UvA-DARE is a service provided by the library of the University of Amsterdam (https://dare.uva.nl)
UvA-DARE (Digital Academic Repository)
Radiation induced lung damage
Seppenwoolde, Y.
Publication date
2002
Link to publication
Citation for published version (APA):
Seppenwoolde, Y. (2002). Radiation induced lung damage.
General rights
It is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s)
and/or copyright holder(s), other than for strictly personal, individual use, unless the work is under an open
content license (like Creative Commons).
Disclaimer/Complaints regulations
If you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please
let the Library know, stating your reasons. In case of a legitimate complaint, the Library will make the material
inaccessible and/or remove it from the website. Please Ask the Library: https://uba.uva.nl/en/contact, or a letter
to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam, The Netherlands. You
will be contacted as soon as possible.
Chapterr 6
Optimizingg radiation treatment plans for
lungg cancer using lung perfusion
information n
Yvettee Seppenwookie, Martijn Engelsman, Katrien De Jaeger, Sara H. Muller,
Paull Baas, Daniel L. McShan, Benedick A. Fraass, Marc L. Kessler,
Joséé S.A. Belderbos, Liesbeth J. Boersma, Joos V. Lebesque
Optimizingg radiation treatment plans for lung cancer
usingg lung perfusion information
Too study the impact of incorporation of lung perfusion information in the optimization of radicall radiotherapy (RT) treatment plans for patients with medically inoperable Non-Small Celtt Lung Cancer (NSCLC). the treatment plans for a virtual phantom and for five patients withh typical defects of pre-RT lung perfusion were optimized. Geometrically determined parameterss as the mean lung dose (MLD), the lung volume receiving more than 20 Gy (V20),, and thefurtctional equivalent of the MLD were minimized, using perfuston-weighted dose-volumee histograms. For me patients the (perfuston-weighted) optimised plans were comparedd to the ciintcaHy applied treatment r^ns. Tr» feastbiWy of perftision-weighted optimizationn was demonstrated in the phantom. Using perfusion information resulted in an increasee of the weights of those beams mat were directed through the hypo-perfused lung regionss bom for the phantom and for the studied patients. The automatically optimized dosee distributions were improved with respect to lung toxicity compared to the clinical treatmentt plans. For patients with one hypo-perfused hemMhorax, the estimated gain in post-RTT king perfusion was 6% of the prescribed dose compared to the geometrically optimizedd plan. For patients wtth smaller perfusion defects, perfusion-weighted optimizationn resulted in the same plan as the geometrically optimized plan. Perfusion-weightedd optimization resulted in clinically well applicable treatment plans, which cause lesss radiation damage to functioning lung for patients with large perfusion delects,
Introduction n
Forr patients with medicalfy inoperable or locally advanced non-small cell lung cancer (NSCLC),, local control remains poor after treatment with conventional radiotherapy doses (Schaaké-Koningg 1983) up to 66 Gy (Perez 1986, 1987, Pigott 1993, Saunders 1996, Dosoretzz 1996, Arriagada 1997). In the study of Martel et al. (1999), based on tumor control probabilityy model calculations, it was found that the dose required to achieve a better (50%) tumorr control at three years is probably in the order of 85 Gy. Because the incidence of grade
III radiation pneumonitis was found to be relatively low for prescribed standard doses up to 65 Gyy {Armstrong 1993), dose-escalation studies are ongoing in our and other institutions to increasee the tumor dose while keeping lung toxicity within certain limits (Armstrong 1997, Robertsonn 1997, BekJerbos 2000, Mehta 2001, Hayman 2001). In these studies two models aree used to estimate the incidence of radiation pneumonitis for a patient and to guide optimizationn of the dose-distribution. In these models, parameters as the mean fung dose (MLD,, (Kwa 1998a» or the volume of lung receiving more than a threshold dose (Marks 1997a,, Armstrong 1997, Graham 1999) of 20, 25 or 30 Gy are used to estimate the incidence off radiation pneumonitis. The shape of the accompanying normal tissue complication probabilityy (NTCP) curve is based on patient data of large (multi-center) studies (Kwa 1998a, Grahamm 1999).
Besidess the development of radiation pneumonitis, reduction in overall pulmonary function or lungg perfusion due to the treatment can be a complication as well. The amount of pulmonary functionn loss is especially important for patients with medically inoperable non-small-ceil lung cancerr who often have a reduced lung function before treatment because of chronic obstructivee pulmonary disease (COPD), intra-tboracic tumor or because they are heavy
Chapters Chapters
smokers.. The extent of damage to the lung due to these pre-existent diseases is not always
reflectedd in CT images. Single Photon Emission Computed Tomography (SPECT) lung
perfusionn scans provide additional information in three dimensions about local functionality of
lungg tissue and might give additional benefit to design the plan that minimizes the complication
riskrisk for perfusion damage for an individual patient (Marks 1993, 1995). The effect of
inhorriogeneouss dose distributions on lung perfusion can be predicted using a dose-effect
relationn for perfusion damage (Seppénwootde 2000). Changes in overall lung perfusion are
correlatedd with reduction in pulmonary function tests for patients with breast cancer and
malignantt lymphoma (Theuws 1998b, 2000). Recently, the same was found for lung cancer
patientss (Fan 2001). The group of Marks et al. (1999) suggested that the perfusion weighted
dose-volumee histogram (where the volume receiving a certain dose is weighted with the
averagee perfusion in that dose-region) could be a valuable tool in designing the optimal RT
plan. .
