Radiat Environ Biophys (1992) 31:241-245
Radiation and Environmental Biophysics © Springer-Verlag 1992
3D treatment planning for heavy charged particles* Michel Goitein Massachusetts General Hospital, Department of Radiation Oncology, Harvard Medical School, Boston MA 02114, USA Received December 16, 1991 / Accepted February 17, 1992
Summary. The comments herein describe, at a necessarily superficial level, a number of issues which must be addressed in developing plans for heavy charged particle therapy. Programs now exist which provide the needed capabilities. The challenge now is to make the planning process easier and faster- and possibly more elective. It seems likely that this will be achieved in the next few years.
Introduction Heavy charged particles (heavier than electrons) have been used for radiation therapy for over three decades. However, it has only been since the advent of computed tomography (CT) in the mid 1970's, and the availability by the beginning of the 1980's affordable fast computers and interactive image and graphics displays, that is has been possible to plan and deliver charged particle radiation with confidence anywhere in the body. This is because, if particles are to have any advantage over photons, their energy must be tailored so that they do not exit the patient distally but, rather, are caused to stop within or at most very slightly beyond the target volume. For this to be achieved one must know where the target margins are and one must have a " m a p " of the patient's tissues to ascertain where particles of a given energy will come to rest. Until CT was available, these things could be known with confidence only in quite restricted anatomic sites (mainly the in the brain). Because the need was so acute, the technologies mentioned above were first taken advantage of by those engaged in heavy charged particle therapy - as soon as the technologies became available. However, it was immediately clear that the capabilities so developed had much broader application to conventional * Invited paper given on the fourth workshop on "Heavy Charged Particles in Biology and Medicine" GSI, Darmstadt, FRG, September 23-25, 1991
242 x-ray therapy. As a result, programs which use imaging information to develop 3D approaches to delivering radiation are now beginning to be used widely in radiation therapy and, no doubt, their development will feed back into, and enrich, particle therapy. The topic of planning treatment with heavy charged particles was addressed in 1982 by Goitein et al. [1] and much of the relevant literature up to that time was cited there. The following discussion does not duplicate that information, to which the reader is referred; only subsequent references are cited here. The requirements for planning therapy for those heavy charged particles which exhibit high LET behaviour are very similar to those for low LET charged particles (primarily protons) - but have additional and complex requirements relating to the usually non-uniform spatial (and even temporal) distribution of LET and related quantities. These additional issues are not further discussed here.
Compensation for inhomogeneities In treating a specified target volume the beam penetration must generally be tailored over the field area to account for three things: (I) the shape of the entrance (usually skin) surface; (2) the shape in depth of the distal surface of the target volume; and (3) the density distribution throughout the traversed tissues. In static beam treatments, this is generally done by designing a compensating bolus. As already alluded to, it is vital in most situations to have a measurement of the patient's tissue densities and then to take them into account in designing a compensating bolus. CT scan data, suitably scaled to represent effective density rather than linear x-ray attenuation coefficients, is ideal for this purpose [1]. The technique we use to fabricate boli in the M G H / H C L program has been described by Wagner et al. [3]; this approach or something quite similar to it seems to be widely used. In addition to their influence on the design of the beam penetration map, inhomogeneities also affect the beam modulation (the extent to which dose is spread out over depth). Inhomogeneities must be taken into account both for treatments employing fixed modulation, and for scanned beam approaches where the modulation can be varied over the beam cross section. In both cases, the modulation depth depends on the inhomogeneities within the target volume(s).
