- Email: [email protected]
Evaluation of margining algorithms in commercial treatment planning systems
Radiotherapy and Oncology 86 (2008) 43–47 www.thegreenjournal.com
Treatment planning
Evaluation of margining algorithms in commercial treatment planning systems Alistair M. Poolera,*, Helen M. Maylesa, Olivia F. Naismithb, John P. Sagec,1 David P. Dearnaleyd, on behalf of CHHIP collaborators a
Physics Department, Clatterbridge Centre for Oncology NHS Foundation Trust, Wirral, UK, bDepartment of Physics, The Royal Marsden NHS Foundation Trust, London, UK, cDepartment of Physics, Arden Cancer Centre, Walsgrave Hospital, Coventry, UK, dDepartment of Academic Radiotherapy, Institute of Cancer Research and Royal Marsden NHS Foundation Trust, Surrey, UK
Abstract Introduction: During commissioning of the Pinnacle (Philips) treatment planning system (TPS) the margining algorithm was investigated and was found to produce larger PTVs than Plato (Nucletron) for identical GTVs. Subsequent comparison of PTV volumes resulting from the QA outlining exercise for the CHHIP (Conventional or Hypofractionated High Dose IMRT for Prostate Ca.) trial confirmed that there were differences in TPS’s margining algorithms. Margining and the clinical impact of the different PTVs in seven different planning and virtual simulation systems (Pinnacle, Plato, Prosoma (MedCom), Eclipse (7.3 and 7.5) (Varian), MasterPlan (Nucletron), Xio (CMS) and Advantage Windows (AW) (GE)) is investigated, and a simple test for 3D margining consistency is proposed. Methods: Using each TPS, two different sets of prostate GTVs on 2.5 mm and 5 mm slices were margined according to the CHHIP protocol to produce PTV3 (prostate + 5 mm/0 mm post), PTV2 (PTV3 + 5 mm) and PTV1 (prostate and seminal vesicles + 10 mm). GTVs and PTVs were imported into Pinnacle for volume calculation. DVHs for 5 mm slice plans, created using the smallest PTVs, were recalculated on the largest PTV dataset and vice versa. Since adding a margin of 50 mm to a structure should give the same result as adding five margins of 10 mm, this was tested for each TPS (consistency test) using an octahedron as the GTV and CT datasets with 2.5 mm and 5 mm slices. Results: The CHHIP PTV3 and PTV1 volumes had a standard deviation, across the seven systems, of 5% and PTV2 (margined twice) 9%, on the 5 mm slices. For 2.5 mm slices the standard deviations were 4% and 6%. The ratio of the Pinnacle and the Eclipse 7.3 PTV2 volumes was 1.25. Rectal doses were significantly increased when encompassing Pinnacle PTVs (V50 = 42.8%), compared to Eclipse 7.3 PTVs (V50 = 36.4%). Conversely, fields that adequately treated an Eclipse 7.3 PTV2 were inadequate for a Pinnacle PTV2. AW and Plato PTV volumes were the most consistent (0.3%) and ( 0.4%). However, the 1 · 50 mm margin in Pinnacle produced a 15.9% larger volume than 5 · 10 mm margins, while for Eclipse 7.3 the single margined volume was 14.3% smaller. These inconsistencies were reduced to 5% by adjusting the superior/inferior margins. Conclusions: Accurate margin algorithms are necessary to ensure that volume expansion does not add extra uncertainty to the radiotherapy planning process. We have found significant differences in the 3D margining algorithms of TPSs, devised a simple test to predict inconsistency and suggested corrective action to minimise the variation. c 2007 Elsevier Ireland Ltd. All rights reserved. Radiotherapy and Oncology 86 (2008) 43–47.
Keywords: Margins; Margining algorithms; Treatment planning systems; Quality assurance
Algorithms that allow the user to grow a region of interest (ROI) by a specified 3D margin are an important feature of modern treatment planning and virtual simulation systems. Their most common use is to obtain a planning target volume (PTV) from a gross tumour volume (GTV) where the user specifies the required margins in each of the six directions (right, left, anterior, posterior, superior and inferior). 1
Present address: Department of Radiotherapy, University Hospitals of Leicester, Leicester, UK.
The mathematical basis for automatic margining tools is morphological dilation, the equations for which have been discussed elsewhere [1]. The methods used by the original 3D automatic margining tools, for radiotherapy planning, were reported over a decade ago [2,3] and their superiority over 2D margining tools and manual PTV outlining has been discussed [4–6]. During the commissioning of the Pinnacle treatment planning system (Philips Medical Systems) it became apparent that margins grown by this system were larger than
0167-8140/$ - see front matter c 2007 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.radonc.2007.11.006
44
Evaluation of margining algorithms in planning systems
those grown for identical starting volumes by the Plato treatment planning system (Nucletron). The accurate growing of margins is particularly important when participating in clinical trials of conformal radiotherapy, as larger volumes treated may lead to different outcomes. As part of the quality assurance programme for the UK National Cancer Research Network prostate trial, CHHIP (Conventional or Hypofractionated High Dose IMRT for Prostate Cancer), participating centres were asked to create the planning target volumes by using 3D computer margining algorithms to expand the GTV in a reproducible way. Two test data sets were provided by the Royal Marsden Hospital, and participating centres were asked to grow appropriate margins. A comparison of the results confirmed that the margining algorithms from all the different planning systems involved produced different results. A systematic study was therefore initiated at Clatterbridge Centre for Oncology (CCO) to compare margining algorithms on a number of different treatment planning systems.
