptive planning.
Armed with knowledge from CBT studies, an alternative has been examined for bladder patients at the Christie Hospital [28]. In this approach setup errors are addressed and then three quantized margins are added in turn to the bladder CTV seen on the RTP-CT scan to generate a set of three PTVs. The PTVs are designed to encompass all the bladder movements that might be seen in a population and are planned in advance of treatment. Since the greatest changes to the bladder are usually seen in the superior direction, the superior margin ranges from 5 mm to 15 mm in 5 mm increments. The anterior margin is set at 15 mm and all other margins at 10 mm. During treatment a CBT scan is used for edge based registration to the RTP-CT scan and the best margin fit to the bladder seen in the CBT is chosen from the three available plans. However, CBT has shown that there are occasions when unexpected deformations thwart even the best laid plans. Figure 8 shows a CBT coronal section of a bladder patient displayed against the primary planning data within the Pinnacle RTP system.
View larger version (117K):
[in this window]
[in a new window]
|
Figure 8. Coronal plane cone beam tomography(CBT) of a bladder, registered with the radiotherapy planning (RTP)-CT scan (inset) showing the displacement and deformation of organs in the target zone due to unexpectedly large rectal distension. The central graphic defines the clinical target volume (CTV) and the outer graphic the planning target volume (PTV).
|
|
Point by point dosimetric modelling requires cone beam reconstructions with grey levels that are a close approximation to the CT numbers found in conventional CT planning scans, despite minute long scanning times. This is particularly demanding given the scatter from a wide field, minimally collimated, kilovoltage X-ray cone beam. From the data given previously for the Synergy system at the Christie Hospital it can be seen that the cone beam imaging angles range from 15° to 30°, which is much larger than in diagnostic multislice CT scanners. Furthermore, the transducers in the multislice scanners differ markedly from the thin, indirect detection panels commercially available for radiotherapy. Clearly the development of superior image transducers will help.
Cone beam in transition
The Christie Hospital is using kilovoltage CBCT for patients with disease of the central nervous system, head and neck, breast, lung, upper gastrointestinal tract, rectum, bladder, prostate, cervix and peripheral soft tissue sarcomas. At the time of writing approximately 1000 CBT scans have been completed on a single pre-production machine.
CBT is being deployed to support IMRT, including individuals selected from the large pool of breast patients seen at the Christie Hospital each year. The Synergy treatment head has a touch guard that is 43 cm from the isocentre, which is closer than the CBT panel imager at nearly 54 cm. Hence, the equivalent "bore" for CBT scanning is 86 cm, which is comparable with widest bore CT-simulators. This provides sufficient space to acquire whole body scans for most treatment techniques. Figure 9 shows a wide field lung scan without motion gating. However, the geometric limits to cone beam scanning are reached with breast patients who lie supine on a breast board that can be inclined to 20° with the patient's arm raised and immobilized. For these patients it is not possible to perform a full rotation scan to generate a 50 cm, whole body RROI. Instead, a half scan is performed with the patient shifted vertically from the treatment centre. This produces a field of view optimally encompassing breast, heart, lungs and lateral body surface. The shift is recorded and transferred to the hospital's Pinnacle planning system where cross modality comparison with planning scans is performed, assisted by common colour overlays of radiotherapy structures, as in Figure 10.
View larger version (57K):
[in this window]
[in a new window]
|
Figure 9. Matched radiotherapy planning(RTP)-CT and cone beam tomography (CBT) image volumes on the Pinnacle treatment planning system. (a) RTP-CT axial and coronal, 5 mm sections with inner clinical target volume (CTV) and planning target volume (PTV). (b) CBT axial and coronal, 1 mm sections with registered CTV and PTV. (c) CBT axial and coronal, 5 mm equivalent sections for comparison.
|
|
View larger version (33K):
[in this window]
[in a new window]
|
Figure 10. Comparison of breast simulator-CT and cone beam tomography (CBT) in the treatment room. (a) Transverse sim-CT section with breast and lung contours. (b) Treatment room CBT axial (1 mm equivalent slice thickness) overlaid with simulation contours. (c) CBT coronal section (1 mm equivalence slice thickness).
|
|
|
The future
|
|---|
A particular concern of the authors has been that cone beam imaging in radiotherapy looked set to become an additional process executed before or after treatment. Hence, the configuration of target and OARs during the process of beam delivery itself would not be known. To address this problem, we have collected X-ray projections during the (segmented) gantry rotations between the beam deliveries of a single treatment fraction, and assembled these partial profile sequences into a contiguous data set for CBT reconstruction. The results have been encouraging. Figure 11 demonstrates the comparable quality of CBT reconstructions generated from continuous and between beam acquisitions. An added bonus comes in the form of the potential for quantification of patient shift during treatment using measurements from the reconstructed bony detail in the cone beam image volume itself.
