software for plan quality control - wordpress.com...also has the underlying framework in place for...

19
Software for plan quality control D La Russa *1 , R Taylor 1 , E Henderson 1 , K Lekx 1 , D Mason 2 , (1) The Ottawa Hospital Cancer Centre, Ottawa, ON, (2) R. S. McLaughlin Durham Regional Cancer Centre, Oshawa, ON Introduction As a result of continuous developments in radiation therapy, treatment objectives and constraints are in a constant state of flux. Further, with few exceptions, comparing treatment plans with pre-selected objectives and constraints is often manual, time-consuming, error-prone, and necessarily case-dependent. For these reasons plan quality control (QC) is difficult to standardize and is susceptible to human factors modes of failure. In an effort to improve the consistency of the plan QC process, we have developed software that automates the comparison between patient dose-volume histogram (DVH) information and pre-selected dosimetric objectives and constraints. This software also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software The software presented here is a free, open-source web application developed using the Python programming language in combination with the Django web framework. A single installation is hosted on a server connected to the institution’s network, although local, non-networked installations are also an option. Compatibility with the most common operating system/server/database combinations has been established, and the single network installation simplifies version control and IT support. The underlying codebase is such that design, content, and Python code are inherently separate to better accommodate institution-specific requirements and customization, and foster community-supported development. These features are in keeping with recommended design considerations [1], and share many characteristics of other software solutions designed for similar purposes [2]. Users interact with the software via any web browser on any network-connected device with access to the server. User authentication is required for both login and document/plan approval. The former serves to restrict access to patient information while the latter acts partially as a forcing function in the plan QC process. Several user groups are supported to distinguish between the rights and roles of treatment planners, physicians, and physicists/system administrators. In this initial version of the software, treatment plans are evaluated via a comparison of patient DVH data with pre- set DVH objectives and constraints. Users select a set of objectives and constraints (i.e. a careplan) from a database together with DVH data exported from the treatment planning system (TPS). The XiO, Monaco, and TomoTherapy TPSs are currently supported, but users have the freedom to add support for additional TPSs via configurable plugins. The sets of objectives and constraints are defined by users with administrative rights and can include both “hard” and “soft” limits. However, end-users can override default values to accommodate instructions specific to a particular plan. Plans that fail to meet careplan criteria, or that are compared with non default objectives and constraints, require an electronic approval from a physician before a report can be generated. Figure 1 features the intuitive interface via an example of matching a plan with a careplan. The user selects the planning system they are working with, enters the patient ID, and selects the plan (there can be more than one) and corresponding careplan. Menu options are consolidated and kept to a minimum to guide work-flow, maintain efficiency, and minimize errors. Figure 1: Example of selecting a treatment plan/careplan combination.

Upload: others

Post on 02-Aug-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

Software for plan quality control

D La Russa*1, R Taylor1, E Henderson1, K Lekx1, D Mason2, (1) The Ottawa Hospital Cancer Centre, Ottawa, ON,

(2) R. S. McLaughlin Durham Regional Cancer Centre, Oshawa, ON Introduction As a result of continuous developments in radiation therapy, treatment objectives and constraints are in a constant state of flux. Further, with few exceptions, comparing treatment plans with pre-selected objectives and constraints is often manual, time-consuming, error-prone, and necessarily case-dependent. For these reasons plan quality control (QC) is difficult to standardize and is susceptible to human factors modes of failure. In an effort to improve the consistency of the plan QC process, we have developed software that automates the comparison between patient dose-volume histogram (DVH) information and pre-selected dosimetric objectives and constraints. This software also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software The software presented here is a free, open-source web application developed using the Python programming language in combination with the Django web framework. A single installation is hosted on a server connected to the institution’s network, although local, non-networked installations are also an option. Compatibility with the most common operating system/server/database combinations has been established, and the single network installation simplifies version control and IT support. The underlying codebase is such that design, content, and Python code are inherently separate to better accommodate institution-specific requirements and customization, and foster community-supported development. These features are in keeping with recommended design considerations [1], and share many characteristics of other software solutions designed for similar purposes [2]. Users interact with the software via any web browser on any network-connected device with access to the server. User authentication is required for both login and document/plan approval. The former serves to restrict access to patient information while the latter acts partially as a forcing function in the plan QC process.  Several user groups are supported to distinguish between the rights and roles of treatment planners, physicians, and physicists/system administrators. In this initial version of the software, treatment plans are evaluated via a comparison of patient DVH data with pre-set DVH objectives and constraints. Users select a set of objectives and constraints (i.e. a careplan) from a database together with DVH data exported from the treatment planning system (TPS). The XiO, Monaco, and TomoTherapy TPSs are currently supported, but users have the freedom to add support for additional TPSs via configurable plugins. The sets of objectives and constraints are defined by users with administrative rights and can include both “hard” and “soft” limits. However, end-users can override default values to accommodate instructions specific to a particular plan. Plans that fail to meet careplan criteria, or that are compared with non default objectives and constraints, require an electronic approval from a physician before a report can be generated. Figure 1 features the intuitive interface via an example of matching a plan with a careplan. The user selects the planning system they are working with, enters the patient ID, and selects the plan (there can be more than one) and corresponding careplan. Menu options are consolidated and kept to a minimum to guide work-flow, maintain efficiency, and minimize errors. Figure 1: Example of selecting a treatment plan/careplan combination.

Page 2: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

Figure 2 shows how results of a comparison between a treatment plan of the head-and-neck region and its corresponding careplan are summarized. The software automatically matches structures from the plan with structures in the careplan if they share the same name. Unmatched structures can be selected from dropdown menus. Each objective and constraint is listed next to the result from the plan, provided it is associated with a patient structure. Otherwise the user has the option to ignore (skip) the constraint. The comparison in figure 2 also shows the effect of overriding a hard constraint, and adding soft constraints if defaults are not already present. Figure 2: List of constraints for a head-and-neck careplan compared with DVH data from a head-and-neck treatment plan.