Method Margining algorithms in seven treatment planning and virtual simulation systems were tested; Philips Pinnacle (6.2, 7.6 and 8.0), Nucletron Plato (2.7.4), MedCom Prosoma (3.0), Varian Eclipse (7.3 and 7.5), Nucletron MasterPlan (3.0), CMS Xio (4.3.1) and GE Advantage Windows (4.2). The first five were tested at CCO, CMS Xio was tested at Addenbrookes (Cambridge) and GE Advantage Windows at Cookridge (Leeds). The CHHIP protocol requires that three planning target volumes are grown based on the GTV outlined by the Radiation Oncologist, as illustrated in Fig. 1. The relative doses prescribed to these three PTVs are also given in Fig. 1. The PTVs provide examples of anisotropic (PTV3) and isotropic (PTV1) margins as well as double margining (PTV2). The two CHHIP trial sample prostate datasets, which were created at the Royal Marsden Hospital using the Eclipse treatment planning system (Varian Medical Systems), were used as the basis of this study. They consisted of a set of CT images with prostate (P) and prostate plus seminal vesicles (P + SV) outlined; one with 5 mm slice separation and the other with 2.5 mm slice separation. The datasets were transferred to the seven planning systems using DICOM RT transfer and the following margining was performed: a margin of 5 mm, except 0 mm posterior, was applied to P to create PTV3. A uniform margin of 5 mm
PTV3 = prostate + 5mm / 0mm towards rectum Minimum dose coverage: 95% PTV2 = prostate + 10mm / 5mm towards rectum Minimum dose coverage: 91% PTV1 = prostate+SV + 10mm margin Minimum dose coverage: 76%
Fig. 1. CHHIP protocol PTVs and their minimum dose coverage.
was applied to PTV3 to create PTV2. Finally, a uniform margin of 10 mm was applied to P + SV to produce PTV1. In order to provide a reference volume set, margins were grown using the CTV2PTV algorithm (Version 1.10) written by Stroom and Storchi [2], which we had previously implemented in conjunction with the Plato treatment planning system. This algorithm had previously been carefully validated in conjunction with the RT01 prostate study. There were differences of up to 11% in the absolute volumes of P and P + SV as calculated by the different planning systems due to different methods of measuring ROI volumes. In order to eliminate these differences from the margining algorithm comparison, all the structure sets (GTVs and PTVs) were imported into one system, Pinnacle, for the volume calculations. So as to examine where the algorithms differed and to check how self-consistent they were, the following tests were done: an octahedron was drawn on 5 mm CT slices and a uniform margin of 10 mm was applied. The resulting structures were compared to those obtained using the Stroom and Storchi CTV2PTV program. Each system was then used to margin the octahedron once by 50 mm and five times by 10 mm and the resulting volumes were compared. A final test was carried out using the 5 mm sample prostate dataset. A forward planned IMRT plan was created in Pinnacle for the largest PTVs (Pinnacle) using a three-field technique with a concurrent boost. The primary fields were fitted to PTV1, with the boost fields fitted to PTV3, and the plan was optimised to meet the CHHIP minimum PTV doses shown in Fig. 1, while minimising the doses to rectum and bladder. A feature of the CHHIP trial is that all plans must conform to a set of DVH constraints for PTVs and organs at risk, i.e. rectum, bladder and femoral heads. A further plan was created in Pinnacle with the fields fitted to the smallest PTVs (Eclipse 7.3) but with all other parameters identical to the first plan. Rectal dose was compared between the two plans using dose volume histograms. The Eclipse 7.3 plan was copied onto the Pinnacle structure set and the PTV coverage of the resulting plan was compared using DVHs.
Results CHHIP outlines – prostate volumes A comparison of the three PTVs produced by margining according to the CHHIP protocol is shown in Tables 1 and 2 and Fig. 2a and b. The volume obtained for PTV1 using CTV2PTV is shown as a horizontal line, which lies close to the mean value for all the systems. Most of the other volumes lie within 5% of this mean. However, for 5 mm slice separation, Pinnacle’s PTV1 and PTV3 volumes were 107% and 106% of the mean and Eclipse’s (7.3) were 93% and 92%. Changing the slice separation to 2.5 mm reduced Pinnacle’s volumes to 106% and 105% and increased Eclipse’s (7.3) volumes to 94% and 95%. The differences are magnified when a volume is margined twice, i.e. PTV2, so that for 5 mm slice separation only Plato produced a volume within 5% of the mean. For 2.5 mm slice separation the differences were greatly reduced and only Pinnacle and Eclipse 7.3’s PTV2 volumes lie outside the 5% boundaries.
The results for Eclipse show how the algorithm apparently changed between versions 7.3 and 7.5.
CHHIP outlines – effect on plans Fig. 3a compares the rectal DVHs for the plans based on the Eclipse 7.3 and Pinnacle PTVs, together with the CHHIP rectal DVH constraints. This shows that planning using the Pinnacle PTVs (V50 = 42.8%) results in a greater dose to the rectum than planning using the Eclipse 7.3 PTVs (V50 = 36.4%). Fig. 3b shows how the PTV2 coverage is reduced when a plan that adequately treats the Eclipse PTVs is transferred to a Pinnacle structure set. The field sizes would no longer be considered large enough since one of the criteria for an ‘adequate’ CHHIP plan is that the minimum dose to PTV2 should be 91%.