View larger version (99K):
[in this window]
[in a new window]
|
Figure 11. (a) Cone beam tomography (CBT) coronal image generated from profiles gathered between treatment beam deliveries. (b) CBT coronal image for the same patient generated from image profiles gathered in a single continuous post-treatment scan.
|
|
Algebraic reconstruction algorithms [29] have already been researched as alternatives to FDK filtered back projection. Their versatility in the face of scatter and their potential to produce reconstructions with CT-number accuracy makes their practical development attractive. Furthermore, they marry well with the out of plane collection of projection data that improves the information content beyond the central reconstruction plane. Nutation of the X-ray source and image transducer during the execution of an otherwise circular scanning trajectory would allow the extra data to be gathered.
The value of taking projections from trajectories other than a strict circular rotation suggests that integration of the kilovoltage imaging system with the treatment gantry may be too restrictive. Indeed, as long as it is possible to co-register the cone beam imaging device with the local treatment machine coordinates there is no absolute reason for gantry integration. Furthermore, freed from the restriction on speed of therapy gantry rotation, a separate cone beam device could offer fast scanning that would improve work flow and reduce motion artefacts in the reconstructed image volumes. It would also avoid treatment machine down time caused by maintenance of the imaging system. Solberg et al [30] have reported their assessment of such a device from Siemens (Erlangen, Germany). This technology is a mobile, kilovoltage C-arm device that could be shared across a clinical department.
|
Conclusions
|
|---|
X-ray CBT for IGRT looks certain to expand into mainstream clinical use because it allows 3D setup and soft tissue visualization that is both practical and superior to the alternative megavoltage modalities. Imaging technology in the treatment room is now at least as capable as that deployed for treatment planning, taking into account the large 3D contiguous image volumes that can be produced. Frequent deployment of CBT in support of patient setup, disease targeting and adaptive treatment planning is unlikely to be limited by technology or dose considerations. It appears to be only a matter of time before the multiway flow of reconstructed volumetric images between planning and treatment departments will require a supporting infrastructure every bit as impressive as that already found in diagnostic radiology.
The authors gratefully acknowledge funding support from the Christie Hospital Centenary Appeal Fund, Elekta Oncology Systems, Cancer Research UK and The Royal College of Radiologists.
Received for publication July 1, 2005.
Revision received October 10, 2005.
Accepted for publication January 25, 2006.
|
References
|
|---|
- Feldkamp LA, Davis LC, Kress JW. Practical cone beam algorithm. J Opt Soc Am 1984;1:612–9.
- Lewis DG, Swindell W, Morton EJ, Evans PM, Xiao ZR. A megavoltage CT scanner for radiotherapy verification. Phys Med Biol 1992;37:1985–99.[CrossRef][Medline]
- ICRU Report 50. Prescribing, recording and reporting photon beam therapy. Washington, DC: International Committee on Radiation Units and Measurements, 1993
- Antonuk LE. Electronic portal imaging devices: a review and historical perspective of contemporary technologies and research. Phys Med Biol 2003;47:R31–R65.
- Antonuk LE, El-Mohri Y, Huang W, Jee KW, Siewerdsen JH, Maolinbay M, et al. Initial performance evaluation of an indirect-detection, active matrix flat-panel imager (AMFPI) prototype for megavoltage imaging. Int J Radiat Oncol Biol Phys 1998;42:437–54.[CrossRef][Medline]
- Turbell H, 'Cone Beam Reconstruction using Filtered Back Projection', Dissertation 672, February 2001, Linköping Studies in Science and Technology, Department of Electrical Engineering, Linköping University, Sweden
- Mosleh-Shirazi MA, Evans PM, Swindell W, Webb S, Partridge M. A cone-beam megavoltage CT scanner for treatment verification in conformal therapy. Radiother Oncol 1998;48:319–28.[CrossRef][Medline]
- Jaffray DA, Siewerdsen JH, Wong JW, Martinez AA. Flat-panel cone-beam computed tomography for image-guided radiation therapy. Int J Radiat Oncol Biol Phys 2002;53:1337–49.[CrossRef][Medline]
- European Standard EN 60601-2-1 Medical electrical equipment Part 2-1: Particular requirements for the safety of electron accelerators in the range of 1MeV to 50 MeV (IEC 60601-2-1:1998)
- Fahrig R, Holdsworth DW. Three-dimensional computed tomographic reconstruction using a C-arm mounted XRII: image-based correction of gantry motion nonidealities. Med Phys 2000;27:30–8.[CrossRef][Medline]
- Amer A, Sykes J, Moore CJ. Imaging panel skew correction and autofocussing in radiotherapy cone beam imaging. Procs IEEE Computer Society, Procs VIIIth Information Visualisation, 2004 July: 92-97
- Von Smekal L, Katchelriess M, Stepina E, Kalender WA. Geometric misalignment and calibration in cone-beam tomography. Med Phys 2004;31:3242–65.[CrossRef][Medline]
- Cho Y, Moseley DJ, Siewerdsen JH, Jaffray DA. Accurate technique for complete geometric calibration of cone-beam computed tomography systems. Med Phys 2005;32:968–83.[CrossRef][Medline]
- Hsieh J. Computed tomography; principles design, artifacts and recent advances. Washington, USA: SPIE Press, 2003
- The Ionising Radiation (Medical Exposure) Regulations 2000 UK. Statutory Instrument SI 2000, No 1059. London: HMSO, 2000
- Amer A, Marchant T, Sykes J, Henry A, Mcbain C, Moore CJ. The potential of X-ray volume imaging in adaptive radiotherapy to reduce the small bowel high dose volume in bladder cancer treatment. ESTRO-23; 2004 24-28 October; Amsterdam. Radiother Oncol 2004;73(Suppl-1): Abstr-513, S230
- Letourneau D, Wong J, Oldham M, Gulam M, Watt L, Jaffray D, et al. Cone-beam-CT guided radiation therapy: technical implementation. Radiother Oncol 2005. 75:279–86.
