Table of contents

Editorial

Dear colleagues,

From a professional perspective there is nothing quite as enjoyable as attending a good conference with colleagues that share not only interest for the same topic but also a similar vision for their field of work. Proceedings cannot replace the direct contact and lively conversations between participants who are at the same time at the same place sharing ideas and discussion. However, proceedings complement the actual conference by

1. giving the ideas, research findings and debates which characterise the conference lasting presence

2. giving participants an opportunity to refresh their memory about issues they would like to follow up on and

3. providing an opportunity for others who were not able to attend the meeting to share in the thoughts and issues discussed at the conference.

For the proceedings of the International Conferences on the Use of Computers in Radiation Therapy (ICCR) all this has been important in the past as the high citation rates for ICCR proceedings show. We hope that also the present proceedings will appeal to participants and others who are interested in cancer and ways to treat patients affected by it. It is exciting to note that what sounded like a small niche field (''Computers in Radiation Therapy'') has become a very broad forum to discuss all aspects of cancer diagnosis and therapy. Of particular interest for the 17th ICCR have been data and how to collect, organise and use it effectively. This is relevant for most areas in medicine and we believe that radiation therapy with its focus on evidence based practice and measurable and standardised activities has an important and possibly leading role to play.

The 100 manuscripts combined in the present proceedings can hopefully contribute to this. The proceedings are organised into five different streams which reflect the foci of the conference:

• Dose Calculation

• Imaging for treatment planning and Motion Management

• Treatment planning and optimisation

• Verification, Risk assessment and IGRT

• Trials, registries and more

Each of the streams had its own theme editor and we would like to thank Peter Greer, Sarah Gulliford, Tomas Janssen, Joerg Lehmann and Peter Metcalfe and for their invaluable contributions. I believe we all shared the joy of reading the huge variety of manuscripts which reflect a dynamic field and a truly international authorship. All manuscripts have been peer reviewed by at least two independent referees. Still, proceedings are different from journal articles. The focus of the proceedings is on interest for others, novelty and creativity. While trying to ensure a common format to all contributions we have also aimed to maintain the style and approach of the individual authors. We would like to thank all authors and reviewers for their contribution to these ICCR proceedings.

On behalf of the editorial team we would like to invite you to explore the proceedings and hope that you enjoy reading them as much as we did editing them.

Yours sincerely

Annette Haworth and Tomas Kron

Editors, Proceedings of the 17th ICCR

'ICCR 2013 in Images' - A pictorial essay containing images from the conference can be viewed in the pdf.

All papers published in this volume of Journal of Physics: Conference Series have been peer reviewed through processes administered by the proceedings Editors. Reviews were conducted by expert referees to the professional and scientific standards expected of a proceedings journal published by IOP Publishing.

While Monte Carlo (MC) simulation is believed to be the most reliable method of dose calculation in particle therapy, the simulation time is critical in attaining sufficient statistical accuracy for clinical applications. Therefore, parallelization of simulations is essential. This paper describes a common platform of MC dose calculation in grid-distributed computing environments. The platform is flexible and effective for dose calculation in both clinical and research applications for particle therapy. The platform consists of the universal grid interface (UGI) and the Geant4-based particle therapy simulation framework (PTSIM). The UGI, written in Python, provides a command-line interface for job submission, file manipulation, and monitoring in multiple-grid middleware environments. The PTSIM is a single software application for modeling a treatment port with patient data obtained from CT images. The common platform was constructed in grid computing environments using the computing resources in five institutions. The platform utilized these resources through the NAREGI grid middleware under UGI to provide stable computing resources and a common environment for MC dose calculation in particle therapy.

We introduce the concept of analytical probabilistic modeling (APM) to calculate the mean and the standard deviation of intensity-modulated proton dose distributions under the influence of range uncertainties in closed form. For APM, range uncertainties are modeled with a multivariate Normal distribution p(z) over the radiological depths z. A pencil beam algorithm that parameterizes the proton depth dose d(z) with a weighted superposition of ten Gaussians is used. Hence, the integrals ∫ dz p(z) d(z) and ∫ dz p(z) d(z)² required for the calculation of the expected value and standard deviation of the dose remain analytically tractable and can be efficiently evaluated. The means μ_k, widths δ_k, and weights ω_k of the Gaussian components parameterizing the depth dose curves are found with least squares fits for all available proton ranges. We observe less than 0.3% average deviation of the Gaussian parameterizations from the original proton depth dose curves. Consequently, APM yields high accuracy estimates for the expected value and standard deviation of intensity-modulated proton dose distributions for two dimensional test cases. APM can accommodate arbitrary correlation models and account for the different nature of random and systematic errors in fractionated radiation therapy. Beneficial applications of APM in robust planning are feasible.

Purpose: In ion beam therapy electronic stopping power data enter in different disciplines, e.g., dose planning, dosimetry, and radiobiology. However, relevant stopping power data are only known within an accuracy of 2%-10%. We started the software library project libdEdx to unify data from several well-known stopping power sources into one ready-to-use package being 1) freely available and 2) easy accessible via a web-based front end.

Methods: Currently, stopping power data from PSTAR, ASTAR, MSTAR and ICRU49+73 are implemented along with a version of the Bethe formula. The library is programmed in the language C to provide broad portability and high performance. A clean API provides full access to the underlying functions and thread safety in multi-threaded applications. The possibility to define arbitrary materials complements the list of predefined ICRU materials. Furthermore, we introduced a collection of tools, e.g., inverse stopping power look-up as well as CSDA range calculation and its inverse.

Results: On a standard desktop PC libdEdx calculates 22 million look-ups/sec. A web GUI (available at http://dedx.au.dk) provides easy access to libdEdx and download of stopping data and graphs. For compounds, we observe that stopping power data are robust for variations in the mean excitation potential of the constituents as long as the total mean excitation potential is fixated.