Consistency tests using an octahedron Fig. 4 shows the percentage difference of the volume resulting when the same octahedron was margined once by 50 mm relative to the volume obtained by margining the octahedron five times by 10 mm. There is a correlation between systems that have a small percentage difference when this test is done and those that produce PTVs within 5% of the mean (Fig. 2). Pinnacle’s tendency to produce large volumes and Eclipse 7.3 tendency to produce small
Xi o as te rp la n Ec lip se 7. 3 Ec lip se 7. 5
o at Pl
M
5 7.
3 se lip Ec
se lip Ec
te
rp
la
7.
n
o
83.9 –
80% Xi
150.7 –
85%
as
196.2 198.3
90%
M
Mean volume CTV2PTV
95%
o
87.9 86.8 85.9 82.1 83.8 83.1 79.3 82.2
at
163.6 158.9 155.7 145.0 148.2 149.0 136.0 149.1
100%
Pl
209.1 204.1 196.8 191.6 195.1 193.4 185.1 194.5
Stroom PTV1
105%
W
Pinnacle 7.6 Prosoma 3.0 Advantage Win 4.2 Plato 2.7.4 Xio 4.3.1 MasterPlan 3.0 Eclipse 7.3 Eclipse 7.5
PTV1
110%
0
PTV3
PTV2
AD
PTV2
PTV3
115%
Pr
PTV1
80%
Pi
b Table 2 Volumes in ml of PTVs produced by each planning system for 2.5 mm slices
85%
3.
129.5 –
90%
E
214.2 –
95%
a
254.2 252.9
100%
G
Mean volume CTV2PTV
Stroom PTV1
105%
om
137.7 134.7 132.5 126.6 126.0 125.3 118.6 134.6
os
238.8 227.8 228.7 205.9 200.7 203.9 181.4 226.6
nn ac le os om a 3. 0 G E AD W
273.0 265.0 255.2 249.2 249.0 247.0 237.4 257.9
le
Pinnacle 7.6 Prosoma 3.0 Advantage Win 4.2 Plato 2.7.4 Xio 4.3.1 MasterPlan 3.0 Eclipse 7.3 Eclipse 7.5
PTV1
110%
Pi
PTV3
PTV2
Pr
PTV2
PTV3
45
115%
nn ac
PTV1
a
Volume normalised to mean volume
Table 1 Volumes in ml of PTVs produced by each planning system for 5 mm slices
Volume normalised to mean volume
A.M. Pooler et al. / Radiotherapy and Oncology 86 (2008) 43–47
Fig. 2. Volumes of PTVs created by margining the GTVs on seven different systems normalised to the mean volumes across the systems. (a) Results from the first CHHIP QA dataset (5 mm slices) at the top and (b) second QA dataset (2.5 mm slices) at the bottom. VolumePTV1 produced by Stroom and Storchi CTV2PTV shown by the horizontal line.
ones is reflected in the results of this test. Fig. 5 illustrates where the differences occur.
Discussion The results of the octahedral margining test illustrate the source of the differences. Since CT scans are performed in discrete slices, the edges of structures in the sagittal or coronal direction are open to interpretation. In Pinnacle (versions 6.2, 7.6 and 8.0), structures are treated as stacked discs, each disc having a thickness equal to the slice separation (see Fig. 6b). A margin is applied by sliding a sphere (or ellipsoid) centred on the edge (see Fig. 6b), around the structure, which results in an over-estimation of the margin. Stroom and Storchi’s CTV2PTV algorithm takes the central slice position, interpolates the contour from one central slice position to the next and slides a sphere (or ellipsoid) around this. Careful examination of the octahedron in the sagittal plane in Eclipse 7.3 shows that, although this system treats structures in the same way as Stroom and Storchi, setting equal margins of 10 mm causes a sphere of 9 mm radius to slide around the structure (Fig. 6a). In Eclipse’s lat-
46
Evaluation of margining algorithms in planning systems
a
Eclipse 7.3
Pinnacle
CHHIP Constraints
Normalised Volume
100% 80% 60% 40% 20% 0% 0
2000
4000
6000
8000
Fig. 5. Sagittal views of structures created using 5 · 10 mm margin (blue) and 1 · 50 mm margin (green) for Pinnacle (left) and Eclipse (right).