Registration of the patient with the treatment beam and other uncertainties There are number of uncertainties in beam delivery and dose estimation. Probably the foremost among these is the uncertainty in the registration between the treatment beam and the patient. Of course, this uncertainty exists in conventional x-ray therapy too and is handled primarily by expanding the treatment field margins. The problem is exacerbated in heavy charged particle therapy for two reasons: (I) because compensation for inhomogeneities relies on appropriate a registration between bolus and patient - a misregistration can give
243 rise to either hot or cold spots, or both [1]; and (2) the application of heavy charged particle therapy is predicted on taking advantage of the possibility of closely tailoring the treatment volume to the target volume - one therefore is desirous of minimizing the volume of tissue irradiated for non-medical indications. Techniques have therefore been developed [4] to assess quantitatively the uncertainties in treatment delivery, including those of misregistration, and of designing compensating boli taking into account misregistration and multiple scattering problems [5] - the smearing technique of [5] is the three-dimensional analogue of the field margin enlargement approach in x-ray therapy. These uncertainties must also be taken into account in designing the beam modulation. Definition of anatomy and design of beams
The approaches needed for the delineation of patient anatomy, both for normal tissues and target volumes, and for the design of treatment fields are very analogous to those needed in conventional x-ray therapy - or, considering the historical sequence in which 3D planning developed, one might express this v i s a versa. One needs: interactive drawing tools; automated feature extraction; margining tools; beam's eye view (BEV) assessment of anatomy and field direction; BEV design (both manual and automatic) of blocks and apertures; BEV design of compensating boli; sectional display of beams and beam modifying devices; etc. [6, 7]. In addition, now that they are readily available, one must be able to use MRI studies for these purposes and must be able to register CT and MRI studies with one another. In some ways BEV planning with heavy charged particles is easier than with x-rays because the resulting doses are more readily intuited. Rarely does an experienced planner need to see computed dose distributions to conceptualize the impact of a given beam on the plan. One aspect that is different and requires additional support is the use of patched fields to augment one another [8]. Here additional planning tools are needed, although none are yet available. Dose calculation
There are three broad approaches to computing the radiobiologically effective dose of heavy charged particles: (1) broad beam models; (2) pencil beam models; and (3) Monte Carlo approaches. These are distinguished by an ascending order of both accuracy and computational complexity, hence time. To date, broad beam models have been mostly used in routine clinical practice, with dose distributions as measured in cuboid water phantoms being modified by line-of-sight corrections for inhomogeneities and allowance for the distance from the effective source to the point of interest. While broad beam models have been valuable for practical clinical work, they have some important limitations. They do not model a number of the dose perturbations induced by finely structured inhomogeneities (although an error analysis [4] can generally ensure that the actual dose is properly bracketed by the uncertainty bounds placed on the calculated dose [5]); and they do not take proper account of the penumbra-blurring effect of bolus/patient air gaps [9].
244 The increased computing capabilities of modern affordable computers, such as workstations, are rapidly making it possible to implement more accurate algorithms. Pencil beam approaches, such as those used in the work described by Scheib and Pedroni below and that recently reported by Petti et al. [10], are likely to improve the accuracy of dose calculations - and hence potentially further reduce treatment volumes. Monte Carlo calculations have to date been confined to the exploration of generic (i.e. not specific to an individual patient) issues - primarily those having to do with compensation for inhomogeneities [1]. However, it seems likely that it will not be too long before it becomes practicable to perform Monte Carlo calculations for individual patient treatments.
Dose analysis: assessment and comparison of plans A discussion of 3D treatment planning would not be complete without mention of the problem of assessing a 3D treatment plan - and, more importantly, the problem of comparing two or more rival plans. As Urie describes in the paper below, heavy charged particle therapy has - both for the historical reasons already alluded to, and for some intrinsic reasons - played a leading role in this area. Advocates of heavy charged particle therapy have used comparative treatment planning approaches, in which heavy charged particle plans have been compared with plans using conventional x-rays, in order to assess objectively the possible advantages of heavy charged particles. Given that the plan assessment tools must in these situations be applied to both the heavy charged particle and x-ray plans, one can see that, almost by definition, the tools for plan assessment are not particle-specific (excepting only the assessment of high L E T effects). The tools in question have included: side-by-side presentation of dose distributions in sectional images; use of color-wash dose displays; dose-difference displays; dose-volume histograms; the presentation of dose statistics; and biophysical modelling of both tumor control probabilities and normal tissue complication probabilities [11]. As Urie's paper makes clear, these tools also make possible invaluable quantitative analysis of clinical experience.
