Robust optimization strategies for lung...
Transcript of Robust optimization strategies for lung...
Robust optimization strategies for lung SBRT
Travis McCaw, PhD, DABRNCCAAPM Spring Meeting
April 26, 2019
photon
Objectives
• Motivation for robust optimization• Approaches to robust optimization• Review of literature on applications of RayStation robust optimization• Phantom-based study of robust optimization in photon lung SBRT
Conventional radiation therapy workflow
• Simulation• Contouring• Treatment planning• Plan review and QA• Treatment
7-10 business days
Potentially > 8 weeks
Treatment uncertainties
• Target segmentation• Weight change• Mechanical uncertainty• Setup error• Intrafraction motion
Treatment uncertainties
• Target segmentation• Weight change• Mechanical uncertainty• Setup error• Intrafraction motion
Hong et al, Radiother Oncol 103, 2012
Treatment uncertainties
• Target segmentation• Weight change• Mechanical uncertainty• Setup error• Intrafraction motion
Templeton et al, Med Phys 42, 2015
Treatment uncertainties
• Target segmentation• Weight change• Mechanical uncertainty• Setup error• Intrafraction motion Addressed with PTV margin
Planning Target Volume
• ICRU 62– “The PTV is a geometrical
concept used for treatment planning, and it is defined to select appropriate beam sizes and beam arrangements, to ensure that the prescribed dose is actually delivered to the CTV.”
Static dose cloud approximation
• Assumption that spatial dose distribution is unperturbed by variation in patient position– Delivered dose given by
convolution of planned dose distribution with patient position PDF
– Invalid near heterogeneities (Craig et al, 2003)
Craig et al, Med Phys 30, 2003
Treatment planning with respiratory motion
• ITV and PTV treat all possible tumor positions– Inaccurate representation of treated patient anatomy
• Multiple anatomy optimization has potential to reduce normal tissue dose with equivalent target coverage
Watkins et al, Med Phys 41, 2014
Robust optimization
• Conventional optimization assumes static matrix mapping fluence to dose
• Robust planning addresses uncertainty in dose distribution from a given incident fluence
𝑘𝑘 ∈
Robust optimization techniques
• Stochastic programming– Minimizes expected value of objective
function
• Composite worst-case minimax
• Objective-wise worst-case minimax
• Voxel-wise worst-case minimax
Comparison of worst-case minimax techniques
• No difference between methods without target coverage/OAR sparing conflict• Composite
– Can disregard “easy” scenarios– Best worst-case objective values
• Voxelwise– Overly conservative interpretation of DVH objectives– Better for severe target coverage/OAR sparing conflict
Fredriksson and Bokrantz, Med Phys 41, 2014
Phantom-based comparison of planned and measured doses for several lung VMAT planning techniques
• Compared several planning strategies– ITV with PTV expansion– GTV with PTV expansion– ITV with PTV expansion on average dataset– Robust optimization using all 4DCT phases– Robust optimization using mid and extreme 4DCT phases– Robust optimization on single dataset with positioning offsets– ITV with PTV expansion using intermediate density override of ITV-GTV
• Evaluated all plans using recalculation on individual 4DCT phases
Archibald-Heeren et al, J Appl Clin Med Phys 18, 2017
Phantom-based comparison of planned and measured doses for several lung VMAT planning techniques
Archibald-Heeren et al, J Appl Clin Med Phys 18, 2017
Robust optimization for setup uncertainty
• Compared ITV planning using robust optimization to ITV planning with PTV margin– 20 lung patients– Median ITV: 10.39 cc (3.29-107.23 cc)– Tumor motion: 1.45 cm (0.54-3.4 cm)
• Prescribed to PTV (if present) or ITV D95
Zhang et al, J Appl Clin Med Phys 19, 2018
Robust optimization for setup uncertainty in lung SBRT
• Evaluated use of robust optimization for setup uncertainty– Compared ITV planning using robust optimization to ITV planning with
PTV margin– Investigated impact of tumor volume and motion amplitude– Phantom study + 10 SBRT lung patients– ITV: 4.65 cc (1.74-7.76 cc)– Motion amplitude: 0.5 cm (0.2-1.5 cm)
• No renormalization of plans for consistent target coverage
Liang et al, Prac Rad Oncol 9, 2019
Robust optimization for setup uncertainty in lung SBRT
• Robust plans: 2199 MU (1917-2615 MU)• PTV plans: 2219 MU (1997-2700 MU)
Liang et al, Prac Rad Oncol 9, 2019
Questions after review of literature
• Importance of planning scan selection?• How to normalize robustly optimized dose distributions for comparison
with conventional planning practice?• 4D robust optimization versus ITV planning for management of
respiratory motion?• Robust optimization versus PTV planning for management of setup
uncertainty?• How to evaluate quality of robust treatment plans?