In addition to producing a summary of the compliance with a careplan, users can optionally display imported DVH data for each patient structure. This DVH display can be included in an optional pdf report generated after results are submitted to the database. Results stored in the database include the comparison with the associated careplan, the date and time of submission, and the DVH data for each patient structure. Multiple plans can be stored for the same patient, and no patient identifying information is stored apart from the patient ID. If a user overrides one or more default constraints, as shown in figure 2, approval by a designated physician user will be required prior to submitting to the database and generating a report. Thus, the software acts as a forcing function in the plan QC process by ensuring a physician acknowledges the failure to meet the careplan specifications. This step that is distinct from approval of the treatment plan itself. The data stored in the database is conducive for quantitative analysis. By storing the DVH data, institutions will have the ability to track or trend a variety of dosimetric values, such as minimum, maximum, mean, or median doses, the volume of a given structure covered by a clinically relevant dose, etc. Additionally, users can analyze careplan compliance rates by plan, structure, or by individual objectives/constraints. Furthermore, the structure of this software is such that it can easily incorporate radiobiological evaluation of DVH data (TCP and NTCP calculations), as implemented by others [3, 4], with the potential to correlate with biological endpoints.

Page 3: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

Summary We have developed software to automate the process of comparing DVH data from patient plans with a predefined set of objectives and constraints (a careplan). Case-by-case deviations from predefined careplans are gracefully handled by allowing constraints to be overridden while also enforcing that these deviations are both documented and have received physician approval. In addition the software allows for trending and tabulation of various aggregated dosimetric metrics, and analysis of careplan compliance rates. Further, by storing DVH data in an accessible database, this software provides a framework for radiobiological modeling and for correlating planning outcomes with clinical endpoints. This software will be released as an open source project and made freely available to other clinics via an online repository. Software support and feedback from users will be provided via an online forum.

References [1] L. Zhang et al, Computer Methods and Programs in Biomedicine, 110, 528 – 537, 2013 [2] M. Biermann, Computer Methods and Programs in Biomedicine, 114, 70 – 79, 2014 [3] B. Sanchez-Nieto and A. Nahum, Medical Dosimetry, 25, 71 – 76, 2000 [4] J. Uzan and A. Nahum, The British Journal of Radiology, 85, 1279 – 1286, 2012

Page 4: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

Treatment planning workflow for very high-energy electron beam radiotherapy

Magdalena Bazalova*1, Bradley Qu1, Bianey Palma1, Bjorn Hårdemark2, Elin Hynning2, Peter Maxim1, Billy Loo1, (1) *Stanford University, Stanford, CA, USA (2) RaySearch Laboratories AB, Stockholm, Sweden

1. Introduction Deep-seated tumors are typically treated with 6-15 MV photon beams delivered by medical linear accelerators. Previous work has demonstrated in principle dosimetric advantages of using very high-energy (50-250 MeV) electron (VHEE) beams for radiation therapy of deep-seated tumors [1-5]. VHEE plans are not only superior to IMRT plans, but they can be delivered by orders of magnitude faster. We have proposed to build a VHEE treatment machine with a compact electron accelerator to deliver radiotherapy with VHEE scanning beam. In this paper, we present a workflow for treatment planning for intensity-modulated scanning VHEE beam therapy. 2. Methods A treatment planning tool for radiotherapy with VHEE scanning beam has been developed by linking Monte Carlo (MC) beamlet dose calculations in EGSnrc codes with RayStation (RaySearch Laboratories AB, Stockholm Sweden), a commercially available treatment planning system (Figure 1). To facilitate treatment planning for a large number of cases and treatment machine parameters, a Matlab (The Mathworks, Nattick, MA) graphical user interface (GUI) was developed (Figure 2a).

2.1. Matlab GUI First, patient CT images with the DICOM structure file are loaded in the GUI (Figure 2a) and the target structure to be treated is chosen. The user then selects treatment-planning parameters, such as the electron beam energy, beamlet size and spacing (distance between neighboring beamlets at isocenter), and the number of uniformly distributed beams (Table 1). The GUI then automatically generates MC input files, runs dose calculations, and writes a RayStation header file with beamlet information for each beam of the plan.

Figure 2: Matlab GUI for MC beamlet generation (a), VHEE spot scanning inverse treatment planning optimization in RayStation (b) for the studied pediatric case. The PTV is shown in red.

Figure 1: Workflow diagram for treatment planning delivered with VHEE scanning beam. EGSnrc input files are generated with Matlab for the user-selected parameters such as beamlet size and spacing. MC beamlet doses are calculated on a 64-CPU cluster and optimized in RayStation. Four 1-mm 100 MeV beamlets spaced by 8 mm are shown.

Page 5: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

Table1: Treatment planning parameters selectable in the GUI and studied for the presented pediatric case. Note that bore radius defined as the distance from the machine exit window to the isocenter was kept at 50 cm in the GUI.

Energy (MeV)

Number of beams Bore radius (cm)

Beamlet size (mm)

Beamlet spacing (mm)

GUI 50-150 arbitrary 50 1-7 1-7 Pediatric case 60, 80, 100 13, 17, 36 40, 50 0.1, 1.0, 3.0 2.0, 2.5, 3.5

2.2. Monte Carlo calculations BEAMnrc was used to pre-generate beamlets of various energies (50-150 MeV) and sizes (0.1-7 mm) defined by the full width at half maximum (FWHM) at the machine exit window. Phase-space files of monoenergetic VHEE beams collected at the machine exit window were used for patient dose calculations in DOSXYZnrc MC code. The DOSXYZnrc input files for each beamlet were generated based on the GUI user-specified treatment parameters. For each beamlet simulation, 5×104 electrons were modeled resulting in statistical uncertainty of less than 2% along the beamlet path. 2.3. Inverse treatment planning optimization Spot-scanning inverse treatment planning optimization was performed in a research version of RayStation. For each plan, MC-simulated beamlets were imported and the RayStation beam setup was generated using a python script and the GUI-generated header file. Target and organ-at-risk constraints and objectives were determined based on the DVHs of the clinical treatment plan. Once the objectives were satisfied, the dose to normal tissue was decreased within the limits of the VHEE treatment plan. VHEE treatment plans for a number of treatment parameters were compared to each other and to the clinical plan. 2.4. Pediatric case and studied treatment parameters The presented workflow for VHEE treatment planning is demonstrated on a pediatric case. The patient’s 4.3 cm3 brain PTV wrapped around the brainstem was clinically treated with 6 MV VMAT. The prescription dose to the target was 3620 cGy to 95% volume. The pediatric case was planned with VHEE beams using combinations of treatment parameters summarized in Table 1.