Inn this paper we investigated whether the additional information obtained by SPECT lung
perfusionn scans in the treatment planning process resulted in better treatment plans for
patientss with inoperable non-small cell lung cancer.
Methodss and materials
Perfusion-weightedd optimization was first simulated on a phantom as a proof of principle and
thenn applied to a number of representative lung cancer patients.
Too apply perfusion-weighted optimization, the automatic beam weight optimization method of
thee University of Michigan was used. With this method, a set of randomly distributed points
wass generated for both the PTV, lung (CT-defined lung volume minus the GTV) and spinal
cord.. The dose per beam was calculated in each point To allow perfusion-weighted
optimization,, the dose in each of the randomly distributed points in the lung was weighted with
thee normalized pre-treatment perfusion of that point (see Appendix). Fast simulated annealing
(Magerass 1993, Oldham 1995) was used to optimize the weights of the beams by minimizing
user-definedd cost functions (costiets). These costlets were composed for each relevant organ
(targett and organs at risk). The costlets can be defined as a function of an «valuator, such as
aa point in a dose-volume histogram or a dose parameter (e.g. the mean lung dose). The total
costt of a certain dose distribution is then computed by summation of the costiets. The details
off the algorithm are published elsewhere (Kim 1995, Kassier 1998). The constraints used to
optimizee the treatment plans for the PTV were: a minimum dose of 95%, a maximum of 107%
andd an average of 100% of the prescribed dose (according to the ICRU (ICRU Report 50
1993)).. When violating the constraints a proportional cost was applied. For the spinal cord the
appliedd dose was not allowed to be higher than 50 Gy.
Forr a phantom and five patients, beam weights were optimized by (individually) minimizing
differentt lung parameters besides the constraints for the PTV and the spinal cord. These lung
parameterss were the mean lung dose (MLD), the relative volume of lung receiving more than
200 Gy (V20, a DVH-point) and the mean perfusion-weighted lung dose (MpLD) using
perfusion-weightedd dose-volume histograms. The MpLD is a measure for the perfusion
damagee when the local dose-effect relation for perfusion changes is linear, while the Vp20 is
ann approach for a step-tike local dose-effect relation (see Appendix). The constraints used to
optimizee thé treatment plans for normal tissue damage were an M(p)LD of 0 Gy or a V(p)20 of
0%.. When violating these constraints a proportional cost was applied, so that the
perfusion-Forr the patiënte the Vp20 could not be calculated in the beam optimization program.
Perfusion-weightedPerfusion-weighted optimization
Thee optimized beam weights for the different plans were re-entered into the U-MPIan treatmentt planning system (University of Michigan, (Fraass 1987b)) and the full 3D dose distributionn of each optimized plan and the averaged remaining perfusion after treatment was calculated.. For the phantom the dose distributions and the dose-volume parameters were comparedd to the treatment plan with the highest conformity.
Phantom m
AA virtual phantom was constructed in the treatment planning system. The phantom consisted off three concentric cylinders: one inner cylinder with a radius of 2.5 cm consisting of unit densityy material representing a lung tumor, one with a radius of 8 cm, density 0.3 g/cm3, representingg lung tissue and an outer cylinder with a radius of 10 cm of unit density representingg the patient's body contour (Figure 1A).
functioningg . \ igg \ \ :: if ymÊ \G4—I \p=0.3glcm\p=0.3glcmii[l[l GTV | | II WP=I g/cm3// \\ Functioning "" * * * 5
<-A <-A
FigureFigure 1. A. Cross section of the circular phantom consisting of three concentric cylinders: one with a radius of 2.5 cm
consistingconsisting of unit density material representing the tumor, one with a radius of 8 cm, density 0.3 g/cm3, representing lunglung and one with a radius of 10 cm, unit density representing the patient contour. The PTV is a 0.5 cm expansion of thethe GTV. The right upper quarter of the lung has no perfusion (function = 0, dark grey area), while the other part of the lunglung is functioning 100%. B. Beam setup consisting of 7 coplanar 8 MV photon beams at equally spaced angles. Sevenn beam directions (8 MV) at equally spaced angles were set up around the tumor. The dosee to regions at risk could be substantially reduced using seven ports of beam incidence. Increasingg the number of ports beyond seven produced only minor further gain (Sauer 1999). Thee seven beams consisted of one open beam and two wedged beams with opposite wedge directionss in the plane of planning. The gross tumor volume (GTV) plus 0.5 cm margin yielded thee planning target volume (PTV). The margin between PTV and field edge was adapted in suchh a way that the 95% isodose line was fitted as close as possible around the PTV (Figures 1BB and 2A). A hypo-perfused region was constructed representing 29% of the total lung volumee (shaded area in Figure 1A). The hypo-perfused region in the phantom was comparable too the patterns of hypo-perfusion seen in the patients of group 3 and 4 (See Results section).