Conclusion: We released libdEdx (version number 1.2.1: http://sf.net/projects/libdedx/) with a web-based GUI. Future development will focus on implementing further stopping powers sources (e.g., for electrons and nuclear stopping) and relativistic effects.

Purpose: The Monte Carlo (MC) code SHIELD-HIT simulates the transport of ions through matter. Since SHIELD-HIT08 we added numerous features that improves speed, usability and underlying physics and thereby the user experience. The "-A" fork of SHIELD-HIT also aims to attach SHIELD-HIT to a heavy ion dose optimization algorithm to provide MC-optimized treatment plans that include radiobiology.

Methods: SHIELD-HIT12A is written in FORTRAN and carefully retains platform independence. A powerful scoring engine is implemented scoring relevant quantities such as dose and track-average LET. It supports native formats compatible with the heavy ion treatment planning system TRiP. Stopping power files follow ICRU standard and are generated using the libdEdx library, which allows the user to choose from a multitude of stopping power tables.

Results: SHIELD-HIT12A runs on Linux and Windows platforms. We experienced that new users quickly learn to use SHIELD-HIT12A and setup new geometries. Contrary to previous versions of SHIELD-HIT, the 12A distribution comes along with easy-to-use example files and an English manual. A new implementation of Vavilov straggling resulted in a massive reduction of computation time. Scheduled for later release are CT import and photon-electron transport.

Conclusions: SHIELD-HIT12A is an interesting alternative ion transport engine. Apart from being a flexible particle therapy research tool, it can also serve as a back end for a MC ion treatment planning system. More information about SHIELD-HIT12A and a demo version can be found on http://www.shieldhit.org.

Amorphous silicon (a-Si) electronic portal imaging devices (EPIDs) are x-ray detectors frequently used in radiotherapy imaging and dosimetry applications. EPIDs employ a copper plate and gadolinium oxysulfide phosphor screen with an array of a-Si photodiodes to indirectly detect incident radiation. In this study, a previously developed Monte Carlo (MC) model of an a-Si EPID has been extended for transit dosimetry. The GEANT4 MC toolkit was used to integrate an a-Si EPID model with two phantoms and a 6 MV x-ray source. A solid water phantom was used to simulate EPID transmission factors, field size output factors and relative dose profiles and results were compared to experimental measurements. An anthropomorphic head phantom was used to qualitatively compare simulated and measured portal images of humanoid anatomy. Calculated transmission factors and field size output factors agreed to within 2.0% and 1.9% of experimental measurements, respectively. A comparison of calculated and measured relative dose profiles yielded >98% of points passing a gamma analysis with 3%/3 mm criterion for all field sizes. The simulated anthropomorphic head phantom image shows macroscopic anatomical features and qualitatively agrees with the measured image. Results validate the suitability of the MC model for predicting EPID response in transit dosimetry.

To obtain accurate Monte Carlo simulations of small radiation fields, it is important model the initial source parameters (electron energy and spot size) accurately. However recent studies have shown that small field dosimetry correction factors are insensitive to these parameters. The aim of this work is to extend this concept to test if these parameters affect dose perturbations in general, which is important for detector design and calculating perturbation correction factors.

The EGSnrc C++ user code cavity was used for all simulations. Varying amounts of air between 0 and 2 mm were deliberately introduced upstream to a diode and the dose perturbation caused by the air was quantified. These simulations were then repeated using a range of initial electron energies (5.5 to 7.0 MeV) and electron spot sizes (0.7 to 2.2 FWHM).

The resultant dose perturbations were large. For example 2 mm of air caused a dose reduction of up to 31% when simulated with a 6 mm field size. However these values did not vary by more than 2 % when simulated across the full range of source parameters tested.

If a detector is modified by the introduction of air, one can be confident that the response of the detector will be the same across all similar linear accelerators and the Monte Carlo modelling of each machine is not required.

The purpose of this study was to systematically evaluate dose distributions computed with 5 different dose algorithms for patients with lung cancers treated using stereotactic ablative body radiotherapy (SABR). Treatment plans for 133 lung cancer patients, initially computed with a 1D-pencil beam (equivalent-path-length, EPL-1D) algorithm, were recalculated with 4 other algorithms commissioned for treatment planning, including 3-D pencil-beam (EPL-3D), anisotropic analytical algorithm (AAA), collapsed cone convolution superposition (CCC), and Monte Carlo (MC). The plan prescription dose was 48 Gy in 4 fractions normalized to the 95% isodose line. Tumors were classified according to location: peripheral tumors surrounded by lung (lung-island, N=39), peripheral tumors attached to the rib-cage or chest wall (lung-wall, N=44), and centrally-located tumors (lung-central, N=50). Relative to the EPL-1D algorithm, PTV D95 and mean dose values computed with the other 4 algorithms were lowest for "lung-island" tumors with smallest field sizes (3-5 cm). On the other hand, the smallest differences were noted for lung-central tumors treated with largest field widths (7-10 cm). Amongst all locations, dose distribution differences were most strongly correlated with tumor size for lung-island tumors. For most cases, convolution/superposition and MC algorithms were in good agreement. Mean lung dose (MLD) values computed with the EPL-1D algorithm were highly correlated with that of the other algorithms (correlation coefficient =0.99). The MLD values were found to be ~10% lower for small lung-island tumors with the model-based (conv/superposition and MC) vs. the correction-based (pencil-beam) algorithms with the model-based algorithms predicting greater low dose spread within the lungs. This study suggests that pencil beam algorithms should be avoided for lung SABR planning. For the most challenging cases, small tumors surrounded entirely by lung tissue (lung-island type), a Monte-Carlo-based algorithm may be warranted.

