Radiotherapy Unit, Royal Preston Hospital, Sharoe Green Lane North, Fullwood, ... patient and plan data between radiotherapy treatment planning systems ...
T he British Journal of Radiology, 70 (1997), 740–749
© 1997 The British Institute of Radiology
Methods for transferring patient and plan data between radiotherapy treatment planning systems 1J L BEDFORD, 1M OLDHAM, 2A HOESS, 1P M EVANS, 1,3G S SHENTALL and 1S WEBB 1Joint Department of Physics, Institute of Cancer Research and Royal Marsden NHS Trust, Downs Road, Sutton, Surrey SM2 5PT , UK, 2Abt. Medizinische Physik, Deutsches Krebsforschungszentrum (German Cancer Research Centre), Im Neuenheimer Feld 280, D-69120 Heidelberg, Germany, and 3Lakeland Radiotherapy Unit, Royal Preston Hospital, Sharoe Green Lane North, Fullwood, Preston PR2 9HT , UK Abstract. The effectiveness of conformal radiotherapy can ultimately only be assessed by the use of clinical trials. As large multicentre clinical trials become more widespread, methods of transferring patient and plan data between radiotherapy treatment planning systems become increasingly important. In this paper, the general strategy for the transfer of data is discussed, and also illustrated with reference to two specific systems: TARGET 2 (GE Medical Systems) and VOXELPLAN (DKFZ-Heidelberg). The transfer method involves using a computer program to translate the data formats used by each of the two systems for CT scans, patient outlines, plan information and block descriptions. This paper does not address the question of transferring beam data between systems: beam data must first be entered separately into both machines. The physical concepts encountered when transferring plans are described, with specific reference to the two planning systems used. Differences in the strategies used by the two planning systems for definition of irregular field shapes are compared. The dose calculations used by the two systems are also briefly evaluated. Isodoses produced by VOXELPLAN around a circular target volume are found to be up to 3 mm different in location to those produced by TARGET 2, owing to the use of a smooth field shape contour as opposed to a stepped field shape which closely models the leaves of a multileaf collimator. In general, dose distributions generated by both systems are comparable, but some differences are found in the presence of large tissue inhomogeneities. It is concluded that the transfer of patient and plan data between two different treatment planning systems is feasible, provided that any differences in field shape definition methods or dose calculation methods between the two systems are understood.
Introduction The goal of conformal radiotherapy is to irradiate a primary tumour with a sufficiently high dose to provide local control without irradiating surrounding normal tissues and sensitive structures, which would promote complications. The most common techniques by which this is performed in practice are the use of custom blocks shaped to the beam’s eye view of the planning target volume (PTV), and the appropriate use of a multileaf collimator (MLC). The success of the treatment can be further improved if a threedimensional (3D) treatment planning system is used to design novel beam configurations which have the possibility of delivering a high dose to the target while avoiding the nearby sensitive structures, or organs-at-risk (OARs). The immediate benefit of conformal radiotherapy is that the dose to the normal tissues can be reduced. Moreover, if the target dose in a Received 18 October 1996 and in revised form 23 January 1997, accepted 21 February 1997. This work is funded by the Cancer Research Campaign. 740
conformal treatment is escalated then, in principle, the dose to the normal tissues remains as it was with conventional radiotherapy. However, the target dose is higher, with a consequent improvement in the probability of local control. This is the basis of a dose escalation trial for radiotherapy of the prostate, currently being conducted at the Royal Marsden NHS Trust [1]. The success of conformal radiotherapy, 3D treatment planning and dose escalation can ultimately only be assessed by the outcome of a large number of patient treatments, involving large clinical trials. The number of patients treated in such trials can be significantly increased if several centres co-operate in the same trial. This raises technical difficulties as the different centres invariably use different planning systems and different treatment machines. Nevertheless, multicentre clinical trials are becoming more widespread, and methods for transferring treatment planning data between different systems are therefore becoming increasingly important [2, 3]. Although standard protocols are desirable, there is still a substantial amount of work to be carried T he British Journal of Radiology, July 1997
Radiotherapy treatment planning
out to achieve such protocols. This paper outlines some of the most significant physical concepts involved when transferring data between two different planning systems. The approach adopted here is that each centre has different treatment machines and must therefore collect its own beam data and enter it into its own treatment planning system. Conversely, patient and plan details are machine-independent and these data can therefore be transferred from centre to centre, regardless of the exact machine characteristics to be found at each centre. Each centre wishing to use a given treatment plan needs to recalculate the dose distribution using its own planning system and beam data. The principles of this approach are described and illustrated with reference to two specific treatment planning systems. It is stressed that although reference is made to specific planning systems for illustration purposes, the problems are general, so that analogous concepts will be involved when transferring data between any other twodimensional (2D) or 3D planning systems. For the above concept to be successful, it is necessary that dose calculations on both systems yield approximately the same results, and that the methods used for modelling field shape devices, such as MLCs and conformal blocks, are compatible. For example, if a plan has been optimized for one planning system and its beam data, then the dose calculation algorithm employed by another planning system, to which the plan is transferred, must be comparable or the plan will no longer be optimal. These types of issues are briefly discussed in the latter sections of this paper.