Patients s
Too investigate the value of perfusion-weighted optimization for real treatment plans, we performedd a planning study using chamfer matched CT and normalized SPECT lung perfusion scanss (Seppenwoolde 2000) of patients with non-small cell lung cancer. The lung perfusion
ChapterChapter 6
patternss of 116 patients from the Netherlands Cancer Institute could retrospectively be divided intoo six groups (See Results section). The subdivision in groups was based on the overall appearancee of their perfusion pattern; the localization and size of the perfusion defect. Left and right-mirroredd perfusion defects were put in the same group.
FigunFigun 2. A. Dose distribution of the plan with the highest conformity to the PTV. All beams have the same weight (no
wedges);wedges); 100, 95, 50 and 29% (20 Gy) iso-dose lines are represented (the prescribed dose is 70 Gy). B. Dose distributiondistribution optimized for the MpLD. C. Dose distributbn optimized for the V20. D. Dose distribution optimized for the Vp20. Vp20.
Too exclude the effect of automatic optimization compared to manual treatment planning, the clinicallyy applied treatment plan of a representative patient from each group was first optimized onn geometry (based on the MLD or V20) and then compared to perfusion-weighted optimized planss (based on the MpLD), with and without varying the number of beams and beam incidencee directions. Next to the clinically used beams, one or more extra beams were set up throughh hypo-perfused lung areas or at more or less equidistant angles. The prescription dose wass 70 Gy, according to the treatment planning protocol that was used at the time these patientss were treated. The beams were shaped using a multi-leaf collimator to conform field shapee to the PTV in the beam's eye view. All beams consisted of one open beam and two wedgedd beams with opposite wedge directions in the plane of planning. The isocenter, chosen nearr the center of the PTV is the ICRU reference point, which receives the prescribed dose. Forr the patients the dose constraint for the spinal cord of 50 Gy was included in the optimizationn procedure. The treatment plans created with (perfusion-weighted) optimization weree calculated retrospectively in this study. All patients were treated with conventional plans.
Dosee calculation
CT-basedd dose calculations were performed as described previously (Boersma 1994), using aa 3D treatment planning system (U-MPIan) in which the clinically applied Octree/Edge model withh tissue inhomogeneity correction (equivalent path length algorithm) is incorporated.
(Perfusion-weighted)) dose per beam
Forr the phantom and all tested patients, the MLD and MpLD per single beam were calculated. Thesee values represent the geometrical and functional usefulness of using that beam for the treatmentt plan: when the MLD of one beam is low compared to others, this beam irradiates a relativelyy small lung volume. The values for the MLD per beam are determined by the geometryy of the lungs. When the MpLD of one beam is low compared to others, this beam irradiatess a relatively small perfused lung volume. The values for the MpLD per beam are determinedd both by the geometry of the lungs and the perfusion distribution of the particular
Perfusion-weightedPerfusion-weighted optimization
patient:: when a large amount of lung is irradiated by one particular beam, but most of the lung tissuee in the beam's-eye-view of that beam is not perfused, the value for the MpLD will be low. Thee ratio of the MpLD and the MLD per beam indicates the gain for that beam of using perfusionn information and is independent of the lung geometry. If the ratio is small, the extra perfusionn information is useful, when the ratio is larger, the beam irradiates well-perfused lung andd thus is not favorable. When the ratios for all beams are comparable, the perfusion-weightedd optimization will yield similar results to the non-weighted optimization. When the ratio iss small for one or more beams, the perfusion-weighted optimization will prefer to use these beamss to create an optimal plan. However, the use of less favorable beams can sometimes bee necessary to create a homogeneous dose distribution in the target.