Figure 3: VHEE DVHs and dose distributions as a function of beam energy for 60 MeV (dash-dotted), 80 MeV (dashed), and 100 MeV (solid) plans with 13 beams (a) and as a function of number of beams for 13 (dash-dotted), 17 (dashed), and 36 beam (solid) plans for 80 MeV (b).

Page 6: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

3. Results and Discussion VHEE treatment plans as a function of two studied parameters are described in the following section. Additionally, the best VHEE treatment plan is compared to the clinical 6 MV VMAT plan. 3.1. Parameter study Electron beam energy and the number of beams had the largest effect on the quality of dose distributions (Figure 3). Critical structure dose sparing increased with increasing beam energy and with increasing number of beams. While bore radius and beamlet spacing had a minor effect on dose distributions, the studied beamlet sizes generated nearly identical treatment plans. 3.2. Comparison to VMAT plan The best VHEE plan calculated based on the studied parameters was identified as a plan with 100 MeV 1 mm beamlets spaced by 2 mm delivered with 36 beams in a 40 cm bore radius. Figure 4 presents a comparison of the best VHEE plan with the VMAT plan by means of dose differences in an axial and coronal slice and by means of DVHs. The VHEE dose is tighter around the target compared to the VMAT dose, which is also reflected by more than 30% of mean dose decrease in both temporal lobes and globes. Dose to the cochleae is decreased by up to 70% in the VHEE plan. Additionally, the VHEE integral dose is by 33% lower than the VMAT integral dose.

4. Conclusions We have presented a workflow for treatment planning of VHEE radiotherapy delivered with scanning beam. We have furthermore demonstrated the dosimetric advantage of VHEE radiotherapy over photon VMAT radiotherapy for a pediatric patient. References 1. DesRosiers C, Moskvin V, Bielajew AF, Papiez L. 150-250 MeV electron beams in radiation therapy. Physics in Medicine and Biology. 2000;45(7):1781-805. 2. Papiez L, DesRosiers C, Moskvin V. Very high energy electrons (50-250 MeV) and radiation therapy. Technology in cancer research & treatment. 2002;1(2):105-10. 3. Yeboah C, Sandison GA, Moskvin V. Optimization of intensity-modulated very high energy (50-250 MeV) electron therapy. Physics in Medicine and Biology. 2002;47(8):1285-301. 4. Yeboah C, Sandison GA. Optimized treatment planning for prostate cancer comparing IMPT, VHEET and 15 MV IMXT. Physics in Medicine and Biology. 2002;47(13):2247-61. 5. Bazalova M, Maxim P, Tantawi S, Colby E, Koong A, Loo B W. Monte Carlo Simulations and Experimental Validation of Rapid Dose Delivery with Very High-Energy Electron Beams. Medical Physics. 2012;39(6):3944.

Figure 4: Dose distributions (a), dose difference (b), and DVHs (c) for the best VHEE plan and the clinical 6MV VMAT plan. Normal tissue dose sparing (a,b) and clear separation of DVHs for all critical organs is evident (c).

Page 7: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

Automated delivery and quality assurance of a modulated electron radiation therapy plan

T Connell*1

, A Alexander2, P Papaconstadopoulos

1, M Serban

3, S Devic

4, J Seuntjens

1,3

1 Medical Physics Unit, McGill University, Montreal General Hospital, 1650 Cedar Avenue, Montreal, Québec, H3G 1A4, Canada

2 Medical Physics Department, Saskatoon Cancer Center, 20 Campus Drive, Saskatoon, Saskatchewan, S7N 4H4, Canada 3 Medical Physics Department, McGill University Health Center, Montreal, Québec, H3G 1A4, Canada

4 Department of Radiation Oncology, Jewish General Hospital, Montréal, Québec, H3T 1E2, Canada

Introduction

Modulated electron radiation therapy (MERT) offers the potential to improve healthy tissue sparing over currently

offered techniques through energy modulation to conform the prescribed isodose line to the distal edge of superficial

targets. Challenges remain, however, in accurate beamlet dose calculation, plan optimization, collimation method

and delivery accuracy. Up until this work, full end-to-end deliveries based on automated beam collimation of

inverse optimized plans have not yet been achieved. In this study, we investigate the accuracy and efficiency of a

complete MERT delivery using an in-house built automated collimation device used to deliver an inverse optimized

plan.

Methods

The MERT treatment planning process was performed on a previously treated whole breast irradiation case that

included an electron boost for invasive ductal carcinoma. The goal of the MERT plan was to replace the electron

boost which consisted of 10 Gy in 4 fractions at the 90% isodose line. All dose calculations were performed using

Monte Carlo methods through the MMCTP treatment planning interface and beam weights were determined using

direct aperture optimization implemented through a research-based inverse optimization engine1. The calculated

dose distribution can be seen in Figure 1 (a), and the MU histogram broken down by energy can be seen in Figure

1(b). Electron beam collimation was achieved using an in-house built jaw-based automated collimation device,

known as the FLEC, positioned at the level of the electron cutout2. Field shaping involved simultaneous motion of

FLEC jaws and accelerator photon jaws. The combined all-energy plan was delivered onto a Solid Water phantom

containing EBT3 film for verification against the plan dose that was recalculated onto a solid water slab. The same

plan was also delivered to the MapCHECK 2 QA device (Sun Nuclear Corporation, Florida, USA) as individual

energies for comparison against the planned distributions.