This study reports on Gafchromic EBT2 film skin dose measurements for lung stereotactic body radiotherapy. These measurements were compared to near surface skin doses predicted by Eclipse treatment planning system (TPS) using the Analytical Anisotropic Algorithm (AAA) for a 6 MV photon beam. The accuracy of the predicted near surface dose for 3×3, 5×5 and 10×10 cm² fields was assessed using an Attix chamber and EBT2 film in a Virtual Water phantom and compared to Monte Carlo calculation. The maximum near surface dose and its location were identified from the patient's treatment plan. For phantom measurements, the TPS dose (nominal 0 mm depth) was higher than the Attix chamber data by up to 24.0 % but in closer agreement with the EBT2 film measurement, which was up to 10.3 % higher than the Attix data. The MC calculated dose values were higher than the Attix data by up to 3.5% for the depths up to 2 mm. The maximum patient skin dose estimated from in vivo EBT2 film measurements was 7.5 – 19.5 Gy per course and depended on the number of overlapping fields, beam weight and/or contact with immobilisation devices. The TPS predicted dose for patient plans was mostly higher than the in vivo dose, by as much as 69.3%, but in two cases was lower, by as much as -23.1%.

Purpose: Exploration of historical data for regional organ dose sensitivity is limited by the effort needed to (sub-)segment large numbers of contours. A system has been developed which can rapidly perform autonomous contour sub-segmentation and generic dose-volume computations, substantially reducing the effort required for exploratory analyses.

Methods: A contour-centric approach is taken which enables lossless, reversible segmentation and dramatically reduces computation time compared with voxel-centric approaches. Segmentation can be specified on a per-contour, per-organ, or per-patient basis, and can be performed along either an embedded plane or in terms of the contour's bounds (e.g., split organ into fractional-volume/dose pieces along any 3D unit vector). More complex segmentation techniques are available. Anonymized data from 60 head-and-neck cancer patients were used to compare dose-volume computations with Varian's Eclipse^TM (Varian Medical Systems, Inc.).

Results: Mean doses and Dose-volume-histograms computed agree strongly with Varian's Eclipse^TM. Contours which have been segmented can be injected back into patient data permanently and in a Digital Imaging and Communication in Medicine (DICOM)-conforming manner. Lossless segmentation persists across such injection, and remains fully reversible.

Conclusions: DICOMautomaton allows researchers to rapidly, accurately, and autonomously segment large amounts of data into intricate structures suitable for analyses of regional organ dose sensitivity.

This work presents the CPU implementation of GMC ['gimik]: a fast yet accurate one-variable Monte Carlo dose algorithm for proton therapy to be incorporated into our in-house treatment planning system, Astroid. GMC is based on a simple mathematical model using the formulated proton scattering power and tabulated data of empirical depth-dose distributions. These Bragg peaks determine the energy deposited along the particle's track. The polar scattering angle is based on the particle's local energy and the voxel's density, while the azimuthal component of that scattering angle is the single variable in GMC, uniformly distributed from 0 to 2π. The halo effect of the beam, currently not implemented, will consider large scattering angles and secondary protons for a small percentage of the incident histories. GMC shows strong agreement with both the empirical data and GEANT4-based simulations. Its current CPU implementation runs at ~300 m.s^−-1, approximately ten times faster than GEANT4. Significant speed improvement is expected with the upcoming implementation of multi-threading and the portage to the GPU architecture. In conclusion, a one-variable Monte Carlo dose algorithm was produced for proton therapy dose computations. Its simplicity allows for fast dose computation while conserving accuracy against heterogeneities, hence drastically improving the current algorithms used in treatment planning systems.

Monte Carlo radiation transport models are increasingly being used to simulate biological damage. However, such radiation biophysics simulations require realistic molecular models for water, whereas existing Monte Carlo models are limited by their use of atomic cross-sections, which become inadequate for accurately modelling interactions of the very low-energy electrons that are responsible for biological damage. In this study, we borrow theoretical methods commonly employed in molecular dynamics simulations to model the molecular wavefunction of the water molecule as the first step towards deriving new molecular cross-sections. We calculate electron charge distributions for molecular water and find non-negligible differences between the vapor and liquid phases that can be attributed to intermolecular bonding in the condensed phase. We propose that a hybrid Monte Carlo – Molecular Dynamics (MC-MD) approach to modelling radiation biophysics will provide new insights into radiation damage and new opportunities to develop targeted molecular therapy strategies.

Previous investigators of EPID dosimetric properties have ascribed the backscatter, that contaminates dosimetric EPID images, to its supporting arm. Accordingly, Monte-Carlo (MC) EPID models have approximated the backscatter signal from the layers under the detector and the robotic support arm using either uniform or non-uniform solid water slabs, or through convolutions with back-scatter kernels. The aim of this work is to improve the existent MC models by measuring and modelling the separate backscatter contributions of the robotic arm and the rear plastic housing of the EPID. The EPID plastic housing is non-uniform with a 11.9 cm wide indented section that runs across the cross-plane direction in the superior half of the EPID which is 1.75 cm closer to the EPID sensitive layer than the rest of the housing. The thickness of the plastic housing is 0.5 cm. Experiments were performed with and without the housing present by removing all components of the EPID from the housing. The robotic support arm was not present for these measurements. A MC model of the linear accelerator and the EPID was modified to include the rear-housing indentation and results compared to the measurement. The rear housing was found to contribute a maximum of 3% additional signal. The rear housing contribution to the image is non-uniform in the in-plane direction with 2% asymmetry across the central 20 cm of an image irradiating the entire detector. The MC model was able to reproduce this non-uniform contribution. The EPID rear housing contributes a non-uniform backscatter component to the EPID image, which has not been previously characterized. This has been incorporated into an improved MC model of the EPID.