Results s
Ass a proof of principle the method of (perfusion-weighted) optimization was first applied to the cylindricall phantom (Figure 1). The integral dose will essentially be the same for all beam arrangementss for the non-weighted case since the problem is totally symmetric. One of the
140 140 120 0 100 0 «0 0 6 0 0 4 0 0 A.. Phantom o o 1 1
1 1
2 2ê ê
3 3* *
4 4£ £
5 5 è è 6 66 6
7 7 140 0 120 0 100 0 *>1 1 D.. Group 3 33 4 5ff 2 2 - 2 . 2
MpLD/MLD OO MLD (%) AA MpLD (%) 500 100 150 200 250 300 350 500 100 150 200 250 120 0 100 0 80 0 6 0 0 2 0 0 B.. Group 1 »» 11 2** *
3 3 k k 4 4 A A 5 5 Ë Ë 6 6£ £
E.. Group 4 22 3 4 AAAA A 55 622 a
SO,, 100 150 200 250 300 350 500 100 150 200 250 2 0 0 00 0 80 0 6 0 0 40 0 20 0 C.. Group 2 ft ft 1 1 2 2 3 3 ** A 44 5 ft ft 6 6Beamm angle (degrees)
Groupp 5
500 100 150 200 250
Beamm angle (degrees)
FigureFigure 3. The mean (perfusbn weighted) lung dose per beam, (relative to the prescription dose) and the ratio between
ChapterChapter 6
solutionss was a plan with all beam weights equal to 1 (Figure 2A). When the MLD per single beamm (normalized on 100% in the isocenter) was calculated (Figure 3A), the values for the MLDD per beam are equal, another way to represent that the choice of beam weight combinationss is arbitrary in a symmetrical phantom.
Minimizingg the V20 resulted in an approximation of a plan-parallel irradiation, because the geometryy of the 7 beams did not allow a pure parallel beam setup. In this dose-distribution the V200 is as small as possible (Figure 2C). The MLD for this plan is about the same as any of the planss optimized on the MLD. Because of the symmetry of the phantom, rotation of the dose distributionn resulted in the same V20.
Whenn perfusion information was considered during optimization, the choice of a certain dose distributionn became better determined. The MpLD and the Vp20 were minimal in the configurationss as shown in Figure 2B and D, respectively. In both cases, the beams with the highestt weights were directed through the lung region that had no perfusion. The values for thee MpLD per single beam and the ratio between the MpLD and MLD varied with the amount off non-perfused lung in the beam's-eye-view (Figure 3A). Beams 1 and 2 had the lowest ratio andd were therefore chosen as the optimal combination of beams. Opposing beam nr. 5 ensuredd a better PTV coverage and had also a low ratio. The other beams with a much smaller contributionn to the absolute dose were necessary to ensure a more homogeneous PTV dose. Forr a smaller tumor (2.5 cm diameter) similar beam weights were obtained. However, the resultss will differ when the relative size of the GTV to the lung decreases further. Then, the best plann for the V20 will approach an arc when the number of beams is larger than the ratio of the prescriptionn dose and the threshold dose (in this case 20 Gy), provided that there is minimal overlapp between the entrance beams.
Patients s
Inn the cylindrical phantom only the GTV and lung volumes were considered. To apply the optimizationn method in more realistic situations, perfusion-weighted optimization was applied too representative NSCLC-patients with different kinds of perfusion defects. The 116 patients couldd be easily divided into groups according to Figure 4, based on the size and appearance off their perfusion pattern:
FigureFigure 4. Each patient is assigned to a group according to the characteristics of the perfusion pattern. The white
ellipsesellipses represent the tumor location.
Groupp 1: hypo-perfusion only at the site of the tumor (44% of the patients). Groupp 2: hypo-perfusion adjacent to the tumor (5% of the patients). Groupp 3: hypo-perfusion ventral of the tumor (13% of the patients). Groupp 4: hypo-perfusion dorsal of the tumor (9% of the patients).
Groupp 5: hypo-perfusion of the entire ipsi-lateral lung (16% of the patients).
Groupp 6: miscellaneous (13% of the patients), including patients with bullae and emphysema. Thee patterns of hypo-perfusion were inhomogeneous for these patients.
Perfusbn-weigbtedPerfusbn-weigbted optimization
Fromm group 1 to 5, one representative patient who had an average sized tumor was chosen
forr perfuston-weighted optimization (Figure 5-9). For every representative patient, the clinically
usedd beams (in the C panels of Figure 5-9), plus one to four additional beams were set up
aroundd tine tumor for the optimization. The resulting treatment plans will be discussed below
a.. for the M(p)LD and b. for tire V20. In all these plans the constraints for tumor coverage and
spinall cord were met. For all patients, except for patient 3, the automatically optimized dose
distributionss were improved with respect to lung toxicity compared to the clinical treatment
plans.. However, optimization on one parameter did not necessarily optimize the other
parameters,, for example in group 1 the plan optimized for the MpLD has a higher V20 than
plann optimized on the MLD (Table 1).
Tablee 1. Dosé-volume parameters for S patients (in % of the prescribed dose for the MLD and MpLDD and in % of the lung volume for the V20) for the plans optimized for the different parameters.