Method The two treatment planning systems being used for this work are VOXELPLAN (DKFZHeidelberg) and TARGET 2 (GE Medical Systems). VOXELPLAN is chiefly comprised of TOMAS, which is a 3D outlining program [4], and VIRTUOS, which is a 3D virtual-simulation program [5]. Tools have been incorporated into this system to provide for beam weight optimization and calculation of dose–volume histograms, tumour control probability, and normal tissue complication probability [6]. A facility for beam orientation optimization is also currently in the final stages of implementation. The paradigm used in this paper to represent a multicentre collaboration is as follows. A treatment plan is prepared on TARGET according to the normal clinical protocols, with suitable radiation beam parameters (e.g. number of beams, beam orientation, beam weight) decided upon by the experience of the planner. In particular, a pair of opposed blocks is used to represent MLC leaves fitted around the beam’s-eye-view of the planning T he British Journal of Radiology, July 1997
target volume. The plan is then transferred to VOXELPLAN for 3D visualization, optimization and calculation of treatment statistics. To do this, CT scans, PTV and OAR outlines, beam parameters and block details are extracted from TARGET using purpose-designed computer software. The information is converted to VOXELPLAN file format and transferred to the computer on which VOXELPLAN is resident, where the plan can be optimized and dose statistics calculated. For example, beam weights and orientations can be adjusted or optimized, and new beams can also be added. By defining two coincident beams, one of which is wedged and the other of which is unwedged, the effective wedge angle can also be optimized [7]. The modified beam and block parameters are then transferred back to TARGET where a final dose computation takes place. The CT scan and outlines are not transferred from VOXELPLAN back to TARGET, as these data are already resident on the TARGET system. Specific issues relating to the transfer of data between any two treatment planning systems are now discussed, with specific reference to the above paradigm.
Operating system Treatment planning programs run on a variety of computers, and under several operating systems. Although the precise type of computer and vendor are immaterial for the purposes of data transfer, the operating system is important. The most common operating systems used currently for treatment planning computers are DOS, Unix and VMS. Attention has to be given to the means by which data can be transferred meaningfully between these operating systems. For the case of our experiment, TARGET runs on a Sun workstation under Unix, while VOXELPLAN runs on a DEC Alpha under OpenVMS. The transfer procedure for moving data between TARGET and VOXELPLAN must therefore make the transition between the two operating systems as well as converting between the file formats of the two systems. We have chosen to use File Transfer Program (FTP) to make the transition between the computers hosting TARGET and VOXELPLAN. This conveniently allows access to remote machines over the Internet, handling changes in operating system in the process.
File format The main issue confronting users of multiple treatment planning systems is that of file formats. Each different planning system makes use of a different data format, and if data are to be exchanged between two systems, the format has to 741
J L Bedford, M Oldham, A Hoess et al
be changed. This is a perennial problem in all areas of medical computing, including medical imaging. In the imaging field, the DICOM standard has been developed to overcome the problem, but is complex and by no means definitive. Similarly, a DICOM-RT protocol has been suggested for the transfer of radiotherapeutic information, but has not yet been fully developed. In the absence of a standard data format for radiotherapy, we have developed our own strategy for data transfer, called TV-TRANSFER. This consists of a computer program, written in C++, which resides on the TARGET workstation. It extracts the relevant files from the TARGET patientoriented database (POD) and converts them into the format required by VOXELPLAN, leaving the files in a single directory on the TARGET machine. The program is structured so that each data modality (i.e. CT, outlines, plan, blocks) is handled by a separate C++ derived class. Each derived class inherits two base classes, one of which models the TARGET version of the data and the other the VOXELPLAN version. Further details of the physical principles underlying the data transfer process are now given.