Purpose of this work was to implement Monte Carlo (MC) dose computation in realistic patient geometries with raster-scanning, the most advanced ion beam delivery technique, combining magnetic beam deflection with energy variation. FLUKA, a Monte Carlo package well-established in particle therapy applications, was extended to simulate raster-scanning delivery with clinical data, unavailable as built-in feature. A new complex beam source, compatible with FLUKA public programming interface, was implemented in Fortran to model the specific properties of raster-scanning, i.e. delivery by means of multiple spot sources with variable spatial distributions, energies and numbers of particles. The source was plugged into the MC engine through the user hook system provided by FLUKA. Additionally, routines were provided to populate the beam source with treatment plan data, stored as DICOM RTPlan or TRiP98's RST format, enabling MC recomputation of clinical plans. Finally, facilities were integrated to read computerised tomography (CT) data into FLUKA. The tool was used to recompute two representative carbon ion treatment plans, a skull base and a prostate case, prepared with analytical dose calculation (TRiP98). Selected, clinically relevant issues influencing the dose distributions were investigated: (1) presence of positioning errors, (2) influence of fiducial markers and (3) variations in pencil beam width. Notable differences in modelling of these challenging situations were observed between the analytical and Monte Carlo results. In conclusion, a tool was developed, to support particle therapy research and treatment, when high precision MC calculations are required, e.g. in presence of severe density heterogeneities or in quality assurance procedures.

This study investigates the variation of photon field penumbra shape with initial electron beam diameter, for very narrow beams. A Varian Millenium MLC (Varian Medical Systems, Palo Alto, USA) and a Brainlab m3 microMLC (Brainlab AB. Feldkirchen, Germany) were used, with one Varian iX linear accelerator, to produce fields that were (nominally) 0.20 cm across. Dose profiles for these fields were measured using radiochromic film and compared with the results of simulations completed using BEAMnrc and DOSXYZnrc, where the initial electron beam was set to FWHM = 0.02, 0.10, 0.12, 0.15, 0.20 and 0.50 cm. Increasing the electron-beam FWHM produced increasing occlusion of the photon source by the closely spaced collimator leaves and resulted in blurring of the simulated profile widths from 0.24 to 0.58 cm, for the MLC, from 0.11 to 0.40 cm, for the microMLC. Comparison with measurement data suggested that the electron spot size in the clinical linear accelerator was between FWHM = 0.10 and 0.15 cm, encompassing the result of our previous output-factor based work, which identified a FWHM of 0.12 cm. Investigation of narrow-beam penumbra variation has been found to be a useful procedure, with results varying noticeably with linear accelerator spot size and allowing FWHM estimates obtained using other methods to be verified.

Different treatment planning system (TPS) algorithms calculate radiation dose in different ways. This work compares measurements made in vivo to the dose calculated at out-of-field locations using three different commercially available algorithms in the Eclipse treatment planning system. LiF: Mg, Cu, P thermoluminescent dosimeter (TLD) chips were placed with 1 cm build-up at six locations on the contralateral side of 5 patients undergoing radiotherapy for breast cancer. TLD readings were compared to calculations of Pencil Beam Convolution (PBC), Anisotropic Analytical Algorithm (AAA) and Acuros XB (XB). AAA predicted zero dose at points beyond 16 cm from the field edge. In the same region PBC returned an unrealistically constant result independent of distance and XB showed good agreement to measured data although consistently underestimated by ~0.1 % of the prescription dose. At points closer to the field edge XB was the superior algorithm, exhibiting agreement with TLD results to within 15 % of measured dose. Both AAA and PBC showed mixed agreement, with overall discrepancies considerably greater than XB. While XB is certainly the preferable algorithm, it should be noted that TPS algorithms in general are not designed to calculate dose at peripheral locations and calculation results in such regions should be treated with caution.

Purpose: Accurate on-line reconstruction of in-vivo volume dose that accounts for both machine and patient discrepancy is not clinically available. We present a simple reference-dose-perturbation algorithm that reconstructs in-vivo volume dose fast and accurately. Methods: We modelled the volume dose as a function of the fluence map and density image. Machine (output variation, jaw/leaf position errors, etc.) and patient (setup error, weight loss, etc.) discrepancies between the plan and delivery were modelled as perturbation of the fluence map and density image, respectively. Delivered dose is modelled as perturbation of the reference dose due to change of the fluence map and density image. We used both simulated and clinical data to validate the algorithm. The planned dose was used as the reference. The reconstruction was perturbed from the reference and accounted for output-variations and the registered daily image. The reconstruction was compared with the ground truth via isodose lines and the Gamma Index. Results: For various plans and geometries, the volume doses were reconstructed in few seconds. The reconstruction generally matched well with the ground truth. For the 3%/3mm criteria, the Gamma pass rates were 98% for simulations and 95% for clinical data. The differences mainly appeared on the surface of the phantom/patient. Conclusions: A novel reference-dose-perturbation dose reconstruction model is presented. The model accounts for machine and patient discrepancy from planning. The algorithm is simple, fast, yet accurate, which makes online in-vivo 3D dose reconstruction clinically feasible.