Groupp 1 MLD(%) ) MpLDD (%) V2QQ (%) Groupp 2 MLDD (%) MpLDD (%) V200 (%) Groupp 3 MLDD (%) MpLDD (%) V200 (%) Groupp 4 MLDD (%) MpLDD (%) V200 (%) Groupp 5 MLDD (%) MpLDD (%) V200 (%) Clinicall plan 22 2 21 1 19 9 21 1 23.5 5 21.5 5 11 1 5.5 5 9.5 5 27 7 15 5 33 3 28.5 5 23.5 5 38.5 5 MLD D 16.5 5 17 7 19 9 19.5 5 19 9 18 8 10.5 5 5.5 5 9.5 5 25 5 13.5 5 30 0 26 6 19 9 37.5 5 Optimizedd for MpLD D 16,5 5 17 7 20 0 19.5 5 19 9 18 8 10.5 5 5.5 5 9.5 5 25.5 5 13 3 30.5 5 26 6 13 3 21.5 5 V20 0 19 9 19 9 16.5 5 20.5 5 20.5 5 18 8 11 1 5.5 5 9.5 5 28 8 15.5 5 30 0 26 6 13 3 21.5 5
85 5
ChapterChapter 6
Groupp 1
a.. For the patient of group 1 (Figure 5, Table 1), there was, as expected, almost no difference betweenn perfusion-weighted and MLD-optimization. Because the hypo-perfused region is locatedd only at the tumor site, and the mean lung dose and the mean perfusion-weighted lung dosee are calculated based on the lung volume minus the gross tumor volume, the MLD and thee MpLD are equal. This is reflected in Figure 3B as well; the ratio of the MpLD and the MLD perr single beam did not vary more than 10%. Only beam 6 had a small ratio, but for this beam thee absolute values for the MLD and MpLD were larger than for the other beams, so that in thee optimization this beam was omitted.
b.. Optimizing the V20 resulted in a treatment plan that was more parallel opposed and less conformall (see the 95% isodose line at the arrow in Figure 5F) than the plan in Figure 5D.
FigureFigure 5. A. The beam setup for a patient of group 1. B. The pre-RT perfusion pattern of this patient, hypo-perfusion is present
onlyonly at the site of the tumor. C. The 3 field clinically applied treatment plan, the 20,35,50,66.5 (8596) and 70 Gy (100%) isodose lineslines are shown in solid, dashed, solid, dotted, and solid lines, respectively. The PTV and GTV are delineated in solid lines. In all figuresfigures the beam numbers of the beams that were used for the plans are indicated (for beams with a relatively small weight, the numbersnumbers are indicated in a smaller font). D. Treatment plan optimized for the MLD. E. Including perfusion information (optimized forfor the MpLD) yielded a dose distribution very similar to plan D. F. Treatment plan optimized for the V20. The arrow indicates thethe lower conformity in the V20 optimized plan.
Groupp 2
a.. For the patient of group 2 (Figure 6, Table 1), the beam incidence direction of beams 4 and 5,, directed through hypo-perfused lung, seemed to be smart at first sight but attributed to damagee to the healthy contra-lateral lung. The ratio between the MpLD and the MLD (Figure 3C)) for beams 4 and 5 was much higher than for beams 1,2,3 and 6. The low variation in the ratioo between the MpLD and the MLD for these beams can explain why for this patient there wass very little difference between the MLD-optimized and perfusion-weighted optimized treatmentt plan.
Perfusion-weightedPerfusion-weighted optimization
b.. Using the V20 as the optimization parameter, a little different dose distribution was obtained. Thee V20 of this plan was equal to that of the plan optimized for the MLD, however, the values forr the MLD and MpLD are a little higher.
FigureFigure 6. A. The beam set-up for a patient of group 2. B. The pre-RTperfusion pattern of this patient, hypo-perfusion is present
atat the site of and adjacent to the tumor. C. The clinically applied treatment plan with five fields. Note the relative high dose in thethe contra-lateral lung at the arrow that is reduced in the other optimized plans. D. A treatment plan optimized for the MLD. IncludingIncluding perfusion information did not change the dose distribution. E. A treatment plan optimized for the V20. Including perfusionperfusion information did not change the dose distribution.
Groupp 3
a.b.. Optimization on the MLD, the MpLD and V20 of a 6-field plan (Figure 7, Table 1) resulted inn three plans that were almost identical to the clinical plan. Beams nr 4 and 6 had a low ratio betweenn the MpLD and the MLD (Figure 3D), but also a low value for the MLD and the MpLD byy itself, meaning that these beams were both geometrically as functionally advantageous (passedd a small volume of (perfused) lung tissue). Beam nr. 2 was used to ensure adequate tumorr coverage. Although the 4 plans had a slightly different dose-distribution, they were identicall with regard to MLD, V20, MpLD, Vp20 and tumor coverage.
Groupp 4
a.. The high dose area was transferred a little into the hypo-perfused region (Figure 8E, arrow), whenn perfusion information was included in the optimization. Beams 1,2 and 5 have a lower ratioo between the MpLD and the MLD per single beam than the other beams (Figure 3E). For thiss patient especially beam 5 was already preferable in the non-weighted optimization and becamee even more preferable when the perfusion information was included. This beam got a higherr weight in the MpLD optimized treatment plan.
b.. Optimizing on the V20 and Vp20 yielded a different dose distribution: note the 50% isodose linee that encompasses a larger area than in the other dose distributions (Figure 8F, arrow).