Results and Discussion

The automated delivery, which covered 4 electron energies, 196 subfields and 6183 total MU was completed in 25.8

minutes, including 6.2 minutes of beam-on time. The remainder of the delivery time was spent on collimator jaw

motion and the automated interfacing with the accelerator through key-stroke emulation in service mode. The time

spent in each phase is summarized in Error! Reference source not found..

Table 1: A summary of the time required for each phase of the delivery in the current implementation and with

possible optimizations to leaf sequence and FLEC leaf speed.

Stage of delivery Current time (min) Potential time with field sequencing and

faster FLEC leaf speed (min)

Beam time 6.2 6.2

FLEC leaf travel time 7.5 2.2

Remainder 12.1 12.1

Total 25.8 20.5

Page 8: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

Potential optimizations to the collimator leaf sequencing and leaf travel speed could reduce the overall treatment

time to 20.5 minutes offering shorter delivery times. Further reductions in the delivery time could be achieved by

eliminating low-MU fields and re-optimizing on the remaining beamlets. Also, reductions to the beam-on time could

be achieved by redesigning the scattering foils to reduce the spread of the beam, thereby increasing the electron

fluence per MU. In addition to reduced bremsstrahlung dose, we estimate that the beam-on time could be reduced by

a factor of 2-3 using this technique.

Table 2: Gamma pass rates for the film analysis

Energy (MeV) ⁄ (%) ⁄ (%)

9 62.1 92.3

12 99.8 100.0

16 97.8 99.9

20 98.3 99.9

Combined Plan 98.7 99.9

The 3% / 3 mm and 5% / 5 mm gamma pass rates of the planned and delivered film doses are shown in Error!

Reference source not found.. The low pass rate of the 9 MeV 3% / 3 mm gamma value was attributed to the low

dose deposited to the film. This resulted in a mean systematic dose offset of 2.75 cGy between the planned and film

distributions. A disagreement of this magnitude falls within the normal uncertainty expected from EBT3 film

dosimetry at relatively low doses and should not be interpreted as a negative result on the part of the MERT

planning or delivery process. The combined-energy film delivery (a) and the corresponding Gamma analysis (b)

are shown in Figure 2.

Delivery validation was also performed using a MapCHECK 2 (Sun Nuclear Corporation, Melbourne, FL) device

for individual energies. The results of the 3% / 3 mm and 5% / 5 mm gamma pass rates are summarized in Table 3.

Figure 3 shows the results of the MapCHECK report for (a) 9 MeV, (b) 12 MeV, (c) 16 MeV, and (d) 20 MeV.

For each energy, the left panel shows the dose distribution, with pixels not satisfying the ⁄ gamma

criterion marked with a blue square. The right panel shows the cross-plane profile across the central axis

with the plan being represented by the solid black line (Set 2) and the MapCHECK dose points represented by

yellow circles (Set 1). These results are consistent with the film results and are again generally quite good.

Three detectors within the 8x8 cm2 area of the MapCHECK were identified as being defective and could not be

fixed by recalibration or be excluded from the reported gamma results, therefore, it is expected that actual

values would be slightly higher.

Table 3: Gamma pass rates for the MapCHECK 2 analysis

Energy (MeV) ⁄ ⁄

9 88.8 98.0

12 86.1 95.2

16 89.4 95.7

20 94.8 96.7

Conclusions

These results showed that accurate delivery of MERT utilizing an add-on tertiary electron collimator is possible

using Monte Carlo calculated plans and inverse optimization. Delivery time was found to be clinically acceptable,

although more work remains to be done on device optimization and planning tools in order to further reduce the

time. These results show that MERT is moving closer to becoming a viable option for physicians in the treatment of

superficial malignancies.

Page 9: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

References 1 A. Alexander, F. Deblois, G. Stroian, K. Al-Yahya, E. Heath, J. Seuntjens, "MMCTP: a radiotherapy research

environment for Monte Carlo and patient-specific treatment planning," Phys Med Biol 52, N297-308 (2007) 2 K. Al-Yahya, F. Verhaegen, J. Seuntjens, "Design and dosimetry of a few leaf electron collimator for energy

modulated electron therapy," Med Phys 34, 4782-4791 (2007)

Acknowledgments

This work was partially supported by grants from the Canadian Institutes of Health Research, CIHR MOP 102550

and Natural Sciences and Engineering Research Council Discovery grants numbers 298191 and 386009. Also, this

work was partially supported by the Medical Physics Research Training Network, Natural Sciences and Engineering

Research Council/Collaborative Research and Training Experience initiative no #432290. Slobodan Devic is Senior

Research Scientist supported by the Fonds de Recherche en Santé du Québec (FRSQ).

Figure 1: The MERT dose distribution superimposed on the planning CT (a) and the MU statistics broken down by

energy and MU value (b).

Figure 2: The combined-energy delivered dose on EBT3 film at a depth of 2 cm (a) along with the 5%/5mm gamma

map (b).

Page 10: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

Figure 3: Shown are the 9 MeV (a), 12 MeV (b), 16 MeV (c), and 20 MeV (d) MapCHECK dose distributions on

the left and the crossplane profile on the right.