Open-Source Medical Devices (OSMD) was initiated with the goal of facilitating medical research by developing medical technologies including both hardware and software on an open-source platform. Our first project was to develop an integrated imaging and radiotherapy device for small animals that includes computed tomography (CT), positron emission tomography (PET) and radiation therapy (RT) modalities for which technical specifications were defined in the first OSMD conference held in Madison, Wisconsin, USA in December 2011. This paper specifically focuses on the development of a small animal RT (micro-RT) system by designing a binary micro multileaf collimator (bmMLC) and a small animal treatment planning system (SATPS) to enable intensity modulated RT (IMRT). Both hardware and software projects are currently under development and their current progresses are described. After the development, both bmMLC and TPS will be validated and commissioned for a micro-RT system. Both hardware design and software development will be open-sourced after completion.

Microbeam radiation therapy (MRT) is a new experimental oncological modality, intended for the treatment of inoperable brain tumours, particularly in difficult cases where conventional radiation therapy can cause irreversible damage. MRT consists of an array of highly collimated, quasi-parallel x-ray microbeams aimed at the tumour tissue, delivering high dose within the beam path and low doses in regions between the beams. For reasons still not fully understood, healthy tissue exposed to the microbeam array is able to regenerate while tumour volumes are significantly reduced. Low energy Monte Carlo radiative transport simulations provide new insight into understanding the underlying mechanisms of MRT. In particular, predicting the ionisation cluster distribution, which is a significant cause of lethal damage to cells, would provide insight into the biological responses. Geant4-DNA was used to model an x-ray microbeam of width 20 μm in liquid water. Secondary electrons, predominately responsible for ionisation clustering, were tracked to predict damage to cells within and adjacent to the beams. We find that higher energy beams (100 keV) produce less secondary electrons in the regions outside the beam than low energy beams (30-50 keV).

Purpose: To improve the accuracy of convolution/superposition (C/S) in heterogeneous material by developing a new algorithm: heterogeneity compensated superposition (HCS).

Methods: C/S has proven to be a good estimator of the dose deposited in a homogeneous volume. However, near heterogeneities electron disequilibrium occurs, leading to the faster fall-off and re-buildup of dose. We propose to filter the actual patient density in a position and direction sensitive manner, allowing the dose deposited near interfaces to be increased or decreased relative to C/S. We implemented the effective density function as a multivariate first-order recursive filter and incorporated it into GPU-accelerated, multi-energetic C/S implementation. We compared HCS against C/S using the ICCR 2000 Monte-Carlo accuracy benchmark, 23 similar accuracy benchmarks and 5 patient cases.

Results: Multi-energetic HCS increased the dosimetric accuracy for the vast majority of voxels; in many cases near Monte-Carlo results were achieved. We defined the per-voxel error, %|mm, as the minimum of the distance to agreement in mm and the dosimetric percentage error relative to the maximum MC dose. HCS improved the average mean error by 0.79 %|mm for the patient volumes; reducing the average mean error from 1.93 %|mm to 1.14 %|mm. Very low densities (i.e. < 0.1 g / cm3) remained problematic, but may be solvable with a better filter function.

Conclusions: HCS improved upon C/S's density scaled heterogeneity correction with a position and direction sensitive density filter. This method significantly improved the accuracy of the GPU based algorithm reaching the accuracy levels of Monte Carlo based methods with performance in a few tenths of seconds per beam.

Acknowledgement: Funding for this research was provided by the NSF Cooperative Agreement EEC9731748, Elekta / IMPAC Medical Systems, Inc. and the Johns Hopkins University. James Satterthwaite provided the Monte Carlo benchmark simulations.

Intensity Modulated Radiation Therapy (IMRT) treatments are some of the most complex being delivered by modern megavoltage radiotherapy accelerators. Therefore verification of the dose, or the presecribed Monitor Units (MU), predicted by the planning system is a key element to ensuring that patients should receive an accurate radiation dose plan during IMRT. One inherently accurate method is by comparison with absolute calibrated Monte Carlo simulations of the IMRT delivery by the linac head and corresponding delivery of the plan to a patient based phantom. In this work this approach has been taken using BEAMnrc for simulation of the treatment head, and both DOSXYZnrc and Geant4 for the phantom dose calculation. The two Monte Carlo codes agreed to within 1% of each other, and these matched very well to our planning system for IMRT plans to the brain, nasopharynx, and head and neck.

TOPAS (TOol for PArticle Simulation) is a Monte Carlo particle transport tool being released to a wide variety of proton therapy users worldwide. Because TOPAS provides unprecedented ease in 4D placement of geometry components, beam sources and scoring, including options to place geometry components, beam sources or scorers within each other, Quality Control (QC) for TOPAS is both critical and challenging. All simulation details (geometry, particle sources, scoring, physics settings, time-dependent motions, gating, etc.) are specified in the TOPAS Parameter Control System (which catches many user errors). QC includes Unit and End-to-End Testing. Each code unit is tested (each geometry component, particle source option, scoring option, etc.) and these unit testing procedures are shared with end users so they can reproduce tests. End-to-End testing of several full clinical setups is routinely performed. End-to-End testing presents a challenge since one cannot anticipate all the ways users will combine TOPAS flexible units for their specific project. Automated checking catches geometry overlaps and some other problematic setups, but one can never rule out the potential for problems when users combine units in new setups. QC is ultimately a partnership between the tool developer and the user. Key is that the developer be clear to the end user about what has been tested and what has not.

An investigation was conducted to determine the impact of two carbon fibre couch tops on treatment planning and to establish a method by which these couch tops may be accounted for in the XiO treatment planning system. In this study, the effects of the Civco Universal and the Medical Intelligence iBeam evo carbon fibre couch tops were investigated. Customised templates were developed in Focal to mimic the effects of these patient support devices and each template was then included in calculations performed by the XiO treatment planning system. The accuracy of modelling the couch tops in this manner was investigated and it was shown that, while using the customised couch templates, XiO modelled the increase in the surface dose due to treating through the couch tops at 180° to 1.0% ± 1.7%. The attenuation due to the presence of the couches was modelled to within 0.5% ± 0.4% for angles that pass through the flat central region of the couch tops and to within 1.4% ± 1.2% for angles that passed through their rapidly varying edges. When the couch templates were not included in XiO, the calculated dose at depth was recorded to be, in some cases, up to 5.3% ± 1.0% more than corresponding measured values. It was concluded that using the method described in this study it is possible to accurately model the effect of carbon fibre couch tops in the XiO treatment planning system.