Physical issues It is more than likely that two different treatment planning systems will use CT data of differing resolution, and resampling is therefore necessary. This can be conveniently achieved by scanning a square pixel, of the size required by the destination system, across the existing image, using linear interpolation to determine the required intensity. This process is described in detail in Appendix A for the case of TARGET and VOXELPLAN; the use of CT data in these two systems being representative of that in most treatment planning systems. Outlines of PTV, OARs and heterogeneities are also stored differently in various treatment planning systems. Here the principal issues are (1) organization, i.e. whether by slice or by volume of interest, (2) co-ordinate axes and (3) units of measurement. These factors are briefly discussed in Appendix B for our experimental arrangement. Transferring plan data between two treatment planning systems primarily involves changing the organization of the list of treatment beams, and altering the co-ordinate systems where necessary. Furthermore, the two planning systems may not store exactly the same set of parameters, so some variables may be discarded during the transfer process, and others appropriately generated. Appendix C describes the most important features of plan conversion between TARGET and VOXELPLAN. Methods of specifying blocks and other field shaping devices vary considerably between planning 742
systems. Some of the more advanced planning systems are designed for use in conjunction with an MLC, whereas other systems assume more conventional field shaping methods such as blocks. The exact requirements for the transfer procedure are therefore fairly system specific, but in all cases the field shape or block shape is given as a list of co-ordinates, along with the source–carrier distance. Invariably, some kind of co-ordinate transformation must take place when these co-ordinates are translated from one system to another. These issues are illustrated in more detail in Appendix D.
Geometric accuracy The geometric accuracy of TV-TRANSFER was tested by transferring a variety of plans in both the TARGETVOXELPLAN and VOXELPLANTARGET directions. The integrity of a transferred CT scan was checked by examining the Hounsfield number for several different parts of the image, and also the pixel size. The accuracy of outline transfer was tested by outlining the couch top using TARGET and then checking that the outlines still matched the couch top after transfer to VOXELPLAN. The plan parameters (e.g. gantry angle, couch angle) were checked for a variety of test plans which were transferred in both directions. Similarly, to check the integrity of the field shape transfer, several plans were transferred from TARGET to VOXELPLAN and back again and the geometric shape of the field was examined against the PTV and against the field shape from which it was derived.
Dosimetric comparisons In order for planning tools available within one treatment planning system to be effective when the plans are subsequently transferred to another system, it is essential that the dose calculation methods of the two systems are comparable. If this were not the case, then an optimization, for example, would yield an optimal solution within the one system, but this solution would be less than optimal when transferred to the other. This section therefore briefly describes the dosimetric comparisons made between TARGET and VOXELPLAN. In particular, two important differences between the strategies used by TARGET and VOXELPLAN for the definition of conformal fields are presented in detail, and an extreme comparison involving a heterogeneity is also considered. The remainder of the dosimetric comparisons are summarised very briefly as they are of little interest. The first difference between the calculation methods used by TARGET and VOXELPLAN T he British Journal of Radiology, July 1997
Radiotherapy treatment planning
arises because the Philips MLC used at the Royal Marsden NHS Trust [8, 9] is such that the field boundary is defined partly by the leaves of the MLC and partly by the secondary collimator. This effect can be correctly modelled by a pair of blocks on TARGET, allowing for differences in the penumbral characteristics between the MLC leaves and the secondary collimator. However, VOXELPLAN does not distinguish between the MLC and the collimator, thereby making no distinction between the different penumbrae of these devices. A second important difference between the methods used by TARGET and VOXELPLAN is that the blocking method used on TARGET models the individual leaves of the MLC, whereas VOXELPLAN uses an arbitrary field shape surrounding the beam’s-eye-view of the PTV. For instance, if a plan is transferred from TARGET, then TV-TRANSFER creates the appropriate MLC-shaped field as described above. However, if the beam orientation is subsequently changed within VOXELPLAN, the beam’s-eye-view of the PTV changes and the field shape therefore has to be modified accordingly. In this case, VOXELPLAN generates a smooth field shape with a uniform margin around the PTV: the exact shape of the MLC is lost. For example, in the case of a spherical PTV, TARGET produces a series of stepped edges representing the MLC leaves, while VOXELPLAN produces a smooth, approximately circular field shape. This effect, which causes VOXELPLAN’s representation of the Philips MLC to be less accurate than that of TARGET, arises because VOXELPLAN has been partly designed for planning treatments with an MLC