Dose in cGy
b
Eclipse 7.3 PTV2
Pinnacle PTV2
91%
a
Normalised Volume
100%
Eclipse 7.5
80%
10mm 60%
Eclipse 7.3 40%
Starting Structure
20% 0% 0
2000
4000
6000
8000
Dose in cGy Fig. 3. (a) Top: rectum DVH for two plans. One created using the Eclipse 7.3 structures and one created using the Pinnacle structures. The CHHIP rectum DVH constraints are also shown. (b) Bottom: DVH for the Eclipse PTV2 and the Pinnacle PTV2 using a plan created on the Eclipse structures. The minimum dose coverage was 92% and 88%, respectively.
b
Pinnacle 10mm
Percentage Difference
20%
Eclipse 7.3
15% 10%
Starting Structure
5% 0% -5%
Stroom
Pinnacle 7.6
Prosoma 3.0
Plato
Masterplan Eclipse 7.3 Eclipse 7.5
ADW
Xio
CTV2PTV
-10% -15% -20%
Fig. 4. Percentage difference in volume between 5 · 10 mm (3 · 10 mm for Prosoma) margins and 1 · 50 mm (1 · 30 mm for Prosoma) margin.
est version 7.5 (supplied with ARIA) the radius of the resulting sphere is slightly greater than 10 mm (Fig. 6a). Margining in Eclipse 7.3 and Pinnacle produced the most extreme results and Fig. 6b shows how expanded structures from Eclipse 7.3, Stroom and Storchi, and Pinnacle compare when imported into Pinnacle.
Fig. 6. (a) Top: sagittal slice from Eclipse showing how margining behaviour has changed between Eclipse versions 7.3 and 7.5. (b) Bottom: sagittal slice from Pinnacle showing the differences between margins produced by Pinnacle, CTV2PTV and Eclipse 7.3.
The normalised volumes of PTV1 (isotropic margins) and PTV3 (anisotropic margins) in Fig. 2 are similar, which implies that each margining algorithm handles isotropic and anisotropic margins similarly. Fig. 2 shows that PTV2
A.M. Pooler et al. / Radiotherapy and Oncology 86 (2008) 43–47
had the greatest range of volumes across the systems; clearly the double margining used to create PTV2 has accentuated the differences. Double margining was used to create PTV2 because unlike our simplified test, the CHHIP protocol requires that the rectum is excluded from PTV3 and that PTV2 overlaps the rectum by 5 mm, hence the recommendation that PTV3 is used as the source structure for creating PTV2. A simple empirical correction can be applied to both Pinnacle and Eclipse 7.3 to correct their margining inconsistencies and minimise the difference in their volumes compared to the other planning systems. Pinnacle margins in the superior/inferior direction should be reduced by half the slice spacing, while for Eclipse 7.3, they should be increased by half the slice spacing. If this is done, the normalised PTV volumes are within 5% as are the two volumes used in the consistency test. Different planning systems measure ROI volumes in different ways; hence the use of DICOM RT to transfer structure sets to Pinnacle for volume measuring. However transferring structures using DICOM RT could potentially lead to a loss of information, for example where an expanded contour does not end on a slice. To minimise any potential loss of information, the margins applied were multiples of the slice spacing, and the extent of the expanded contours in the superior and inferior directions was checked. The relative volumes reported in this study are specific for the tested margins and slice separations, which in this case were multiples of each other, and may also depend on the target shapes. The magnitude of the differences would be different if the margins were not multiples of the slice separation and in this case the results might be affected by the process of DICOM RT transfer.
Conclusions Inter-observer and intra-observer variability in the outlining of GTVs [7] are the greatest contributors to PTV differences. However such differences are likely to vary randomly between patients, while these differences in margining algorithms will result in larger (or smaller) volumes for all patients. In the context of a national trial requiring consistent dose volumes, such as CHHIP, these differences
47
could lead to systematic differences in the plans that are produced by the participating radiotherapy centres. Since margining algorithms not only differ between systems, but are affected by software upgrades, it is recommended that tests of the margining behaviour should be part of the commissioning of a software upgrade and the octahedron test of consistency provides a simple way of doing this.
Acknowledgements The authors would like to thank David Smith and Andrew Morgan (Cookridge, Leeds), Andrew Hoole (Addenbrookes, Cambridge) and Martin Sheen and Philip Mayles (Clatterbridge). * Corresponding author. A.M. Pooler, Physics Department, Clatterbridge Centre for Oncology, Clatterbridge Road, Bebington, Wirral CH63 4JY, UK. E-mail address: [email protected] Received 20 September 2007; received in revised form 31 October 2007; accepted 2 November 2007; Available online 3 December 2007
References [1] Serra J. The hit or miss transform: an algebraic study. In: Serra J, editor. Image analysis and mathematical morphology. London: Academic Press; 1982. p. 43–50. [2] Stroom JC, Storchi PR. Automatic calculation of three-dimensional margins around treatment volumes in radiotherapy planning. Phys Med Biol 1997;42:745–55. [3] Belshi R, Pontvert D, Rosenwald JC, Gaboriaud G. Automatic three-dimensional expansion of structures applied to determination of the clinical target volume in conformal radiotherapy. Int J Radiat Oncol Biol Phys 1997;37:689–96. [4] Bedford JL, Shentall GS. A digital method for computing target margins in radiotherapy. Med Phys 1998;25:224–31. [5] Stroom JC, Korevaar GA, Koper PC, Visser AG, Heijmen BJ. Multiple two-dimensional versus three-dimensional PTV definition in treatment planning for conformal radiotherapy. Radiother Oncol 1998;47:297–302. [6] Ketting CH, Austin-Seymour M, Kalet I, Unger J, Hummel S, Jacky J. Consistency of three-dimensional planning target volumes across physicians and institutions. Int J Radiat Oncol Biol Phys 1997;37:445–53. [7] Seddon B, Bidmead M, Wilson J, Khoo V, Dearnaley D. Target volume definition in conformal radiotherapy for prostate cancer: quality assurance in the MRC RT-01 trial. Radiother Oncol 2000;56:73–83.