A graphite calorimeter has been developed as a Japanese primary standard of absorbed dose to water in the high-energy photon beams from a clinical linac. To obtain conversion factors for the graphite calorimeter, the beam characteristics of the high-energy photon beams from the clinical linac at National Metrology Institute of Japan were calculated with the EGS5 Monte Carlo simulation code. To run the EGS5 code on High Performance Computing machines that have more than 1000 CPU cores, we developed the EGS5 parallelisation package "EGS5-MPI" by implementing a message-passing interface. We calculated the photon energy spectra, which are in good agreement with those previously calculated by D. Sheikh-Bagheri and D. W. O. Rogers (Med. Phys.29 3). We also estimated the percentage-depth-dose distributions of photon beams from the linac using the calculated photon energy spectra. These calculated percentage-depth-dose distributions were compared with our measured distributions and were found they are in good agreement as well. We will calculate conversion factors for the graphite calorimeter using our results.

Motion artifacts appear in four-dimensional computed tomography (4DCT) images because of suboptimal acquisition parameters or patient breathing irregularities. Frequency of motion artifacts is high and they may introduce errors in radiation therapy treatment planning. Motion artifact detection can be useful for image quality assessment and 4D reconstruction improvement but manual detection in many images is a tedious process. We propose a novel method to evaluate the quality of 4DCT images by automatic detection of motion artifacts. The method was used to evaluate the impact of the optimization of acquisition parameters on image quality at our institute. 4DCT images of 114 lung cancer patients were analyzed. Acquisitions were performed with a rotation period of 0.5 seconds and a pitch of 0.1 (74 patients) or 0.081 (40 patients). A sensitivity of 0.70 and a specificity of 0.97 were observed. End-exhale phases were less prone to motion artifacts. In phases where motion speed is high, the number of detected artifacts was systematically reduced with a pitch of 0.081 instead of 0.1 and the mean reduction was 0.79. The increase of the number of patients with no artifact detected was statistically significant for the 10%, 70% and 80% respiratory phases, indicating a substantial image quality improvement.

We evaluated automatic three-dimensional intensity-based rigid registration (RR) methods for prostate localization on CBCT scans and studied the impact of rectum distension on registration quality. 106 CBCT scans of 9 prostate patients were used. Each one was registered to the planning computed tomography (CT) scan using different methods: (a) global registration, (b) pelvis bony structure registration, (c) bony registration refined by a local prostate registration using the CT clinical target volume (CTV) expanded with 1, 3, 5, 8, 10, 12, 15 or 20-mm margin. Automatic CBCT contours were generated after propagation of the manual CT contours. To evaluate results, a radiation oncologist was asked to manually delineate the CTV on the CBCT scans (gold standard). The Dice similarity coefficients between propagated and manual CBCT contours were calculated.

Purpose: To estimate in-room breathing motion from a limited number of 2D cone-beam (CB) projection images by registering them to a phase of the 4D planning CT.

Methods: Breathing motion was modelled using a piecewise continuous B-spline representation [1], allowing to preserve the sliding along the thoracic wall while limiting the degrees of freedom. The deformed target 3D image was subsequently used to generate Digitally Reconstructed Radiographs (DRR). The Normalized Correlation Coefficient (NCC) between the measured projection images and the DRR was computed in the 2D projection space. However, the partial derivatives of the NCC relative to the transform parameters were backprojected into the 3D space, avoiding the projection of the transform Jacobian matrix which is computationally intractable [2].

Results: The method was quantitatively evaluated on 16 lung cancer patients. 40 CB projection images were simulated using the end-exhale phase of the 4D planning CT and the geometric parameters of a clinical CB protocol. The end-inhale phase was deformed to match these simulated projections. The Target Registration Error (TRE) decreased from 8.8 mm to 2.0 mm while the TRE obtained from the 3D/3D registration of the reconstructed CBCT was significantly worse (2.6 mm), due to view aliasing artefacts. We also provide the motion compensated image reconstructed from a real CB acquisition showing the quality improvement brought by the in-room deformation model compared to the planning motion model.

Conclusions: We have developed a 2D/3D deformable registration algorithm that enables in-room breathing motion estimation from cone-beam projection images.

The analysis of intra-fraction organ motion is important for improving the precision of radiation therapy treatment delivery. One method to quantify this motion is for one or more observers to manually identify anatomic points of interest (POIs) on each slice of a cine-MRI sequence. However this is labour intensive and inter- and intra- observer variation can introduce uncertainty. In this paper a fast method for non-rigid registration based point tracking in cine-MRI sagittal and coronal series is described which identifies POIs in 0.98 seconds per sagittal slice and 1.35 seconds per coronal slice. The manual and automatic points were highly correlated (r>0.99, p<0.001) for all organs and the difference generally less than 1mm. For prostate planning peristalsis and rectal gas can result in unpredictable out of plane motion, suggesting the results may require manual verification.