Phantom-based study of robust optimization for lung SBRT
• 4DCT acquired of respiratory motion phantom– 3 cm diameter GTV, 2 cm
motion amplitude, 4 sec motion period
Evaluated robust optimization techniques
• Five separate treatment plans prepared using phantom CT datasets• Evaluation dose distributions
– All 4DCT phases + average reconstruction– Perturbed isocenter positions of 0.5 cm magnitude
• Normalization such that mean over 4DCT phases GTV D90 = 60 Gy/5 fxPlan 1 Plan 2 Plan 3 Plan 4 Plan 5
Planning scan AverageMid-exhalation
phaseEnd-exhalation
phaseMid-exhalation
phaseEnd-exhalation
phase
Target volume Composite GTV, PTV
GTV, PTV GTV, PTV GTV GTV
4DCT optimization No Yes Yes Yes YesIsocenter uncertainty None None None 0.5 cm isotropic 0.5 cm isotropic
Optimization objectives
Plan 1
Plans 2 & 3
Plans 4 & 5
Robust optimization settings
Plan evaluation
Comparison of robust optimized plans
Mean DVH metrics from recalculation of each treatment plan on all ten 4DCT phasesPlan 1 Plan 2 Plan 3 Plan 4 Plan 5
MU 1630 1678 1766 1693 1761GTV D98 (Gy) 58.37 (0.73) 57.36 (0.37) 57.38 (0.39) 57.47 (0.33) 56.81 (0.40)GTV D90 (Gy) 59.98 (0.75) 60.00 (0.38) 59.97 (0.47) 60.00 (0.33) 60.03 (0.48)GTV 0.4PITV 0.03 (0.00) 0.03 (0.00) 0.03 (0.00) 0.04 (0.00) 0.04 (0.00)GTV 0.5PITV 0.05 (0.00) 0.05 (0.00) 0.05 (0.00) 0.06 (0.00) 0.06 (0.00)GTV PITV 0.19 (0.00) 0.20 (0.00) 0.20 (0.00) 0.23 (0.00) 0.23 (0.00)PTV D95 (Gy) 49.27 (0.45) 47.65 (0.43) 47.53 (0.71) 44.16 (0.86) 43.88 (0.91)PTV Dmean (Gy) 59.11 (0.81) 59.18 (0.78) 59.43 (0.82) 58.20 (0.91) 58.38 (0.94)Mean DVH metrics from recalculation of each treatment plan with perturbed isocenter positions
Plan 2 Plan 3 Plan 4 Plan 5GTV D98 (Gy) 52.39 (3.17) 51.17 (2.68) 49.76 (3.92) 47.96 (3.38)GTV D90 (Gy) 57.56 (1.90) 56.33 (1.90) 57.20 (1.96) 55.30 (2.39)GTV 0.4PITV 0.03 (0.00) 0.03 (0.00) 0.04 (0.00) 0.04 (0.00)GTV 0.5PITV 0.05 (0.00) 0.05 (0.00) 0.06 (0.00) 0.06 (0.00)GTV PITV 0.20 (0.01) 0.20 (0.01) 0.22 (0.01) 0.22 (0.01)
Future work
• Robust plan evaluation– New evaluation tools in RayStation 8
• Voxel-wise min/max distributions, DVH clusters, clinical goal evaluation over all scenarios– Voxel-wise minimum CTV V95 highly correlated with PTV V95 (Korevaar et al,
2018)• Evaluate performance with nearby critical structure• Prospective identification of cases where robust planning is advantageous
– Robust planning vs motion management
Korevaar et al, Multi-scenario robustness evaluation; transition to a ‘proton proof’ alternative to PTV evaluation, ESTRO 2018
Summary
• Management of setup uncertainty with PTV assumes static dose cloud– Invalid near heterogeneities
• Multiple-anatomy optimization offers potential for improved normal tissue sparing
• Robust optimization minimizes max cost function over all input scenarios– Superior normal tissue sparing observed relative to ITV-based planning– Further investigation needed to determine appropriate clinical use