Treatment planning
Evaluation of margining algorithms in commercial treatment planning systems Alistair M. Poolera,*, Helen M. Maylesa, Olivia F. Naismithb, John P. Sagec,1 David P. Dearnaleyd, on behalf of CHHIP collaborators a
Physics Department, Clatterbridge Centre for Oncology NHS Foundation Trust, Wirral, UK, bDepartment of Physics, The Royal Marsden NHS Foundation Trust, London, UK, cDepartment of Physics, Arden Cancer Centre, Walsgrave Hospital, Coventry, UK, dDepartment of Academic Radiotherapy, Institute of Cancer Research and Royal Marsden NHS Foundation Trust, Surrey, UK
Abstract Introduction: During commissioning of the Pinnacle (Philips) treatment planning system (TPS) the margining algorithm was investigated and was found to produce larger PTVs than Plato (Nucletron) for identical GTVs. Subsequent comparison of PTV volumes resulting from the QA outlining exercise for the CHHIP (Conventional or Hypofractionated High Dose IMRT for Prostate Ca.) trial confirmed that there were differences in TPS’s margining algorithms. Margining and the clinical impact of the different PTVs in seven different planning and virtual simulation systems (Pinnacle, Plato, Prosoma (MedCom), Eclipse (7.3 and 7.5) (Varian), MasterPlan (Nucletron), Xio (CMS) and Advantage Windows (AW) (GE)) is investigated, and a simple test for 3D margining consistency is proposed. Methods: Using each TPS, two different sets of prostate GTVs on 2.5 mm and 5 mm slices were margined according to the CHHIP protocol to produce PTV3 (prostate + 5 mm/0 mm post), PTV2 (PTV3 + 5 mm) and PTV1 (prostate and seminal vesicles + 10 mm). GTVs and PTVs were imported into Pinnacle for volume calculation. DVHs for 5 mm slice plans, created using the smallest PTVs, were recalculated on the largest PTV dataset and vice versa. Since adding a margin of 50 mm to a structure should give the same result as adding five margins of 10 mm, this was tested for each TPS (consistency test) using an octahedron as the GTV and CT datasets with 2.5 mm and 5 mm slices. Results: The CHHIP PTV3 and PTV1 volumes had a standard deviation, across the seven systems, of 5% and PTV2 (margined twice) 9%, on the 5 mm slices. For 2.5 mm slices the standard deviations were 4% and 6%. The ratio of the Pinnacle and the Eclipse 7.3 PTV2 volumes was 1.25. Rectal doses were significantly increased when encompassing Pinnacle PTVs (V50 = 42.8%), compared to Eclipse 7.3 PTVs (V50 = 36.4%). Conversely, fields that adequately treated an Eclipse 7.3 PTV2 were inadequate for a Pinnacle PTV2. AW and Plato PTV volumes were the most consistent (0.3%) and ( 0.4%). However, the 1 · 50 mm margin in Pinnacle produced a 15.9% larger volume than 5 · 10 mm margins, while for Eclipse 7.3 the single margined volume was 14.3% smaller. These inconsistencies were reduced to 5% by adjusting the superior/inferior margins. Conclusions: Accurate margin algorithms are necessary to ensure that volume expansion does not add extra uncertainty to the radiotherapy planning process. We have found significant differences in the 3D margining algorithms of TPSs, devised a simple test to predict inconsistency and suggested corrective action to minimise the variation. c 2007 Elsevier Ireland Ltd. All rights reserved. Radiotherapy and Oncology 86 (2008) 43–47.
Keywords: Margins; Margining algorithms; Treatment planning systems; Quality assurance
Algorithms that allow the user to grow a region of interest (ROI) by a specified 3D margin are an important feature of modern treatment planning and virtual simulation systems. Their most common use is to obtain a planning target volume (PTV) from a gross tumour volume (GTV) where the user specifies the required margins in each of the six directions (right, left, anterior, posterior, superior and inferior). 1
Present address: Department of Radiotherapy, University Hospitals of Leicester, Leicester, UK.
The mathematical basis for automatic margining tools is morphological dilation, the equations for which have been discussed elsewhere [1]. The methods used by the original 3D automatic margining tools, for radiotherapy planning, were reported over a decade ago [2,3] and their superiority over 2D margining tools and manual PTV outlining has been discussed [4–6]. During the commissioning of the Pinnacle treatment planning system (Philips Medical Systems) it became apparent that margins grown by this system were larger than
0167-8140/$ - see front matter c 2007 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.radonc.2007.11.006
44
Evaluation of margining algorithms in planning systems
those grown for identical starting volumes by the Plato treatment planning system (Nucletron). The accurate growing of margins is particularly important when participating in clinical trials of conformal radiotherapy, as larger volumes treated may lead to different outcomes. As part of the quality assurance programme for the UK National Cancer Research Network prostate trial, CHHIP (Conventional or Hypofractionated High Dose IMRT for Prostate Cancer), participating centres were asked to create the planning target volumes by using 3D computer margining algorithms to expand the GTV in a reproducible way. Two test data sets were provided by the Royal Marsden Hospital, and participating centres were asked to grow appropriate margins. A comparison of the results confirmed that the margining algorithms from all the different planning systems involved produced different results. A systematic study was therefore initiated at Clatterbridge Centre for Oncology (CCO) to compare margining algorithms on a number of different treatment planning systems.