Acknowledgements. Dr. Goitein's work was funded in part by a grant from the DHHS (CA 21239).
References 1. Goitein M, Abrams M, Gentry R, Urie M, Verhey L, Wagner M (1982) Planning treatment with heavy charged particles. Int J Radiat Oncol Biol Phys 8 :2065-2070 2. Steward VW, Koehler AM (1973) Proton beam radiography in tumor detection. Science 173:913-914 3. Wagner M (1982) Automated range compensation for proton therapy. Med Phys 9 : 749-752 4. Goitein M (1985) Calculation of the uncertainty in the dose delivered in radiation therapy. Med Phys 12: 608-612 5. Urie M, Goitein M, Wagner M (1984) Compensating for heterogeneities in proton radiation therapy. Phys Med Biol 29 : 553-566 6. Goitein M, Abrams M (1983) Multi-dimensional treatment planning: I. Delineation of anatomy. Int J Radiat Oncol Biol Phys 9 : 777-787
245 7. Goitein M, Abrams M, Rowell D, Pollari H, Wiles J (1983) Multi-dimensional treatment planning: II. Beam's-eye view, back projection through CT sections. Int Radiat Oncol Biol Phys 9: 789-797 8. Castro J et al. (1989) Charged particle radiotherapy for lesions encircling the brain stem or spinal cord. Int J Radiat Oncol Biol Phys 17:477~484 9. Uric M, Sisterton J, Koehler AM, Goitein M, Zoesman J (1986) Proton beam penumbra: effects of separation between patient and beam modifying devices. Med Phys 13:734~741 10. Petti P (1991) Private communication 11. Smith AR, Purdy JA (eds) (1991) Three dimensional photon treatment planning. Pergamon Press, New York; also published as Int J Radiat Oncol Biol Phys 21 : 1-265
Radiation and Environmental Biophysics © Springer-Verlag 1992
3D treatment planning for heavy charged particles* Michel Goitein Massachusetts General Hospital, Department of Radiation Oncology, Harvard Medical School, Boston MA 02114, USA Received December 16, 1991 / Accepted February 17, 1992
Summary. The comments herein describe, at a necessarily superficial level, a number of issues which must be addressed in developing plans for heavy charged particle therapy. Programs now exist which provide the needed capabilities. The challenge now is to make the planning process easier and faster- and possibly more elective. It seems likely that this will be achieved in the next few years.
Introduction Heavy charged particles (heavier than electrons) have been used for radiation therapy for over three decades. However, it has only been since the advent of computed tomography (CT) in the mid 1970's, and the availability by the beginning of the 1980's affordable fast computers and interactive image and graphics displays, that is has been possible to plan and deliver charged particle radiation with confidence anywhere in the body. This is because, if particles are to have any advantage over photons, their energy must be tailored so that they do not exit the patient distally but, rather, are caused to stop within or at most very slightly beyond the target volume. For this to be achieved one must know where the target margins are and one must have a " m a p " of the patient's tissues to ascertain where particles of a given energy will come to rest. Until CT was available, these things could be known with confidence only in quite restricted anatomic sites (mainly the in the brain). Because the need was so acute, the technologies mentioned above were first taken advantage of by those engaged in heavy charged particle therapy - as soon as the technologies became available. However, it was immediately clear that the capabilities so developed had much broader application to conventional * Invited paper given on the fourth workshop on "Heavy Charged Particles in Biology and Medicine" GSI, Darmstadt, FRG, September 23-25, 1991
242 x-ray therapy. As a result, programs which use imaging information to develop 3D approaches to delivering radiation are now beginning to be used widely in radiation therapy and, no doubt, their development will feed back into, and enrich, particle therapy. The topic of planning treatment with heavy charged particles was addressed in 1982 by Goitein et al. [1] and much of the relevant literature up to that time was cited there. The following discussion does not duplicate that information, to which the reader is referred; only subsequent references are cited here. The requirements for planning therapy for those heavy charged particles which exhibit high LET behaviour are very similar to those for low LET charged particles (primarily protons) - but have additional and complex requirements relating to the usually non-uniform spatial (and even temporal) distribution of LET and related quantities. These additional issues are not further discussed here.