ChapterChapter 6
FigureFigure 7. A. The beam setup for a patient of group 3. B. The pre-RTperfusion pattern of this patient, hypo-perfusion is present
atat the site and ventrally of the tumor. C. The clinically applied treatment plan with three fields. D. Treatment plan optimized for thethe MLD. E. Treatment plan optimized for the MpLD. F. Treatment plan optimized for the V20. Including perfusion information diddid not change the dose distribution. Although the dose distributions are a little different, the lung volume parameters as the M(p)LDM(p)LD and V20 were not different for all four plans.
FigureFigure 8. A. The beam setup for a patient of group 4. B. The perfusion pattern of this patient, hypo-perfusion is present at the
sitesite and dorsal of the tumor. C. The 3-field clinically applied treatment plan. D. Treatment plan optimized for the MLD. Note the higherhigher conformity to the PTV. E. Treatment plan optimized for the MpLD. The arrow indicates the higher dose in poorly perfused lung.lung. F. Treatment plan optimized for the V20. The 20 Gy isodose line is similar to that of plan E, however, the 50% isodose line (arrow)(arrow) encompasses a larger volume.
Perfusion-weightedPerfusion-weighted optimization
Groupp 5
a.. Although beam 5 had a high MLD compared to the other beams, in the optimization for the MLDD this beam was chosen to ensure a homogeneous irradiation of the PTV. When the 3D perfusionn information of this patient was included, this beam was omitted because this beam wouldd irradiate too much functional lung. Only the 3 beams with the lowest MpLD/MLD ratio remainedd (plan E).
b.. Minimizing V20 yielded plan E. The V20 of plan D is 37.5%, compared to 21.5% in plan E. Thee relatively higher dose in lung tissue anterior of the tumor in plan E might intuitively appear unfavorablee because it is not conformal to the PTV, but these high local doses do not contributee to a higher MLD, compared to plan D; both dose distributions resulted in almost the samee MLD (26% of the prescribed dose). The maximum dose in the spinal cord remained beloww 50 Gy by using beam 4 in plan E.
Plann E resulted in the combination of the lowest MLD, the lowest V20 and the lowest MpLD withh adequate tumor coverage and no violation of the spinal cord constraint for this particular patient. .
FigureFigure 9. A. The beam setup for a patient of group 5. B. The pre-RT perfusion distribution of this patient, almost the entire right
lunglung is hypo-perfused. C. The clinically applied 4-field treatment plan. D. Treatment plan optimized for the MLD. E. Treatment planplan optimized for the MpLD or V20, these parameters yielded the same dose-distribution.
Groupp 6
Becausee the perfusion patterns of these patients were very diverse and complex, it was not possiblee to select one representative patient. Due to the inhomogeneity of the perfusion patternn of these patients, it was very difficult to select the smartest beam incidence direction priorr to automatic optimization using the current methods.
ChapterChapter 6
Discussion n
Thee feasibility of using perfusion information for optimization of radiotherapy treatment plans forr patients with Non-Small Cell Lung Cancer (NSCLC) was studied in a phantom and for patientss with different types of perfusion defects. Including perfusion information did not yield differentt treatment plans for patients with small perfusion defects. Only for the patient with one hypo-perfusedd hemi-thorax, the perfusion-weighted optimization made a difference of 6% on thee remaining lung perfusion after treatment. For this patient, the effect of plans D and E on thee post-RT perfusion is visualized in Figure 10, based on two different dose-effect relations (aa linear and a step dose-effect relation). The plan optimized for the MLD (plan D) resulted in aa perfusion pattern with gradual perfusion damage (Figure 10A) when the linear dose-effect relationn was used and in sharp edged perfusion damage when a step dose effect relation was usedd (Figure 10B). When plan E was applied on the pre-RT perfusion, both dose-effect relationss did not lead to visible perfusion damage (Figure 10C). Clinical data about the local dose-effectt relation for perfusion damage is available (Garipagaoglu 1999, Seppenwoolde 2000).. These data show a linear dose-effect relation rather than a step-function and therefore thee MpLD appears to be the best predictor for what happens after radiotherapy.
Linear r
. . . - • • " • " ' '
Linearr & step
Predicted d postt -RT
J J
MLDD optimized plan (D) MpLD & V20 optimized Planss (E)
FigureFigure 10. Predicted perfusion distributions after treatment, based on the dose distributions of plan 9D and 9E, with differentdifferent dose-effect relations.