Page 11: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

Respiratory margin derivation and verification in partial breast irradiation

S Quirk*1, L Conroy2, and WL Smith3 (1) Department of Medical Physics, Tom Baker Cancer Centre, Calgary, Alberta

(2) Department of Physics and Astronomy, University of Calgary, Calgary, Alberta (3) Department of Oncology, University of Calgary, Calgary, Alberta

Introduction Partial breast irradiation (PBI) following breast-conserving surgery is emerging as an effective means to achieve local control while also reducing the volume of irradiated breast tissue. Patients are typically planned on a static CT image; however, treatment is delivered while the patient is free-breathing. Respiratory motion may degrade plan quality by reducing target coverage and/or dose homogeneity. Typically, in external beam partial breast radiotherapy the CTV-to-PTV expansion is 10 mm: 5 mm for respiratory motion and 5 mm for setup errors1. These geometric margins are based on typical ranges of respiratory motion and are designed to encompass all motion. Geometric margins are often conservative, and can overestimate the required coverage. Margins can also be derived dosimetrically by determining the impact on dosimetric plan quality parameters (such as dose coverage). Dosimetric margins are often smaller than geometric margins. Typically margins are designed to cover a portion of the population, not 100%, we have chosen 95% margins in both our derivations. We examine geometric and dosimetric respiratory margin calculations in one dimension and verify the adequacy of the typically used 5 mm margin in three dimensions. Methods We used existing patient data from the RAPID (Randomized Trial of Accelerated Partial Breast Irradiation) trial for this investigation. All patients had ipsilateral and contralateral breast, ipsilateral and contralateral lungs, heart, thyroid, and target volumes contoured by treating physicians during the trial and met rigorous review standards. We used this dataset to create PBI inverse optimized intensity modulated radiotherapy (IMRT) plans that met the planning guidelines used in the RAPID trial. We previously examined an extensive database of over 1000 respiratory traces from 21 healthy volunteers and 125 thoracic cancer patients and have statistically described the shape and extent of motion2. With this comprehensive analysis of respiratory motion, we developed a software tool that allows for explicit inclusion of respiratory variability3. Here, we investigate the necessary respiratory margin for PBI IMRT with three different methods. The first method is a simple geometric margin, using the extent of motion from our respiratory database to derive the required margin. A probability density function (PDF) is constructed from the volunteer traces and the extent of motion analyzed on a population-coverage basis. A 95% geometric margin is found by optimizing exhale and inhale points cut from the PDF. The second method determines the dosimetric one-dimensional margin required to account for respiratory motion. Dose profiles were extracted in the anterior-posterior direction from the PBI IMRT patient plans and convolved with the respiratory PDFs from the volunteer study. A 95% dosimetric margin is determined but cutting the top 5%. The third method is a dosimetric verification of the commonly used 5 mm respiratory margin. The 3D margin can only be verified and not derived because of geometric constraints. The seroma volume is often near to lung/chest wall interface or to the anterior air surface. At these interfaces the planning target volume (PTV) is cut back from skin by 5 mm to form the dose evaluation volume. Instead, the existing margin was verified by evaluating plan quality metrics, including hotspot and target coverage. Respiratory traces with peak-to-peak amplitudes of 2-20 mm were used to create probability density functions that were convolved with the static plan fluence to generate delivered planned fluence. The prescribed dose was then recalculated to determine the delivered dose under respiratory motion.

Page 12: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

Results and Discussion Geometric Margin The volunteer population PDF is shown in Figure 1a. Encompassing the entire range of motion of the population would require a 3 cm margin; however, this would be unnecessarily large for the majority of the population. A more reasonable approach is to develop a margin covering 95% of points in the population respiratory PDF (Mgeometric). The geometric margin depends on which 5% of points are cut from the population PDF and Figure 1b shows the resulting geometric margins for cutting 5% of points. This function is minimized when 2% of exhale points and 3% of inhale points are cut resulting in a geometric margin of 3 mm.

a) b) Figure 1: a) Population respiratory PDF with three possible ways to cut 5% of the points to generate a 95% margin: 5% of exhale points (blue), 5% of inhale points (green), or 2% of exhale points and 3% of inhale (pink). b) The resulting geometric 95% margin for various combinations of cutting 5% of points.

Dosimetric 1D Margin Derivation The 1D margin was derived by convolving ten typical patient dose profiles with 21 volunteer respiratory PDFs in the anterior-posterior direction. The maximum margin for all PDFs and all dose profiles was 3.2 mm. Figure 2 shows the required margin for each patient and each respiratory trace. The dot-dash line shows that the top 5% of points cut for the 95%-margin and the resulting 2 mm 95%-margin.

Figure 2: Each coloured dot represents a different respiratory trace. The y-axis is the resulting required margin for a specific patient dose profile and respiratory trace. They x-axis is group by patient dose profile. The solid line is the current 5 mm respiratory margin used in PBI; the dashed line is the geometric 95%-margin of 3 mm derived in the above analysis; and dot-dashed line is resulting dosimetric 95%-margin of 2 mm derived in this analysis.

ï10 0 10 200

5

10

15

20

Displacement (mm)

Prob

abili

ty (%

)

0 1 2 3 4 52.5

5

7.5

10

Exhale points cut (%)M

geom

etric

(mm

)

1 2 3 4 5 6 7 8 9 10

0

1

2

3

4

5

Patient (n = 10)

Requ

ired

Mar

gin

(mm

)

Page 13: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

Dosimetric 3D Margin Assessment - Escalation of Respiratory Amplitude Ten patient PBI IMRT plans were assessed for the adequacy of the 5 mm respiratory margin. We defined the structure CTVsu as the CTV plus 5 mm margin for setup uncertainties but not including the 5 mm for respiratory motion. By excluding the respiratory 5 mm in our evaluation we are dosimetrically testing the robustness of the respiratory margin. There should be no degradation in coverage or plan quality when respiratory motion is introduced if the 5 mm is adequate. Target coverage (CTVsu) and dose homogeneity were evaluated: CTVsu target defined as the percentage of the CTVsu volume receiving 95% of the dose (V95%) and dose homogeneity as the hotspot (maximum dose 1 cm3) to the ipsilateral breast. Dose homogeneity and not target coverage is the limiting factor when respiratory motion is introduced. Increase in hotspot above the 107% of prescribed dose constraint is observed even at the smallest motion amplitudes (Figure 3a), whereas significant target coverage is not observed until respiratory amplitudes of 15 mm and greater (Figure 3b). Based on this analysis we recommend respiratory management if the amplitude of a patient's respiratory motion is above 10 mm.

a) b)

Figure 3: a) Hotspot (1 cm3) to ipsilateral breast where the solid black line is 107% of the prescribed dose and any points above that line do not meet the dose homogeneity criteria. There were two deviations at 2 and 5 mm and five deviations at 10 - 20 mm. b) V95 is the volume of the CTVsu receiving 95% of the prescribed dose. The solid black line shows the 100% volume requirement; the dashed line is the 99% volume. Any points below the dashed line are major deviations in target coverage. The different coloured points on each plot represent the same patient plan.