Despite radical treatment with surgery, radiotherapy and chemotherapy, advanced gliomas recur within months. Geographic misses in radiotherapy planning may play a role in this seemingly ineluctable recurrence. Planning is typically performed on post-contrast MRIs, which are known to underreport tumour volume relative to FDOPA PET scans. FDOPA PET fused with contrast enhanced MRI has demonstrated greater sensitivity and specificity than MRI alone. One sign of potential misses would be differences between gross target volumes (GTVs) defined using MRI alone and when fused with PET. This work examined whether such a discrepancy may occur. Materials and Methods: For six patients, a 75 minute PET scan using 3,4-dihydroxy-6-18F-fluoro-L-phynel-alanine (18F-FDOPA) was taken within 2 days of gadolinium enhanced MRI scans. In addition to standard radiotherapy planning by an experienced radiotherapy oncologist, a second gross target volume (GTV) was defined by an experienced nuclear medicine specialist for fused PET and MRI, while blinded to the radiotherapy plans. The volumes from standard radiotherapy planning were compared to the PET defined GTV. Results: The comparison indicated radiotherapy planning would change in several cases if FDOPA PET data was available. PET-defined contours were external to 95% prescribed dose for several patients. However, due to the radiotherapy margins, the discrepancies were relatively small in size and all received a dose of 50 Gray or more. Conclusions: Given the limited size of the discrepancies it is uncertain that geographic misses played a major role in patient outcome. Even so, the existence of discrepancies indicates that FDOPA PET could assist in better defining margins when planning radiotherapy for advanced glioma, which could be important for highly conformal radiotherapy plans.

Analysing osteolytic and osteoblastic bone lesions in systematically affected skeletons, e.g. in multiple myeloma or bone metastasis, is a complex task. Quantification of the degree of bone destruction needs segmentation of all lesions but cannot be managed manually. Automatic bone lesion detection is necessary. Our future objective is comparing modified bones with healthy shape models. For applying model based strategies successfully, identification and position information of single bones is necessary. A solution to these requirements based on bone medullary cavities is presented in this paper.

Medullary cavities are useful for shape model positioning since they have similar position and orientation as the bone itself but can be separated more easily. Skeleton segmentation is done by simple thresholding. Inside the skeleton medullary cavities are segmented by a flood filling algorithm. The filled regions are considered as medullary cavity objects. To provide automatic shape model selection, medullary cavity objects are assigned to bone structures with pattern recognition. To get a good starting position for shape models, principal component analysis of medullary cavities is performed. Bone identification was tested on 14 whole-body low-dose CT scans of multiple myeloma patients.

Random forest classification assigns medullary cavities of long bones to the corresponding bone (overall accuracy 90%). Centroid and first principal component of medullary cavity are sufficiently similar to those of bone (mean centroid difference 21.7 mm, mean difference angle 1.54° for all long bones of one example patient) and therefore suitable for shape model initialization.

This method enables locating long bone structures in whole-body CT scans and provides useful information for a reasonable shape model initialization.

Realistic modelling of breast deformation requires the breast tissue to be segmented into fibroglandular and fatty tissue and assigned suitable material properties. There are a number of breast tissue segmentation methods proposed and used in the literature. The purpose of this study was to validate and compare the accuracy of various segmentation methods and to investigate the effect of the tissue distribution on the segmentation accuracy. Computed tomography (CT) data for 24 patients, both in supine and prone positions were segmented into fibroglandular and fatty tissue. The segmentation methods explored were: physical density thresholding; interactive thresholding; fuzzy c-means clustering (FCM) with three classes (FCM3) and four classes (FCM4); and k-means clustering. Validation was done in two-stages: firstly, a new approach, supine-prone validation based on the assumption that the breast composition should appear the same in the supine and prone scans was used. Secondly, outlines from three experts were used for validation. This study found that FCM3 gave the most accurate segmentation of breast tissue from CT data and that the segmentation accuracy is adversely affected by the sparseness of the fibroglandular tissue distribution.

Purpose: To modify the single-threshold parametric response map (ST-PRM) method for predicting treatment outcomes in order to facilitate its use for guidance of adaptive dose painting in intensity-modulated radiotherapy.

Methods: Multiple graded thresholds were used to extend the ST-PRM method (Nat. Med. 2009;15(5):572-576) such that the full functional change distribution within tumours could be represented with respect to multiple confidence interval estimates for functional changes in similar healthy tissue. The ST-PRM and graded-threshold PRM (GT-PRM) methods were applied to functional imaging scans of 5 patients treated for hepatocellular carcinoma. Pre and post-radiotherapy arterial blood flow maps (ABF) were generated from CT-perfusion scans of each patient. ABF maps were rigidly registered based on aligning tumour centres of mass. ST-PRM and GT-PRM analyses were then performed on overlapping tumour regions within the registered ABF maps.

Main findings: The ST-PRMs contained many disconnected clusters of voxels classified as having a significant change in function. While this may be useful to predict treatment response, it may pose challenges for identifying boost volumes or for informing dose-painting by numbers strategies. The GT-PRMs included all of the same information as ST-PRMs but also visualized the full tumour functional change distribution. Heterogeneous clusters in the ST-PRMs often became more connected in the GT-PRMs by voxels with similar functional changes.

Conclusions: GT-PRMs provided additional information which helped to visualize relationships between significant functional changes identified by ST-PRMs. This may enhance ST-PRM utility for guiding adaptive dose painting.

Purpose: To investigate the effects of registration error (RE) on parametric response map (PRM) analysis of pre and post-radiotherapy (RT) functional images.

Methods: Arterial blood flow maps (ABF) were generated from the CT-perfusion scans of 5 patients with hepatocellular carcinoma. ABF values within each patient map were modified to produce seven new ABF maps simulating 7 distinct post-RT functional change scenarios. Ground truth PRMs were generated for each patient by comparing the simulated and original ABF maps. Each simulated ABF map was then deformed by different magnitudes of realistic respiratory motion in order to simulate RE. PRMs were generated for each of the deformed maps and then compared to the ground truth PRMs to produce estimates of RE-induced misclassification.