(a)
whose leaf-width is around 1 mm, i.e. much smaller than that of the Philips MLC. With such a small leaf-width, the stepping effect of the leaves becomes negligible and the field edge becomes smooth. When using the Philips MLC with its wider leaves, it is necessary to modify the calculated smooth field shape manually to incorporate the leaf undulations. However, this manual procedure is impractical for the 80 leaf positions which have to be specified for each treatment field, so for the purposes of this paper, VOXELPLAN is considered as producing a smooth field shape. To examine the differences in dose distribution introduced by the smooth field shape used by VOXELPLAN, a treatment plan was produced on TARGET for a 50 mm diameter cylindrical PTV located centrally within a cuboidal water phantom of side 200 mm. The isocentre was located at the centre of the PTV and MLC-shaped blocks were designed, leaving a 6 mm margin around the PTV. The MLC leaves were positioned such that the 50% isodose just encompassed the edge of the margin. With this margin, the nominal field size as defined by the secondary collimators was 63×63 mm (Figure 1a). A single 6 MV radiation beam was applied normal to the phantom surface and the dose distribution was calculated in a plane through the centre of the PTV, transverse to the central axis of the beam. The plan, including the field shape, was then transferred to VOXELPLAN. However, on arrival at VOXELPLAN, the field shape was deliberately disregarded and a new field shape was designed using the automatic field shape calculation facility available within VOXELPLAN itself. This was not the normal procedure, but
(b)
Figure 1. (a) Blocks designed to simulate the leaves of a Philips MLC surrounding the beam’s-eye-view of a circular PTV. ( b) The corresponding irregular field shape as designed using the VOXELPLAN field shape calculation function. The slight lack of symmetry in the field shape is attributed to the coarseness of the pixel grid on which it is calculated (pixel size 1.2 mm). T he British Journal of Radiology, July 1997
743
J L Bedford, M Oldham, A Hoess et al
was carried out to examine the effect of the smooth field shape calculated by VOXELPLAN. VOXELPLAN requires that the aperture delineated by the secondary collimators is larger than that defined by the field shaping device (e.g. conformal blocks). For this reason, the nominal field size as given by the position of the secondary collimators was increased to 79×79 mm to ensure that the irregular field was completely irradiated (Figure 1b). The modified field shape was transferred back to TARGET, with the field shape becoming a pair of conformal blocks with a circular aperture. The dose was then recomputed. As the dose calculations for the MLC field and smooth conformal field were both performed by TARGET, any differences between the dose distributions were attributable to the effects of modelling the MLC either fully or simplistically. Note that this experiment models the situation that occurs when a plan is transferred from TARGET to VOXELPLAN, the beam orientations changed, and the resulting plan transferred back to TARGET. However, if the beam orientations are not changed on VOXELPLAN, as is the case when only beam weights are optimized, then the MLC shape is preserved. To test the similarity of the dose computations performed by TARGET and VOXELPLAN, a number of test plans was compared, including beams perpendicular and oblique to the surface of the above cuboidal water phantom, and wedged beams. In addition, to test the relative performance of the calculation algorithms in the presence of inhomogeneities, a plan was created for a cuboidal water phantom of side 200 mm with a 90×90 mm region of relative electron density 0.26 ( lung
(a)
density) running along its anterior left quadrant such that the left and anterior surfaces of the low density region were 10 mm deep and the right and posterior surfaces coincided with the centrelines of the phantom. A 60×60 mm 6 MV beam was directed squarely at the centre of the anterior surface of the phantom, so that the central axis lay along the edge of the inhomogeneity. A dose distribution was produced by TARGET, then the plan was transferred to VOXELPLAN and the calculation repeated. This was intended to demonstrate the relative performance of the calculation algorithms used by the two planning systems for a particularly demanding case. Note that both systems use a comparable Bentley–Milan model with radiological path length scaling [10, 11]. In the case of VOXELPLAN, the calculation algorithm is DOSREC4, which is not the most sophisticated of the four available VOXELPLAN algorithms, but is currently the most applicable to the work being performed at the Royal Marsden NHS Trust. The same beam data were entered into both TARGET and VOXELPLAN.