Conclusions A variety of methods can be used to determine the required margin for respiratory motion in PBI. Simple evaluation of the geometric respiratory extent of motion to determine the required margin may over-estimate required margins. The current 5 mm respiratory margin was found to be possibly too large (1D geometric, dosimetric) for the majority of patients, potentially resulting in unnecessary normal tissue irradiation. However, assessing margins for coverage alone is insufficient for determining the adequate margin. Our 3D analysis showed that the currently used respiratory margin provides sufficient target coverage; however, coverage is not the most affected plan quality metric under respiratory motion. Increase in hotspots was found even at the smallest respiratory amplitudes. Based on our results we conclude that the 5 mm PBI respiratory margin is adequate for coverage, but due to plan quality degradation effects, respiratory management is recommended for patients with respiratory amplitudes greater than 10 mm. References 1 Baglan et al. Int J Radiat Oncol Biol Phys, 55(2): 302–311, 2003. 2 Quirk et al. J Appl Clin Med Phys, 14(2):90–101, 2013. 3 Quirk et al. Med Phys, 39(8): 4999–5003, 2012.

Static 2 5 10 15 20103

105

107

109

111

Motion (mm)

Hot

spot

(1 c

m3 ) t

o ip

silat

eral

bre

ast

Static 2 5 10 15 2085

90

95

100

105

Motion (mm)

V95

of C

TVsu

Page 14: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

Evaluation of dose difference between VMAT plans with and without jaw tracking

MP Milette*1

,T Teke1,

(1) BC Cancer Agency - Centre for the Southern Interior

Introduction

The goal of this study is to quantify the dose difference between VMAT plans calculated with and without jaw

tracking. A recent study showed some potential advantage of jaw tracking in VMAT1. However this study included

jaw tracking only in the direction of the leaf motion (X direction), meaning that the dosimetric advantage was only

due to leaf transmission. In this study the sites were chosen so that there would be jaw tracking for both the X and Y

jaws. As seen in Figure 1, for plans without jaw tracking in the Y direction there is additional dose (over the leaf

transmission) leaking through abutting leaves that can’t be moved out of the field and therefore move across the

treatment field during delivery.

Figure 1 Jaw positions for a plan with static jaws (yellow) and for a plan with jaw tracking (blue).

Methods

VMAT plans for four head and neck patients with concurrent boost and three pelvis patients with concurrent

prostate boost were used in this study. These two sites were chosen because the complexity of the plan due to

concurrent boost, the larger field sizes and the elongated non spherical PTVs result in jaw tracking in the Y

direction. The Eclipse treatment planning system version 11.031 (Varian Medical System) and the progressive

resolution optimizer (PRO3) were used. Dose distributions were calculated using the AAA algorithm. VMAT plans

with 1 arc and 2 arcs were first generated using the jaw tracking option. The collimator angle was rotated between

30 and 45 degrees for best fit to PTVs. All plans were generated using 6MV arcs. A code was written in Matlab to

convert each VMAT plan with jaw tracking (JT plan) to a VMAT plan with static jaws (SJ plan). The jaw settings

for the SJ plan were set to the maximum jaw settings of the JT plan. The SJ plan dose distribution was then

recalculated and compared to the JT plan dose.

Results and Discussion An example of the X and Y jaws positions vs gantry angle for one of the head and neck patient is shown in Figure 2.

For both head and neck and pelvis plans, the Y jaws position changed more frequently compared to the X jaws. The

mean change in jaw positions ranged from 0.4 cm to 2.3 cm for the X jaws versus 0.7 cm to 3.6 cm for the Y jaws.

For the head and neck plans, the maximum change in jaw positions were higher for the Y jaws (3.7 cm to 11 cm)

compared to the X jaws (2.1 cm to 7.8cm). For the pelvis plans, the maximum change in position was higher for the

X jaws (3.3 cm to 6.2 cm) than the Y jaws (2.8 cm to 3.5 cm).

Page 15: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

Figure 2 Jaw size positions versus gantry angle for the JT plan for one of the head and neck patient.

As seen on Figure 3 and 4 plans with static jaws leave an additional dose trail compared to plans with jaw tracking

at the superior and inferior ends of the treated volume. This additional dose is given to normal tissues and/or critical

structures surrounding the PTV. A summary of the dosimetric comparison of the SJ plan minus the JT plan is

presented in Table 1. For most plans, the maximum point dose difference observed was smaller with 2 arcs plans

versus one arc plans and was as high as 15.4% of the prescribed dose. The average volume receiving an additional

2% of the prescribed dose was 134.3 cc and 113.0 cc for one arc and two arcs plans respectively. The average

volume receiving additional dose of 4% of the prescribed dose was 14.11 cc for one arc plans and 5.7 cc for two arcs

plans.

Max point dose

difference

(% prescribed dose)

V+2%

(cc)

V+4%

(cc)

1 arc 2 arcs 1 arc 2 arcs 1 arc 2 arcs

H&N 1 8.2 8.8 114.4 119.5 12.1 13

H&N 2 7.7 6.7 33.8 6.7 3.7 2.5

H&N 3 9.5 6.3 202.7 199.7 28.2 5.7

H&N 4 15.4 6.9 147.7 86.4 14.3 3.8

Prostate 1 7.5 6.9 145.3 139.9 6.3 6.8

Prostate 2 7.7 5.3 66.5 35.3 4.1 0.5

Prostate 3 12.2 9 229.8 203.3 30.1 7.7

Table 1 Summary of the dosimetric differences between the SJ plan and JT plan. V+2% and V+4% are the volumes

in cc receiving an additional 2% and 4% of the prescribed dose. In each column the numbers for the 1arc plan and

the 2arcs plan are reported.