Method Margining algorithms in seven treatment planning and virtual simulation systems were tested; Philips Pinnacle (6.2, 7.6 and 8.0), Nucletron Plato (2.7.4), MedCom Prosoma (3.0), Varian Eclipse (7.3 and 7.5), Nucletron MasterPlan (3.0), CMS Xio (4.3.1) and GE Advantage Windows (4.2). The first five were tested at CCO, CMS Xio was tested at Addenbrookes (Cambridge) and GE Advantage Windows at Cookridge (Leeds). The CHHIP protocol requires that three planning target volumes are grown based on the GTV outlined by the Radiation Oncologist, as illustrated in Fig. 1. The relative doses prescribed to these three PTVs are also given in Fig. 1. The PTVs provide examples of anisotropic (PTV3) and isotropic (PTV1) margins as well as double margining (PTV2). The two CHHIP trial sample prostate datasets, which were created at the Royal Marsden Hospital using the Eclipse treatment planning system (Varian Medical Systems), were used as the basis of this study. They consisted of a set of CT images with prostate (P) and prostate plus seminal vesicles (P + SV) outlined; one with 5 mm slice separation and the other with 2.5 mm slice separation. The datasets were transferred to the seven planning systems using DICOM RT transfer and the following margining was performed: a margin of 5 mm, except 0 mm posterior, was applied to P to create PTV3. A uniform margin of 5 mm
PTV3 = prostate + 5mm / 0mm towards rectum Minimum dose coverage: 95% PTV2 = prostate + 10mm / 5mm towards rectum Minimum dose coverage: 91% PTV1 = prostate+SV + 10mm margin Minimum dose coverage: 76%
Fig. 1. CHHIP protocol PTVs and their minimum dose coverage.
was applied to PTV3 to create PTV2. Finally, a uniform margin of 10 mm was applied to P + SV to produce PTV1. In order to provide a reference volume set, margins were grown using the CTV2PTV algorithm (Version 1.10) written by Stroom and Storchi [2], which we had previously implemented in conjunction with the Plato treatment planning system. This algorithm had previously been carefully validated in conjunction with the RT01 prostate study. There were differences of up to 11% in the absolute volumes of P and P + SV as calculated by the different planning systems due to different methods of measuring ROI volumes. In order to eliminate these differences from the margining algorithm comparison, all the structure sets (GTVs and PTVs) were imported into one system, Pinnacle, for the volume calculations. So as to examine where the algorithms differed and to check how self-consistent they were, the following tests were done: an octahedron was drawn on 5 mm CT slices and a uniform margin of 10 mm was applied. The resulting structures were compared to those obtained using the Stroom and Storchi CTV2PTV program. Each system was then used to margin the octahedron once by 50 mm and five times by 10 mm and the resulting volumes were compared. A final test was carried out using the 5 mm sample prostate dataset. A forward planned IMRT plan was created in Pinnacle for the largest PTVs (Pinnacle) using a three-field technique with a concurrent boost. The primary fields were fitted to PTV1, with the boost fields fitted to PTV3, and the plan was optimised to meet the CHHIP minimum PTV doses shown in Fig. 1, while minimising the doses to rectum and bladder. A feature of the CHHIP trial is that all plans must conform to a set of DVH constraints for PTVs and organs at risk, i.e. rectum, bladder and femoral heads. A further plan was created in Pinnacle with the fields fitted to the smallest PTVs (Eclipse 7.3) but with all other parameters identical to the first plan. Rectal dose was compared between the two plans using dose volume histograms. The Eclipse 7.3 plan was copied onto the Pinnacle structure set and the PTV coverage of the resulting plan was compared using DVHs.
Results CHHIP outlines – prostate volumes A comparison of the three PTVs produced by margining according to the CHHIP protocol is shown in Tables 1 and 2 and Fig. 2a and b. The volume obtained for PTV1 using CTV2PTV is shown as a horizontal line, which lies close to the mean value for all the systems. Most of the other volumes lie within 5% of this mean. However, for 5 mm slice separation, Pinnacle’s PTV1 and PTV3 volumes were 107% and 106% of the mean and Eclipse’s (7.3) were 93% and 92%. Changing the slice separation to 2.5 mm reduced Pinnacle’s volumes to 106% and 105% and increased Eclipse’s (7.3) volumes to 94% and 95%. The differences are magnified when a volume is margined twice, i.e. PTV2, so that for 5 mm slice separation only Plato produced a volume within 5% of the mean. For 2.5 mm slice separation the differences were greatly reduced and only Pinnacle and Eclipse 7.3’s PTV2 volumes lie outside the 5% boundaries.
The results for Eclipse show how the algorithm apparently changed between versions 7.3 and 7.5.
CHHIP outlines – effect on plans Fig. 3a compares the rectal DVHs for the plans based on the Eclipse 7.3 and Pinnacle PTVs, together with the CHHIP rectal DVH constraints. This shows that planning using the Pinnacle PTVs (V50 = 42.8%) results in a greater dose to the rectum than planning using the Eclipse 7.3 PTVs (V50 = 36.4%). Fig. 3b shows how the PTV2 coverage is reduced when a plan that adequately treats the Eclipse PTVs is transferred to a Pinnacle structure set. The field sizes would no longer be considered large enough since one of the criteria for an ‘adequate’ CHHIP plan is that the minimum dose to PTV2 should be 91%.