Results The results of the TARGET dose calculations for MLC-shaped and circular beam portals are given in Figure 2. The dose distribution for the MLC-shaped beam portal contains irregularities due to the stepped nature of the MLC leaves. As expected, the dose distribution for the circular beam portal is relatively circular and smooth. For the MLC-shaped field generated on TARGET, the PTV is encompassed by the 90% isodose, whereas for the circular field generated on VOXELPLAN,
(b)
Figure 2. Dose distributions calculated using TARGET for the case of a circular PTV in a cuboidal water phantom. The dose plane is perpendicular to the central axis of the single radiation beam. (a) Field delineated by MLC-shaped blocks. ( b) Field surrounded by smooth conformal blocks. Isodoses are in percentages, normalized to the centre of the PTV. 744
T he British Journal of Radiology, July 1997
Radiotherapy treatment planning
the PTV is encompassed by the 95% isodose. This difference is a consequence of the differing methods of defining the field edge. Both plans incorporate a 6 mm margin around the PTV but part of each MLC leaf intrudes within this margin for the MLC-shaped field generated on TARGET and part is without, whereas the conformal blocks originating from VOXELPLAN follow the margin more faithfully. Methods of calculating the nominal 6 mm margin differ between the two systems, with the VOXELPLAN margin exhibiting a slight asymmetry, possibly due to the coarseness of the pixel grid on which it is calculated. The equivalent square for the MLC-blocked plan is thus 57 mm, whereas for the conformal blocked plan it is 60 mm. The effect of the MLC irregularity is to shift the 90% isodose by up to about 3 mm around certain parts of the PTV, relative to the conformal blocks. This effect will be similar in magnitude for all field sizes, but as the PTV chosen here is at the lower end of the range of sizes likely to be encountered clinically, its effect is fairly pronounced. The comparisons of dose distributions produced by TARGET and VOXELPLAN yielded isodose plots in which the dose was comparable to within 2% in low gradient regions for open fields and homogeneous media. For wedged beams the similarity was less, with about 4% difference between the two systems. This was probably due to VOXELPLAN’s use of wedge factors for computation of the wedged depth doses, as opposed to TARGET’s use of measured depth dose data. This discrepancy could also have been explained by differences in normalization strategy. The level of difference was considered acceptable, given the
(a)
International Commission on Radiation Units and Measurements (ICRU) recommendation of 2% maximum difference between calculated and measured dose [12], and an acceptability of 3–4% specified by Van Dyk et al [13]. Our results are relative differences between TARGET and VOXELPLAN so that, in principle, both systems could be in error, but TARGET is clinically validated and its dose calculation is therefore assumed to be correct. Moreover, in the context of this paper it is reasonable to assume that planning systems being used for a clinical trial are clinically validated, with each respective system accurate to within the agreed standards for dose calculation. Hence, it is relative dose results which are important. The outcome of the dose comparison between TARGET and VOXELPLAN for the case of a heterogeneity is shown in Figure 3. In the regions of low dose gradient, the dose computed by the two systems is comparable to within 2%. At the penumbra, the correspondence is to within 2 mm, except in the low density region where the discrepancy is around 3 mm, and below the low density region where a difference of 3–5 mm is observed. Since this example is a particularly extreme case of heterogeneity, it is considered that the two dose distributions are sufficiently comparable to enable VOXELPLAN to be used in conjunction with TARGET.