0

2

4

6

8

10

12

-180 -90 0 90 180

Jaw

po

siti

on

(cm

)

Gantry angle (degree)

X1 X2

0

2

4

6

8

10

12

14

16

18

-180 -90 0 90 180

Jaw

po

siti

on

(cm

)

Gantry angle (degree)

Y1 Y2

Page 16: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

Figure 3 Anterior view of the additional dose received with the SJ plan. Additional doses shown are 2% of

prescribed dose (white, on the left) and 4% of prescribed dose (magenta, on the right). The low dose PTV (red) and

high dose PTV (green) are also shown.

Figure 4 Dose difference (SJ plan – JT plan) for prostate patient (plan with 2 arcs). Isodose difference in % of

prescription dose. Slices superior and inferior of the PTV are shown.

Conclusions

VMAT plans with static jaws leave an additional dose trail compared to VMAT plans with jaw tracking with

significant volumes (up to 230 cc) receiving an additional 2% of the prescription dose. The additional dose trail left

by the 2 arcs VMAT plans was less than the 1 arc VMAT for most plans presented in this study.

Reference 1Kim et al, Assesment of potential jaw-tracking advantage using control point sequences of VMAT planning,

Journal of App1ieds Clininical Medical Physics, 2014, 15(2), 60-168

Page 17: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

Real-Time Interactive Treatment Planning

Q Matthews*1, A Mestrovic1, and K Otto2 (1) BC Cancer Agency - Vancouver Island Centre (2) Physics and Astronomy, University of British Columbia

Introduction Optimization-based external beam radiation treatment planning is a cumbersome process. Dose calculations are slow and optimization algorithms are not capable of realistically assessing the importance of target and healthy tissue trade-offs. In this work we present a novel system for efficient interactive evaluation of patient-specific dosimetric trade-offs, termed Real-Time Interactive Planning (RTIP)1. Patient dose distributions are generated using a novel algorithm that computes a 3D Achievable Dose Estimate (ADE). The ADE algorithm is computationally efficient so as to update dose distributions in effectively real-time while accurately incorporating the limits of what can be achieved in practice. Clinicians interactively manipulate dose-volume histograms (DVHs), isodose, and other dose metrics using a custom graphical user interface while the ADE algorithm continuously updates the 3D dose distribution in near real-time. Once a desirable dose distribution is obtained the monitor units and MLC positions are derived using objectives transferred from the ADE to a conventional optimization-based treatment planning system. In addition to improvements in planning efficiency, we hypothesize that the improved speed and interactivity of RTIP may also result in improved treatment plan quality. We have tested the RTIP system on the clinical scenario of head & neck (H&N) VMAT planning. Eleven H&N patients previously planned and treated with either one-arc or two-arc VMAT were re-planned using the RTIP system. A description of the ADE algorithm and the RTIP system is provided along with a comparison of the H&N VMAT plans created with and without using RTIP. Methods ADE algorithm: VMAT delivery characteristics are simulated using 15 equispaced divergent “beams” positioned around the patient CT/contour data set. Each beam comprises a 2D grid of 2.5x2.5mm beamlets projecting into the patient. These “beams” are surrogate fluence maps, with the degree of beamlet modulation constrained to approximate what is achievable through MLC leaf motions. The dose contribution to each voxel from each beam is calculated by (1) convolving the fluence map with a dose deposition kernel, (2) projecting the result along the fluence map ray lines and (3) assigning the value of the convolved fluence to the voxels that intersect each ray line. The total dose for each voxel is the sum of the contributions from each beam. Importantly, the 2D convolution of fluence and dose deposition kernel is performed in the Fourier domain. Modern CPUs and programming libraries have been optimized to calculate Fourier transforms efficiently, and are exploited here for this purpose. The ADE algorithm completes a 3D dose calculation in 2-20 milliseconds, providing the capability to calculate 50-500 dose distributions per second. Interactive Dose Modification: A Matlab software interface was developed to allow the user to specify and manipulate various dose distribution quantities (e.g., DVHs, isodoses, rate of dose fall-off outside targets) using a mouse or touch screen. As an example, if a dose reduction in a certain DVH point is triggered by the user, (1) the voxels inside the structure within a range of the dose value are identified, (2) the ray-lines originating from each voxel passing through the fluence map of each beam and intersecting the radiation source are calculated, and (3) the fluence matrix elements intersecting these ray-lines are identified and modified proportionately to the desired change (Figure 1). The full 3D dose distribution is then recalculated, and the DVHs are updated. Other dose distribution quantities (e.g., isodoses, rate of dose fall-off outside targets) can be interactively modified in a similar fashion as the DVHs. Both mean dose constraints and DVH point constraints can be added or removed at any time. During a user-triggered dose modification the ADE algorithm continuously adjusts the dose through a rapidly converging feedback process so that no constraints are violated. An example of a DVH manipulation is shown in Figure 2. As the Brainstem DVH is modified multiple sets of DVH curves can be seen, each of which is updated approximately every 40 ms. With each DVH change the voxels within the Brainstem at the specified dose level are identified and

Fig. 1 Ray-lines are projected from the dose modification voxels onto the surrogate fluence maps. Beamlets located at the ray-line intersection are modified according to the magnitude of change desired.