Compensation for inhomogeneities In treating a specified target volume the beam penetration must generally be tailored over the field area to account for three things: (I) the shape of the entrance (usually skin) surface; (2) the shape in depth of the distal surface of the target volume; and (3) the density distribution throughout the traversed tissues. In static beam treatments, this is generally done by designing a compensating bolus. As already alluded to, it is vital in most situations to have a measurement of the patient's tissue densities and then to take them into account in designing a compensating bolus. CT scan data, suitably scaled to represent effective density rather than linear x-ray attenuation coefficients, is ideal for this purpose [1]. The technique we use to fabricate boli in the M G H / H C L program has been described by Wagner et al. [3]; this approach or something quite similar to it seems to be widely used. In addition to their influence on the design of the beam penetration map, inhomogeneities also affect the beam modulation (the extent to which dose is spread out over depth). Inhomogeneities must be taken into account both for treatments employing fixed modulation, and for scanned beam approaches where the modulation can be varied over the beam cross section. In both cases, the modulation depth depends on the inhomogeneities within the target volume(s).
Registration of the patient with the treatment beam and other uncertainties There are number of uncertainties in beam delivery and dose estimation. Probably the foremost among these is the uncertainty in the registration between the treatment beam and the patient. Of course, this uncertainty exists in conventional x-ray therapy too and is handled primarily by expanding the treatment field margins. The problem is exacerbated in heavy charged particle therapy for two reasons: (I) because compensation for inhomogeneities relies on appropriate a registration between bolus and patient - a misregistration can give
243 rise to either hot or cold spots, or both [1]; and (2) the application of heavy charged particle therapy is predicted on taking advantage of the possibility of closely tailoring the treatment volume to the target volume - one therefore is desirous of minimizing the volume of tissue irradiated for non-medical indications. Techniques have therefore been developed [4] to assess quantitatively the uncertainties in treatment delivery, including those of misregistration, and of designing compensating boli taking into account misregistration and multiple scattering problems [5] - the smearing technique of [5] is the three-dimensional analogue of the field margin enlargement approach in x-ray therapy. These uncertainties must also be taken into account in designing the beam modulation. Definition of anatomy and design of beams
The approaches needed for the delineation of patient anatomy, both for normal tissues and target volumes, and for the design of treatment fields are very analogous to those needed in conventional x-ray therapy - or, considering the historical sequence in which 3D planning developed, one might express this v i s a versa. One needs: interactive drawing tools; automated feature extraction; margining tools; beam's eye view (BEV) assessment of anatomy and field direction; BEV design (both manual and automatic) of blocks and apertures; BEV design of compensating boli; sectional display of beams and beam modifying devices; etc. [6, 7]. In addition, now that they are readily available, one must be able to use MRI studies for these purposes and must be able to register CT and MRI studies with one another. In some ways BEV planning with heavy charged particles is easier than with x-rays because the resulting doses are more readily intuited. Rarely does an experienced planner need to see computed dose distributions to conceptualize the impact of a given beam on the plan. One aspect that is different and requires additional support is the use of patched fields to augment one another [8]. Here additional planning tools are needed, although none are yet available. Dose calculation
There are three broad approaches to computing the radiobiologically effective dose of heavy charged particles: (1) broad beam models; (2) pencil beam models; and (3) Monte Carlo approaches. These are distinguished by an ascending order of both accuracy and computational complexity, hence time. To date, broad beam models have been mostly used in routine clinical practice, with dose distributions as measured in cuboid water phantoms being modified by line-of-sight corrections for inhomogeneities and allowance for the distance from the effective source to the point of interest. While broad beam models have been valuable for practical clinical work, they have some important limitations. They do not model a number of the dose perturbations induced by finely structured inhomogeneities (although an error analysis [4] can generally ensure that the actual dose is properly bracketed by the uncertainty bounds placed on the calculated dose [5]); and they do not take proper account of the penumbra-blurring effect of bolus/patient air gaps [9].