Consistency tests using an octahedron Fig. 4 shows the percentage difference of the volume resulting when the same octahedron was margined once by 50 mm relative to the volume obtained by margining the octahedron five times by 10 mm. There is a correlation between systems that have a small percentage difference when this test is done and those that produce PTVs within 5% of the mean (Fig. 2). Pinnacle’s tendency to produce large volumes and Eclipse 7.3 tendency to produce small
Xi o as te rp la n Ec lip se 7. 3 Ec lip se 7. 5
o at Pl
M
5 7.
3 se lip Ec
se lip Ec
te
rp
la
7.
n
o
83.9 –
80% Xi
150.7 –
85%
as
196.2 198.3
90%
M
Mean volume CTV2PTV
95%
o
87.9 86.8 85.9 82.1 83.8 83.1 79.3 82.2
at
163.6 158.9 155.7 145.0 148.2 149.0 136.0 149.1
100%
Pl
209.1 204.1 196.8 191.6 195.1 193.4 185.1 194.5
Stroom PTV1
105%
W
Pinnacle 7.6 Prosoma 3.0 Advantage Win 4.2 Plato 2.7.4 Xio 4.3.1 MasterPlan 3.0 Eclipse 7.3 Eclipse 7.5
PTV1
110%
0
PTV3
PTV2
AD
PTV2
PTV3
115%
Pr
PTV1
80%
Pi
b Table 2 Volumes in ml of PTVs produced by each planning system for 2.5 mm slices
85%
3.
129.5 –
90%
E
214.2 –
95%
a
254.2 252.9
100%
G
Mean volume CTV2PTV
Stroom PTV1
105%
om
137.7 134.7 132.5 126.6 126.0 125.3 118.6 134.6
os
238.8 227.8 228.7 205.9 200.7 203.9 181.4 226.6
nn ac le os om a 3. 0 G E AD W
273.0 265.0 255.2 249.2 249.0 247.0 237.4 257.9
le
Pinnacle 7.6 Prosoma 3.0 Advantage Win 4.2 Plato 2.7.4 Xio 4.3.1 MasterPlan 3.0 Eclipse 7.3 Eclipse 7.5
PTV1
110%
Pi
PTV3
PTV2
Pr
PTV2
PTV3
45
115%
nn ac
PTV1
a
Volume normalised to mean volume
Table 1 Volumes in ml of PTVs produced by each planning system for 5 mm slices
Volume normalised to mean volume
A.M. Pooler et al. / Radiotherapy and Oncology 86 (2008) 43–47
Fig. 2. Volumes of PTVs created by margining the GTVs on seven different systems normalised to the mean volumes across the systems. (a) Results from the first CHHIP QA dataset (5 mm slices) at the top and (b) second QA dataset (2.5 mm slices) at the bottom. VolumePTV1 produced by Stroom and Storchi CTV2PTV shown by the horizontal line.
ones is reflected in the results of this test. Fig. 5 illustrates where the differences occur.
Discussion The results of the octahedral margining test illustrate the source of the differences. Since CT scans are performed in discrete slices, the edges of structures in the sagittal or coronal direction are open to interpretation. In Pinnacle (versions 6.2, 7.6 and 8.0), structures are treated as stacked discs, each disc having a thickness equal to the slice separation (see Fig. 6b). A margin is applied by sliding a sphere (or ellipsoid) centred on the edge (see Fig. 6b), around the structure, which results in an over-estimation of the margin. Stroom and Storchi’s CTV2PTV algorithm takes the central slice position, interpolates the contour from one central slice position to the next and slides a sphere (or ellipsoid) around this. Careful examination of the octahedron in the sagittal plane in Eclipse 7.3 shows that, although this system treats structures in the same way as Stroom and Storchi, setting equal margins of 10 mm causes a sphere of 9 mm radius to slide around the structure (Fig. 6a). In Eclipse’s lat-
46
Evaluation of margining algorithms in planning systems
a
Eclipse 7.3
Pinnacle
CHHIP Constraints
Normalised Volume
100% 80% 60% 40% 20% 0% 0
2000
4000
6000
8000
Fig. 5. Sagittal views of structures created using 5 · 10 mm margin (blue) and 1 · 50 mm margin (green) for Pinnacle (left) and Eclipse (right).
Dose in cGy
b
Eclipse 7.3 PTV2
Pinnacle PTV2
91%
a
Normalised Volume
100%
Eclipse 7.5
80%
10mm 60%
Eclipse 7.3 40%
Starting Structure
20% 0% 0
2000
4000
6000
8000
Dose in cGy Fig. 3. (a) Top: rectum DVH for two plans. One created using the Eclipse 7.3 structures and one created using the Pinnacle structures. The CHHIP rectum DVH constraints are also shown. (b) Bottom: DVH for the Eclipse PTV2 and the Pinnacle PTV2 using a plan created on the Eclipse structures. The minimum dose coverage was 92% and 88%, respectively.
b
Pinnacle 10mm
Percentage Difference
20%
Eclipse 7.3
15% 10%
Starting Structure
5% 0% -5%
Stroom
Pinnacle 7.6
Prosoma 3.0
Plato
Masterplan Eclipse 7.3 Eclipse 7.5
ADW
Xio
CTV2PTV
-10% -15% -20%
Fig. 4. Percentage difference in volume between 5 · 10 mm (3 · 10 mm for Prosoma) margins and 1 · 50 mm (1 · 30 mm for Prosoma) margin.
est version 7.5 (supplied with ARIA) the radius of the resulting sphere is slightly greater than 10 mm (Fig. 6a). Margining in Eclipse 7.3 and Pinnacle produced the most extreme results and Fig. 6b shows how expanded structures from Eclipse 7.3, Stroom and Storchi, and Pinnacle compare when imported into Pinnacle.