Discussion This paper has outlined some of the most significant physical concepts involved when transferring
(b)
Figure 3. Comparison of dose calculations using (a) TARGET and (b) VOXELPLAN for the case of a water phantom with a lung density portion in the anterior left region. The beam has a field size of 60×60 mm and is directed perpendicular to the anterior surface of the phantom. Isodoses (%) are normalized to the dose maximum. T he British Journal of Radiology, July 1997
745
J L Bedford, M Oldham, A Hoess et al
data between two different planning systems, for the purpose of multicentre clinical trials. Although we have referred extensively to two specific examples of such systems, analogous concepts are involved when transferring data between other 2D or 3D systems. For example, work is currently in progress at the Royal Marsden NHS Trust to transfer data between TARGET and CADPLAN ( Varian Associates, Palo Alto, California), and this involves very similar processes to those described above. Each different centre in a large collaboration has its own treatment machines with their own beam characteristics. In this paper, it is therefore assumed that appropriate beam data are entered separately into each of the collaborating treatment planning systems. Before treatment, the plan parameters transmitted from another treatment planning system are used to calculate the dose distribution using the local beam data. This requirement stems from the need to model accurately the radiation beam with which patients are to be treated. In contrast, if the same beam data are entered into two communicating systems, then the dose distributions calculated by both systems should be approximately equal, so that a plan deemed as optimal on one of the systems remains optimal when transferred to the second system. For example, a plan which is optimized for, say, a 6 MV photon beam can be transferred to any other system with comparable beam data and still remain optimal. In practice, some loss of optimality may be expected due to inconsistency of beam data between the two systems, but the effect of this can be readily examined by recalculation of the dose distribution as described above. For the two specific planning systems used as a demonstration, it has been shown that the dose distributions calculated by both systems are comparable, provided that the beam data employed by each system are the same, thus enabling these planning systems to be used conjunctively. However, it has been shown that the method by which an MLC is modelled has an effect on the calculated dose distribution. If the beam portal is assumed to have a smooth outline parallel to the edge of the PTV, then the isodoses are shifted by several millimetres relative to those produced by modelling the precise shape of the MLC leaves. This effect is most noticeable for small PTVs where the shift is larger relative to the PTV diameter. Whether the modification of the isodoses caused by assuming a smooth portal edge is significant or not ultimately depends upon the size of the tumour, the clinical margin and the margin surrounding the PTV, as all of these factors govern the dose accuracy required. In conformal treatments, however, the margins are frequently small, and thus 746
the precise locations of the 90 and 95% contours are important. Clearly, if conformal blocks are used for treatment, then the modelling of the MLC is no longer relevant, and the method used by the planning system for delineation of the fields is less important.
Acknowledgments VOXELPLAN was provided by the German Cancer Research Centre, DKFZ-Heidelberg, as part of a continuing collaboration, and the co-operation of Professor Wolfgang Schlegel and the group at DKFZ is appreciated.
References 1. Shentall GS, Latimer PA, Reise SP. The choice of technique for MLC conformal therapy of the prostate. Radiother Oncol 1995;37, Suppl. 1:S7. 2. Bosch WR. Data management for 3D CRT. Med Phys 1996;23:1093. 3. Torresin A, Moretti R, Palazzi M, et al. Patient data acquisition and definition of volumes of interest. In: Andreucci L, editor. Evaluation of 3-D Treatment Planning Systems for Clinical Use in Radiotherapy. Pisa, Italy: Giardini, 1996:107–47. 4. Pross J. Tool for manual segmentation of volumes in multi modality 3D imaging. In: Minet P, editor. Three-Dimensional Treatment Planning. Geneva: EAR, 1993:245–51. 5. Bendl R, Pross J, Hoess A, et al. VIRTUOS—a program for virtual radiotherapy simulation and verification. In: Hounsell AR, Wilkinson JM, Williams PC, editors. Proceedings of 11th International Congress of Computers in Radiotherapy (Manchester). Manchester: ICCR, 1994:226. 6. Oldham M, Neal A, Webb S. A comparison of conventional ‘forward planning’ with inverse planning for 3D conformal radiotherapy of the prostate. Radiother Oncol 1995;35:248–62. 7. Oldham M, Neal AJ, Webb S. The optimisation of wedge filters in radiotherapy of the prostate. Radiother Oncol 1995;37:209–20. 8. Jordan TJ, Williams PC. The design and performance characteristics of a multileaf collimator. Phys Med Biol 1994;39:231–51. 9. Fernandez EM, Shentall GS, Mayles WPM, Dearnaley DP. The acceptability of a multileaf collimator as a replacement for conventional blocks, Radiother Oncol 1995;36:65–74. 10. Milan J, Bentley RE. The storage and manipulation of radiation dose data in a small digital computer. Br J Radiol 1974;47:115–21. 11. Johns HE, Cunningham JR. The Physics of Radiology (4th edn). Springfield, IL, USA: Charles C Thomas, 1983:391–2. 12. International Commission on Radiation Units and Measurements. Use of computers in external beam radiotherapy procedures with high-energy photons and electrons, ICRU Report 42. Bethesda, USA: ICRU, 1987. 13. Van Dyk J, Barnett RB, Cygler JE, Shragge PC. Commissioning and quality assurance of treatment planning computers. Int J Radiat Oncol Biol Phys 1993;26:261–73. T he British Journal of Radiology, July 1997