- Yamada Y. Image guided intensity modulated radiotherapy with the on board imager. ASTRO 46th Annual Meeting; 2004 October 3-7; Atlanta, USA. http://www.varian.com/orad/prd183.html# [Accessed 28 April 2006]
- Aird EGA. Second cancer risk, concomitant exposures and IRMER(2000). Br J Radiol 2004;77:983–5.[Free Full Text]
- Waddington SP, MacKenzie AL. Assessment of effective dose from concomitant exposures required in verification of the target volume in radiotherapy. Br J Radiol 2004;77:557–61.[Abstract/Free Full Text]
- EC Quality Criteria for Computed Tomography. EC Working Document EUR 1626. Brussels EU, 1998
- Amer A, Marchant T, Sykes S, Czajka J, Davies J, Stratford J, et al. Doses from cone beam CT integrated to a radiotherapy treatment machine. United Kingdom Radiation Oncology Conference; 2005 11-13 April; York, UK
- Moore CJ, Marchant T, Amer A, Sykes J, Henry A. Zonal X-ray volume imaging. ESTRO-23; 2004 24-28 October; Amsterdam, The Netherlands. Radiother Oncol 2004;73(Suppl-1): Abstr-111, S51
- Marchant T, Sykes J, Amer A, Moore CJ, Stratford J, Mccarthy C, et al. Measurement of setup error using image registration of X-ray volume images (XVI) and treatment planning CT. ESTRO-23; 2004 24-28 October; Amsterdam, The Netherlands. Radiother Oncol 2004;73(Suppl-1): Abstr-510, S229
- Sykes JR, Amer A, Czajka J, Moore C. A feasibility study for image guided radiotherapy using low dose, high speed, cone beam Xray volumetric imaging. Radiother Oncol 2005. 77:45–52.
- Sonke JJ, Zijp L, Remeijer P, van Herk M. Respiratory correlated cone beam CT. Med Phys 2005;32:1176–86.[CrossRef][Medline]
- Moore CJ, Marchant T, Amer A, Sykes J, Henry A. Zonal X-ray volume imaging. ESTRO-23; 2004 24-28 October; Amsterdam, The Netherlands. Radiother Oncol 2004;73(Suppl-1): Abstr-111, S51
- Amer A, Marchant T, Sykes J, Henry A, Mcbain C, Moore CJ. The potential of X-ray volume imaging in adaptive radiotherapy to reduce the small bowel high dose volume in bladder cancer treatment. Radiother Oncol 2004;73(Suppl-1): Abstr-513, S230
- Mueller K, Yagel R, Wheller JJ. Anti-aliased 3D cone-beam reconstruction of low contrast objects with algebraic methods. IEEE Trans Med Imaging 1999;18:519–37.[CrossRef][Medline]
- Solberg TD, Chow PE, Sorensen S, Wink N, Mitschke MM, Ritter D, et al. Image guided radiotherapy using a mobile kilo-voltage X-ray device. ASTRO 46th Annual Meeting; Abstr 2459; 2004 October 3-7; Atlanta, USA
This article has been cited by other articles:
|
|
|
|
 
P ROXBY, T KRON, F FOROUDI, A HAWORTH, C FOX, A MULLEN, and J CRAMB
Simple methods to reduce patient dose in a Varian cone beam CT system for delivery verification in pelvic radiotherapy
Br. J. Radiol.,
October 1, 2009;
82(982):
855 - 859.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
A.C. Miracle and S.K. Mukherji
Conebeam CT of the Head and Neck, Part 2: Clinical Applications
AJNR Am. J. Neuroradiol.,
August 1, 2009;
30(7):
1285 - 1292.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
A.C. Miracle and S.K. Mukherji
Conebeam CT of the Head and Neck, Part 1: Physical Principles
AJNR Am. J. Neuroradiol.,
June 1, 2009;
30(6):
1088 - 1095.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
R M Harrison, M Wilkinson, D J Rawlings, and M Moore
Doses to critical organs following radiotherapy and concomitant imaging of the larynx and breast
Br. J. Radiol.,
December 1, 2007;
80(960):
989 - 995.
[Abstract]
[Full Text]
[PDF]
|
|
|
Top
Abstract
Historical background and...