Fig. 6. (a) Top: sagittal slice from Eclipse showing how margining behaviour has changed between Eclipse versions 7.3 and 7.5. (b) Bottom: sagittal slice from Pinnacle showing the differences between margins produced by Pinnacle, CTV2PTV and Eclipse 7.3.
The normalised volumes of PTV1 (isotropic margins) and PTV3 (anisotropic margins) in Fig. 2 are similar, which implies that each margining algorithm handles isotropic and anisotropic margins similarly. Fig. 2 shows that PTV2
A.M. Pooler et al. / Radiotherapy and Oncology 86 (2008) 43–47
had the greatest range of volumes across the systems; clearly the double margining used to create PTV2 has accentuated the differences. Double margining was used to create PTV2 because unlike our simplified test, the CHHIP protocol requires that the rectum is excluded from PTV3 and that PTV2 overlaps the rectum by 5 mm, hence the recommendation that PTV3 is used as the source structure for creating PTV2. A simple empirical correction can be applied to both Pinnacle and Eclipse 7.3 to correct their margining inconsistencies and minimise the difference in their volumes compared to the other planning systems. Pinnacle margins in the superior/inferior direction should be reduced by half the slice spacing, while for Eclipse 7.3, they should be increased by half the slice spacing. If this is done, the normalised PTV volumes are within 5% as are the two volumes used in the consistency test. Different planning systems measure ROI volumes in different ways; hence the use of DICOM RT to transfer structure sets to Pinnacle for volume measuring. However transferring structures using DICOM RT could potentially lead to a loss of information, for example where an expanded contour does not end on a slice. To minimise any potential loss of information, the margins applied were multiples of the slice spacing, and the extent of the expanded contours in the superior and inferior directions was checked. The relative volumes reported in this study are specific for the tested margins and slice separations, which in this case were multiples of each other, and may also depend on the target shapes. The magnitude of the differences would be different if the margins were not multiples of the slice separation and in this case the results might be affected by the process of DICOM RT transfer.
Conclusions Inter-observer and intra-observer variability in the outlining of GTVs [7] are the greatest contributors to PTV differences. However such differences are likely to vary randomly between patients, while these differences in margining algorithms will result in larger (or smaller) volumes for all patients. In the context of a national trial requiring consistent dose volumes, such as CHHIP, these differences
47
could lead to systematic differences in the plans that are produced by the participating radiotherapy centres. Since margining algorithms not only differ between systems, but are affected by software upgrades, it is recommended that tests of the margining behaviour should be part of the commissioning of a software upgrade and the octahedron test of consistency provides a simple way of doing this.
Acknowledgements The authors would like to thank David Smith and Andrew Morgan (Cookridge, Leeds), Andrew Hoole (Addenbrookes, Cambridge) and Martin Sheen and Philip Mayles (Clatterbridge). * Corresponding author. A.M. Pooler, Physics Department, Clatterbridge Centre for Oncology, Clatterbridge Road, Bebington, Wirral CH63 4JY, UK. E-mail address: [email protected] Received 20 September 2007; received in revised form 31 October 2007; accepted 2 November 2007; Available online 3 December 2007
References [1] Serra J. The hit or miss transform: an algebraic study. In: Serra J, editor. Image analysis and mathematical morphology. London: Academic Press; 1982. p. 43–50. [2] Stroom JC, Storchi PR. Automatic calculation of three-dimensional margins around treatment volumes in radiotherapy planning. Phys Med Biol 1997;42:745–55. [3] Belshi R, Pontvert D, Rosenwald JC, Gaboriaud G. Automatic three-dimensional expansion of structures applied to determination of the clinical target volume in conformal radiotherapy. Int J Radiat Oncol Biol Phys 1997;37:689–96. [4] Bedford JL, Shentall GS. A digital method for computing target margins in radiotherapy. Med Phys 1998;25:224–31. [5] Stroom JC, Korevaar GA, Koper PC, Visser AG, Heijmen BJ. Multiple two-dimensional versus three-dimensional PTV definition in treatment planning for conformal radiotherapy. Radiother Oncol 1998;47:297–302. [6] Ketting CH, Austin-Seymour M, Kalet I, Unger J, Hummel S, Jacky J. Consistency of three-dimensional planning target volumes across physicians and institutions. Int J Radiat Oncol Biol Phys 1997;37:445–53. [7] Seddon B, Bidmead M, Wilson J, Khoo V, Dearnaley D. Target volume definition in conformal radiotherapy for prostate cancer: quality assurance in the MRC RT-01 trial. Radiother Oncol 2000;56:73–83.