Clinicall relevance and patient selection
Whenn the dose distributions for the five representative patients were automatically optimized, itt appeared that the resulting treatment plans were improved compared to the clinical treatmentt plan, regardless of the optimization parameter. Perfusion-weighted optimization
PetfusJon-weightedPetfusJon-weighted optimization
however,, attributed to a small extra improvement compared to the geometrical optimization
(MLDD and V20). To select patients with one hypo-perfused lung that benefit most from the 3D
perfuston-weightedd optimization, planar V/Q images, made before applying the technique, can
help.. Furthermore, based on the following factors, the usefulness of perfusion-weighted
optimizationn for an individual patient can be estimated:
Geometricall considerations
Laterall beams directed through hypo-perfused lung adjacent to the tumor usually exit in
contra-laterall (well-perfused) lung. To take maximal advantage of the perfusion-weighted
optimization,, beams mat only pass through ipsMaterai (hypo-perfused) lung are preferred, for
examplee in a parallel-opposed setup. However, these beams often deliver their dose to the
spinall cord where the maximal prescribed dose is restricted to 50 Gy (2 Gy/fraction). One has
alsoo to consider organs at risk such as the heart and esophagus, which further limit the beam
incidencee directions that can be chosen. For central tumors, the beam direction with the
shortestt path through lung is preferred in the optimization because these beams will result in
thee lowest mean lung dose. This is fn general just the region where hypo-perfusion is situated,
ass hypo-perfusion often is located distally from the (central) tumor (Marks 1993,
Seppenwoofdee 2000). For more distally located tumors, the region of hypo-perfusion is often
veryy small and will not contribute significantly in the optimization.
Tumorr size
Largee tumors require large treatment fields and thus the damage to (perfused) lung tissue will
alsoo be large. In these cases perfusion-weighted optimization will be important in terms of
functionall outcome for the patient especially because the pre-RT lung function that is reflected
inn the perfusion is also more likely to be reduced for patients with large tumors. When the
tumorr is smalt, both the tumor and the irradiation affect only a small part of the lungs and
thereforee the improvement that can be gained by perfusion-weighted optimization is of less
significancee for the patient
Dosee calculation
Thee Octree/Edge dose calculation algorithm incorporated in the treatment planning system
usedd in this study does not take lateral scatter adequately into account and therefore in lung
tissuee the dose is not calculated correctly. This algorithm underestimates beam penumbra in
lung;; film measurements show a flattening of the beam penumbra (Engelsman 2001b). This
meanss that in the penumbra region where more than 50% of the dose is planned, a lower dose
iss delivered than predicted by the treatment planning system. Because the actual dose is lower
inn the high dose regions and on the edge of the PTV, larger fields are necessary to ensure
adequatee tumor coverage. Due to these larger fields the mean functional lung dose will rise
andd thus the gain of perfusion-weighted optimization may be of more importance for the
patientt Only recently, algorithms became available that accurately predict dose distributions
inn inhomogeneous media like lung tissue. The effect of the underestimation of the beam
penumbraa can lead to different dose distributions in the automatic optimization. This is the
subjectt of future studies. However, the effect of the miscalculated dose on the difference
betweenn the non-weighted and perfusion-weighted dose distributions wilt be of second order.
Pre-RTT perfusion and reperfusion
Especiallyy for patients with an impaired lung perfusion before treatment it is important to
considerr perfusion damage during optimization of the treatment plan.
Reperfusion.. which can be caused by tumor-regression, might reduce the adverse effect of the
irradiationn (Seppenwoolde 2000). Reperfusion occurred even in regions where a high dose
91 1
Chapters Chapters
wass given (SeppenwookJe 2000). In the studied patient group however, the measured
reperfusionn was not followed by improvement in lung function as measured with classical lung
functionn tests [K. de Jaeger, personal communication]. Therefore, reperfusion was not
consideredd in the perfusion-weighted optimization.
MLDVS.V20 0
Optimizationn on the MLO or the V20 yielded different dose distributions, resulting in a more
parallel-opposedd treatment plan for the V20 constraint with a dose distribution that was less
conformall to the PTV man the MLD-optimized plan.
Inn general, two different NTCP models are being used: one uses the threshold dose parameter
(Grahamm 1999) while the other adheres to the MLD for the estimation of the incidence of
radiationn pneumonitis (Armstrong 1997). The methods differ in the 'inferred' underlying
dose-effectt relation between locally absorbed dose and the development of local lung damage
resultingg in radiation pneumonitis (Seppenwoolde 2001). In the threshold dose model a step
locall dose-effect relation is assumed while thé MLD model assumes a linear relation (Kwa
1998c). .
Too illustrate the effect of automatic and perfusion weighted optimization for all the presented
treatmentt plans, the risk on radiation pneumonitis (grade 2 or higher) is estimated, based on
bothh NTCP models (Table 2). In all the optimized treatment plans the NTCP for radiation
pneumonitiss reduced, regardless of the optimization parameter, compared to the clinical plans
thatt were designed without automatic optimization. The differences in the NTCP values were
smallerr between the automatically optimized plans. The most spectacular improvement in the
NTCPP for radiation pneumonitis based on the V2Ö model is seen in the patient of group S for
plann E, compared to plan C and D. However, the resulting treatment plan has a lower
conformity. .