Radiotherapy treatment planning
Appendix A CT transfer TARGET stores a CT scan as a series of transversal image slices, each of which has a distinct zoffset relative to the zero position of the CT scanner couch. Meanwhile, VOXELPLAN stores the entire volumetric scan in a single contiguous file, with each slice followed immediately by the next, so that the axial position of a transversal slice is inferred by its location within the file. The transfer program therefore merges the TARGET slices together and simultaneously extracts the associated header information to create the required VOXELPLAN image file and separate header file. There is also a change in scan resolution as TARGET images are 320×320 pixels in size, whereas VOXELPLAN images are 256×256 pixels wide. For a typical pelvic scan, this corresponds to a pixel size of about 1.5×1.5 mm on TARGET and 1.9×1.9 mm on VOXELPLAN. TV-TRANSFER therefore re-samples the TARGET CT images to achieve the lower resolution. The low resolution image is generated by scanning through each of the pixel locations in a 256×256 image grid, calculating the intensity from the corresponding location in the high resolution scan. If the two different image grids are considered to be overlaid then, in general, each low resolution pixel overlaps part of up to nine high resolution pixels. The resampled intensity is thus given as I=(a1.I1+a2.I2+…+a9.I9)/A, where I1,…,I9 are the intensities of the nine pixels, a1,…,a9 are the corresponding areas by which they overlap the single larger pixel, and A is the total area of the larger pixel. This procedure yields accurate, but lower resolution images as required by VOXELPLAN.
Appendix B Outline transfer When a volume-of-interest is outlined on TARGET , the resulting set of outlines is stored on a slice-by-slice basis. A given slice of the outline set typically contains contour, tumour, OAR and heterogeneity outlines, all relating to that slice. In contrast, VOXELPLAN stores an outline set according to the volumes-of-interest, with each volume-of-interest containing a series of outlines relating to the slices on which that structure is outlined. The conversion procedure must therefore translate the data between these two organization strategies. Furthermore, TARGET has 2D Cartesian axes whose origin lies at the centre of the CT images, while the z co-ordinate is implied by the slice number, which has an offset relative T he British Journal of Radiology, July 1997
to the reference plane of the CT scanner. Co-ordinates are specified in units of mm/16. However, VOXELPLAN’s co-ordinate system has its origin at the top left-hand corner of the CT images, with the z co-ordinate is implied by the slice number, which has an offset relative to the most inferior CT slice. Co-ordinates are specified in units of (CT pixel width)/16. Co-ordinate transformations to convert between these distinct co-ordinate systems are therefore performed during the transfer process.
Appendix C Plan transfer The file format of a TARGET treatment plan consists primarily of a machine descriptor, which contains the details of up to three treatment machines, and a series of beam descriptors, each of which holds the parameters of one radiation beam. Each beam descriptor is associated with one of the machines by means of a linked list mechanism. The isocentre co-ordinates are in units of mm/16, relative to the co-ordinate system used for the storage of outlines. A VOXELPLAN treatment plan is chiefly comprised of a list of beams, each with the machine information included; isocentre co-ordinates are in units of millimetres relative to the VOXELPLAN outline co-ordinate system. In general, there is a simple correspondence between the variables in each of the two systems, so that conversion is a straightforward one-to-one mapping. However, VOXELPLAN does not store the source-to-surface distance (SSD) in its file structure, so when converting a VOXELPLAN treatment plan to a TARGET treatment plan, the SSD must be constructed for each beam by finding the intersection of the beam axis with the 3D patient contour. Since the patient contour is stored as a set of discrete points, it is difficult to find the true intersection, but as the points are only spaced by a few millimetres, it is sufficient to find the nearest point to the intersection instead. If the contour were a simple convex shape, this could be achieved simply by finding the nearest point to the radiation source, but as the contour may contain concavities, a more sophisticated method must be used (Figure 4): 1. Starting at the source, the nearest contour point, P, is found and recorded, and the distance, D, from the source to this point is also recorded. Note that P may lie on any of the transversal slices. 2. Move a small interval towards the isocentre from the source. Again find the nearest contour 747
J L Bedford, M Oldham, A Hoess et al
Figure 4. Method used to construct the SSD. The program steps from source to isocentre, searching for the contour point which minimizes the distance D. This point is then used to approximate the SSD. (Note: contour points are widely spaced for clarity).