Page 18: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

Fig. 2 The Brainstem DVH is interactively manipulated to lower dose under minimum PTV constraints. The 3D dose distribution is recalculated and new DVHs are displayed every ~40 ms. The Brainstem dose is substantially reduced with small increases observed for RT and LT eyes.

the interactive dose modification algorithm is applied. Each DVH set represents the calculation of a new 3D dose distribution using the ADE algorithm. Changes are greatest at the Brainstem as shown in the DVH and isodose plots. RTIP for H&N VMAT Planning: Structure sets for 11 H&N VMAT patients were imported into the RTIP system. Each patient dose distribution was interactively modified to minimize OAR dose, while constraining the target DVHs to the desired coverage and maximum dose. Once a dose distribution was attained where no further improvements were available the DVH data was transferred into the EclipseTM VMAT optimizer. Conventional VMAT optimization was performed, and the calculated plan was evaluated using clinical acceptability criteria. If any structure(s) violated the criteria, the optimization priorities were judiciously adjusted and the optimization was repeated. Results and Discussion ADE and RTIP Timing Performance: ADE calculation times ranged from 2.4-11.0 milliseconds, depending on the surrogate fluence map size required to cover the extent of the patient contours. Unconstrained isodose and DVH modification times ranged from 3.8-16.8 milliseconds and 10.2-22.6 milliseconds, respectively, depending on the number of voxels requiring modification. For constrained isodose or DVH modifications, equilibrium conditions for a specified change were achieved within a few seconds. For each of the 11 H&N VMAT cases tested, the entire RTIP process was completed in ~2-10 minutes, depending on the number of normal tissue structures and their proximity to targets. Representative H&N VMAT Patient: The results from a representative H&N VMAT patient are presented in Figures 3 and 4. The RTIP generated distribution was used for VMAT optimization, as described above. After one optimization, all planning criteria were easily met or exceeded. However, the max PTV dose was 111%, constituting a minor deviation (≥110% but ≤115%). It was decided that the max PTV dose could be reduced, so the priority on the upper PTV objective was increased and the optimization was repeated. After this second optimization, all criteria were still easily met, and the max PTV dose was reduced to 108.5%. As shown in Figures 3 and 4, the RTIP plan achieved nearly equivalent PTV dose coverage as the clinical plan, but with significantly improved OAR sparing.

Fig. 3 Dose colorwash plots (40-104.9% of prescribed dose) for the clinical and RTIP plans, for a representative H&N VMAT patient. The RTIP plan yielded improved sparing of the spinal canal, parotids, and SMGs.

Clinical Plan RTIP Plan

Left SMG

Left SMG

Spinal Canal

Spinal Canal

ParotidsParotids

Page 19: Software for plan quality control - WordPress.com...also has the underlying framework in place for tracking and/or trending variation in dose-volume metrics. Description of the software

RTIP Plan Performance for H&N VMAT: The results for PTV coverage, PTV mean and max dose, and selected OARs (spinal canal, right parotid, and left SMG), for 11 patients are summarized in figure 5. No RTIP plans had any major deviations, whereas one clinical plan (#10) had a major deviation (PTV max dose). None of the RTIP plans produced clinically inferior organ-at-risk (OAR) sparing. However, 10 of 11 RTIP plans achieved substantially improved sparing of one or more OARs, with the other plan (#2) achieving similar overall OAR sparing. Importantly, 10 of the 11 RTIP plans required only one or two optimizations. The other plan (#9) required four optimizations due to the difficult scenario of the high dose PTV overlapping with the spinal canal planning-at-risk volume (PRV). Interestingly, for 8 out of 11 patients it was evident that the clinical plan was accepted as soon as certain OARs met the minimum criteria (e.g., right parotids & left SMG in figure 5). For each of these cases the RTIP plans directly achieved improved OAR sparing, without compromising dose to targets or other OARs. Conclusions RTIP is a novel system for manipulating and updating achievable dose distributions in real-time. Dosimetric trade-offs are evaluated by direct manipulation of DVHs, isodoses, rate of dose fall-off or any other dose metric. Clinical H&N VMAT plans generated using RTIP show improved OAR sparing while substantially improving the efficiency of the planning process. References 1 K. Otto, “Fast VMAT Planning with Interactive Real-Time Dose Manipulation”, 2013 AAPM Annual Meeting,

Indianapolis, IN. Disclosures / Conflict of Interest Statement: One author has a commercial interest in the presented materials.

Fig. 5 Plan comparison for 11 H&N VMAT patients re-planned using RTIP. Dashed lines indicate the clinical criteria for each quantity (orange – minor deviation, red – major deviation).

1 2 3 4 5 6 7 8 9 10 110

5

10

15

20

25

30

35

Patient Number

Mean

Dose

(G

y)

Left Submandibular Gland Dmean

1 2 3 4 5 6 7 8 9 10 110

10

20

30

Patient Number

Mean

Dose

(G

y)

Right Parotid Dmean

1 2 3 4 5 6 7 8 9 10 110

10

20

30

40

Patient Number

Dose

to 0

.03

cc (

Gy)

Spinal Canal D0.03cc

1 2 3 4 5 6 7 8 9 10 1190

100

110

120

130

Patient Number

Max P

TV

Dose

(%

)PTV hd Dmax (%)

1 2 3 4 5 6 7 8 9 10 1196

98

100

102

104

Patient Number

Mean

PTV

Dose

(%

)

PTV hd Dmean

1 2 3 4 5 6 7 8 9 10 1194

96

98

100

102

Patient Number

Volu

me R

ece

ivin

g 9

5%

Dose

(%

)

PTV hd V95%

Clinical PlanRTIP Plan

PTV V95% PTV Mean Dose PTV Max Dose

Spinal Canal D0.03cc Right Parotid Mean Dose Left SMG Mean Dose

Left SMG - Clinical PlanDmean = 25.7 Gy

Left SMG - RTIP PlanDmean = 15.8 Gy

R Parotid - Clinical PlanDmean = 20.5 Gy

R Parotid - RTIP PlanDmean = 12.8 Gy

Spinal Canal - Clinical PlanD0.03cc = 38.8 GyDmean = 15.1 Gy

Spinal Canal - RTIP PlanD0.03cc = 29.6 GyDmean = 10.0 Gy

PTV - Clinical PlanDmean = 99.3%

PTV - RTIP PlanDmean = 100.5%

Fig. 4 Final DVHs for the plan created using RTIP, shown with DVHs for the clinical plan. DVHs for the right SMG and left parotid have been omitted for clarity. Dmean = mean dose to volume, D0.03cc = max dose to 0.03 cm3 of structure.