Tablee 2. The estimated risk for radiation pneumonitis (grade £ 2) based on two NTCP models
Groupp 1 Groupp 2 Groupp 3 Groupp 4 Groupp 5 Model l 1 1 2 2 1 1 2 2 1 1 2 2 1 1 2 2 1 1 2 2 Clinicall plan 4.9 9 2.5 5 4.2 2 3,5 5 0.6 6 0.6 6 10.2 2 13.2 2 12.4 4 21.7 7 MLD D 2.0 0 2.5 5 3.3 3 2.1 1 0.6 6 0.6 6 7.8 8 9.7 7 8.9 9 20.0 0 Optimizedd for: MpLD D 2.0 0 2.8 8 3.3 3 2.1 1 0.6 6 0.6 6 8.3 3 10.2 2 8.8 8 3.5 5 V20 0 3.0 0 1.7 7 3.9 9 2.1 1 0.6 6 0.6 6 11.7 7 9.7 7 8.9 9 3.5 5 Modell IrtheMLD-nxHJelwtthaTDsoOf 30.5 and an m of 0.30 (Kwa 1998a).
Modell 2: the V20-model with a V20M of 51.3 and an m of 0.32 (Graham 1999).
Too be able to discriminate between the two NTCP models, patient data are needed in which
theree is as little as possible correlation between V20 and the MLD. However, for the 116
NSCLC-patientss in this study a high correlation is found (r
2= 0.9). To collect more clinical data
PeifusJon-migtrtddPeifusJon-migtrtdd optimization
forr a reliable discrimination between the V20 and the MLD method, for example half of the numberr of patients with the same tumor location should be treated with plans optimized on the MLDD and the other half with plans optimized on the V20.
Forr the perfusion-weighted equivalents of the MLD and V20, clinical data about the local dose-effectt relation for perfusion damage is available (Garipagaoglu 1999, Seppenwoolde 2000). Thesee data show a linear dose-effect relation rather than a step-function.
Conclusion n
Onlyy for patiënte with a large pre-treatment perfusion defect, perfusion-weighted optimization resultedd in clinically well applicable treatment plans, which may have caused less radiation damagee to functioning lung, compared to treatment plans that were optimized on the mean lungg dose and a homogeneous target dose alone. For patiënte with small perfusion defects, perfusion-weightedd optimization yielded a treatment plan almost equal to the non-perfusion-weightedd optimized plan. Because the gain of perfusion-weighted optimization depended on thee pre-RT perfusion pattern, planar V/Q scans can be used to select patients mat benefit from 3DD perfusion-weighted optimization.
Chapters Chapters
Appendix x
Too be abte to include functional information in the optimization, two steps have to be taken: the
firstfirst is to define a parameter, which is predictive for the functional outcome of the treatment
(usingg for example a local dose-effect relation for changes in lung perfusion). The second is to
findd a way to incorporate the functional information into the optimization module.
Thee mean lung dose and the mean perfusion-wetghted lung dose of a certain treatment plan
cann be calculated by
M L DD
= iZ
D
.
and
M P L D - Ü C . - D ,
wheree N is the number of voxels and D
nis the local dose in voxel n. C
nis the local number of
perfusionn counts before treatment in voxel n, representing local functionality, normalized using:
cc =
Cts
-Wheree Cts„ is the absolute number of SPECT counts in voxel n.
Inn each voxel n we could predict the remaining perfusion at 3 months after treatment by
calculating: :
PaC**** = Perf^ {l - E(DJ }
wheree E(D
n) is a local dose-effect relation for perfusion damage and:
Perff"^^ %
iss the local perfusion normalized on the well perfused (WP) lung region that has a perfusion
betterr than 60% of the maximum perfusion in the lungs. The average perfusion homogeneity
beforee treatment (PHpre) is the average perfusion throughout the lungs:
PHpree Ü Perf ** = — X * « C
Thee average perfusion after treatment is calculated by
r r
PHpredictedpostt = PHpre
PerfusJon-mightedPerfusJon-mighted optimization
Too ensure the maximum possible lung function for a patient after radiotherapy, this value
shouldd be maximized by varying the number of beams, the beam incidence directions and the
beamm weights. We used two different local dose-effect relations that represent extreme shapes
off a dose-effect relation for perfusion damage:
1.. E ( D J = cD
11whichh is a linear dose-effect relation. With this dose-effect relation the remaining perfusion
afterr treatment can be calculated by:
PHpredictedpostt = PHpre$ - c • MpLD}
whichh is a step dose-effect relation with D*, a threshold dose of, for example, 20 Gy. For local
dosess below the threshold dose there is no perfusion damage and for doses higher than the
thresholdd dose, the local perfusion damage is 100%.
Thee perfusion-weighted equivalent for a threshold dose (DgO of 20 Gy is:
V
P
200
= è£
C
r-®(D
I
-D-n.) =
2
^
E a
^^ **=1 • tot