point. If the distance from the current point on the central axis to this point is lower than the previous distance recorded, update P and D. 3. Repeat this procedure, stepping along the central axis of the beam, from the source to the isocentre. 4. Calculate the distance from the last recorded value of P to the source. This is a close approximation to the SSD. This process succeeds because, at some stage, the current point on the central axis lies in or very near to the patient contour and D becomes minimal. P then records the point in the contour which lies closest to the intersection of the central axis with the contour. The algorithm as implemented steps along the central axis at intervals of one-hundredth of the source–axis distance, which corresponds to an interval length of 10 mm. This typically provides the SSD to an accuracy of about 1 mm. Note that this is a 3D method which involves using all of the 2D transversal contours simultaneously to form a 3D contour.
collimator. The block co-ordinates specify the exact size of the block according to the source–carrier distance. Within VOXELPLAN, an irregularly shaped field is specified simply by a list of co-ordinates giving the outline of the field as it would appear at the isocentric plane. These co-ordinates are in units of CT pixel width and are referred to a co-ordinate system whose origin is at the top left of the beam’s-eye-view window as it appears on the screen, where the X2 collimator appears at the top of the window and the Y2 collimator appears at the left. Figure 5 illustrates the method of converting between a pair of MLC-shaped blocks on TARGET and an irregular field shape on VOXELPLAN. Starting from the point P1 where the edge of the Y1 block crosses the X1 collimator, the algorithm traces around the Y1 block, copying the co-ordinates. At points where the block edge lies behind the Y1 collimator (e.g. point P2), the x co-ordinate of the Y1 collimator is recorded rather than the x co-ordinate of the block edge, since in this case it is the collimator which delineates the field. When the edge of the Y1 block crosses over the X2 collimator (point P3), the algorithm records the x co-ordinate of the Y1 block and the y co-ordinate of the X2 collimator. The algorithm then finds where the Y2 block crosses the X2 collimator (point P4) and stores the x co-ordinate of the Y2 block and the y coordinate of the X2 collimator. It then continues copying the co-ordinates for the second block. Finally, when the edge of this second block crosses the X1 collimator (point P5), the algorithm completes the field boundary by joining up to the
Appendix D Block conversion Using TARGET , there is no explicit means of defining an irregularly shaped radiation field. Hence, a pair of opposing blocks are used to simulate the Y1 and Y2 leaves of a Philips MLC. Conformal fields can also be planned by defining a single block with an aperture in the middle or a pair of adjacent blocks forming the same arrangement. Each block is stored as a list of co-ordinates defining the block shape. The co-ordinates are in units of mm/16 and are referred to a co-ordinate system which has its origin on the central axis of the beam, with the x-axis directed towards the Y1 collimator and the y-axis directed towards the X2 748
Figure 5. Use of two blocks to simulate the Y1 and Y2 leaves of the MLC. The broken lines denote the secondary collimators, the solid lines indicate the MLC leaves, and the bold lines indicate the field shape transferred to VOXELPLAN. Note that the field is delineated partly by the MLC leaves and partly by the secondary collimators. T he British Journal of Radiology, July 1997
Radiotherapy treatment planning
starting position P1. Co-ordinate transformations are then applied to convert between the two different co-ordinate systems. Conversely, when converting a VOXELPLAN irregular field shape into a pair of TARGET blocks, the transfer algorithm finds the extreme x
T he British Journal of Radiology, July 1997
co-ordinates of the field shape and bisects them to find the approximate centreline of the field. The program then copies the field shape co-ordinates for half of the field and generates a block by adding an exterior block outline. This is then repeated for the second half of the outline.
749