244 The increased computing capabilities of modern affordable computers, such as workstations, are rapidly making it possible to implement more accurate algorithms. Pencil beam approaches, such as those used in the work described by Scheib and Pedroni below and that recently reported by Petti et al. [10], are likely to improve the accuracy of dose calculations - and hence potentially further reduce treatment volumes. Monte Carlo calculations have to date been confined to the exploration of generic (i.e. not specific to an individual patient) issues - primarily those having to do with compensation for inhomogeneities [1]. However, it seems likely that it will not be too long before it becomes practicable to perform Monte Carlo calculations for individual patient treatments.
Dose analysis: assessment and comparison of plans A discussion of 3D treatment planning would not be complete without mention of the problem of assessing a 3D treatment plan - and, more importantly, the problem of comparing two or more rival plans. As Urie describes in the paper below, heavy charged particle therapy has - both for the historical reasons already alluded to, and for some intrinsic reasons - played a leading role in this area. Advocates of heavy charged particle therapy have used comparative treatment planning approaches, in which heavy charged particle plans have been compared with plans using conventional x-rays, in order to assess objectively the possible advantages of heavy charged particles. Given that the plan assessment tools must in these situations be applied to both the heavy charged particle and x-ray plans, one can see that, almost by definition, the tools for plan assessment are not particle-specific (excepting only the assessment of high L E T effects). The tools in question have included: side-by-side presentation of dose distributions in sectional images; use of color-wash dose displays; dose-difference displays; dose-volume histograms; the presentation of dose statistics; and biophysical modelling of both tumor control probabilities and normal tissue complication probabilities [11]. As Urie's paper makes clear, these tools also make possible invaluable quantitative analysis of clinical experience.
Acknowledgements. Dr. Goitein's work was funded in part by a grant from the DHHS (CA 21239).
References 1. Goitein M, Abrams M, Gentry R, Urie M, Verhey L, Wagner M (1982) Planning treatment with heavy charged particles. Int J Radiat Oncol Biol Phys 8 :2065-2070 2. Steward VW, Koehler AM (1973) Proton beam radiography in tumor detection. Science 173:913-914 3. Wagner M (1982) Automated range compensation for proton therapy. Med Phys 9 : 749-752 4. Goitein M (1985) Calculation of the uncertainty in the dose delivered in radiation therapy. Med Phys 12: 608-612 5. Urie M, Goitein M, Wagner M (1984) Compensating for heterogeneities in proton radiation therapy. Phys Med Biol 29 : 553-566 6. Goitein M, Abrams M (1983) Multi-dimensional treatment planning: I. Delineation of anatomy. Int J Radiat Oncol Biol Phys 9 : 777-787
245 7. Goitein M, Abrams M, Rowell D, Pollari H, Wiles J (1983) Multi-dimensional treatment planning: II. Beam's-eye view, back projection through CT sections. Int Radiat Oncol Biol Phys 9: 789-797 8. Castro J et al. (1989) Charged particle radiotherapy for lesions encircling the brain stem or spinal cord. Int J Radiat Oncol Biol Phys 17:477~484 9. Uric M, Sisterton J, Koehler AM, Goitein M, Zoesman J (1986) Proton beam penumbra: effects of separation between patient and beam modifying devices. Med Phys 13:734~741 10. Petti P (1991) Private communication 11. Smith AR, Purdy JA (eds) (1991) Three dimensional photon treatment planning. Pergamon Press, New York; also published as Int J Radiat Oncol Biol Phys 21 : 1-265