Main findings: The percentage of voxels misclassified as decreasing, no change, and increasing, increased with RE For all patients, increasing RE was observed to increase the number of high post-RT ABF voxels associated with low pre-RT ABF voxels and vice versa. 3 mm of average tumour RE resulted in 18-45% tumour voxel misclassification rates.

Conclusions: RE induced misclassification posed challenges for PRM analysis in the liver where registration accuracy tends to be lower. Quantitative understanding of the sensitivity of the PRM method to registration error is required if PRMs are to be used to guide radiation therapy dose painting techniques.

The purpose of this study was to test the hypothesis that audiovisual (AV) biofeedback can improve image quality and reduce scan time for respiratory-gated 3D thoracic MRI. For five healthy human subjects respiratory motion guidance in MR scans was provided using an AV biofeedback system, utilizing real-time respiratory motion signals. To investigate the improvement of respiratory-gated 3D MR images between free breathing (FB) and AV biofeedback (AV), each subject underwent two imaging sessions. Respiratory-related motion artifacts and imaging time were qualitatively evaluated in addition to the reproducibility of external (abdominal) motion. In the results, 3D MR images in AV biofeedback showed more anatomic information such as a clear distinction of diaphragm, lung lobes and sharper organ boundaries. The scan time was reduced from 401±215 s in FB to 334±94 s in AV (p-value 0.36). The root mean square variation of the displacement and period of the abdominal motion was reduced from 0.4±0.22 cm and 2.8±2.5 s in FB to 0.1±0.15 cm and 0.9±1.3 s in AV (p-value of displacement <0.01 and p-value of period 0.12). This study demonstrated that audiovisual biofeedback improves image quality and reduces scan time for respiratory-gated 3D MRI. These results suggest that AV biofeedback has the potential to be a useful motion management tool in medical imaging and radiation therapy procedures.

We propose a method to build a fully deformable motion model directly from cone-beam CT (CBCT) projections. This allows inter-fraction variations in the respiratory motion to be accounted for. It is envisaged that the model be used to track the tumour, and monitor organs at risk (OAR), during gated or tracked radiotherapy (RT) treatment of lung cancer. The method is tested on CBCT projections from a simulated phantom in two cases. The simulations are generated from a patient respiratory trace and associated CBCT scanner geometry. Without and with motion correction, l² norm maximum errors were reduced from 24.5 to 0.698 mm in case 1, and 20.0 to 0.101 mm in case 2, respectively.

The pharmacokinetic behavior of iodine contrast agents makes it difficult to achieve significant enhancement during contrast-enhanced cone-beam CT (CE-CBCT). This study modeled this dynamic behavior to optimize CE-CBCT and improve the localization of liver lesions for SBRT. We developed a model that allows for controlled study of changing iodine concentrations using static phantoms. A projection database consisting of multiple phantom images of differing iodine/scan conditions was built. To reconstruct images of dynamic hepatic concentrations, hepatic contrast enhancement data from conventional CT scans were used to re-assemble the projections to match the expected amount of contrast. In this way the effect of various parameters on image quality was isolated, and using our dynamic model we found parameters for iodine injection, CBCT scanning, and injection/scanning timing which optimize contrast enhancement. Increasing the iodine dose, iodine injection rate, and imaging dose led to significant increases in signal-to-noise ratio (SNR). Reducing the CBCT imaging time also increased SNR, as the image can be completed before the iodine exits the liver. Proper timing of image acquisition played a significant role, as a 30 second error in start time resulted in a 40% SNR decrease. The effect of IV contrast is severely degraded in CBCT, but there is promise that, with optimization of the injection and scan parameters to account for iodine pharmacokinetics, CE-CBCT which models venous-phase blood flow kinetics will be feasible for accurate localization of liver lesions.

Improvement in the positional accuracy of irradiation is expected by capturing patient motion (intra-fractional error) during irradiation. The present study reports the construction of a patient observation system using Microsoft® KINECT^TM. By tracking movement, we made it possible to add a depth component to the acquired position coordinates and to display three-axis (X, Y, and Z) movement. Moreover, the developed system can be displayed in a graph which is constructed from the coordinate position at each time interval. Using the developed system, an observer can easily visualize patient movement. When the body phantom was moved a known distance in the X, Y, and Z directions, good coincidence was shown with each axis. We built a patient observation system which captures a patient's motion using KINECT^TM.

We report results of an intensity-based 2D-3D rigid registration framework for patient positioning and monitoring during brain radiotherapy. We evaluated two intensity-based similarity measures, the Pearson Correlation Coefficient (ICC) and Maximum Likelihood with Gaussian noise (MLG) derived from the statistics of transmission images. A useful image frequency band was identified from the bone-to-no-bone ratio. Validation was performed on gold-standard data consisting of 3D kV CBCT scans and 2D kV radiographs of an anthropomorphic head phantom acquired at 23 different poses with parameter variations along six degrees of freedom. At each pose, a single limited field of view kV radiograph was registered to the reference CBCT. The ground truth was determined from markers affixed to the phantom and visible in the CBCT images. The mean (and standard deviation) of the absolute errors in recovering each of the six transformation parameters along the x, y and z axes for ICC were φ_x: 0.08(0.04)°, φ_y: 0.10(0.09)°, φ_z: 0.03(0.03)°, t_x: 0.13(0.11) mm, t_y: 0.08(0.06) mm and t_z: 0.44(0.23) mm. For MLG, the corresponding results were φ_x: 0.10(0.04)°, φ_y: 0.10(0.09)°, φ_z: 0.05(0.07)°, t_x: 0.11(0.13) mm, t_y: 0.05(0.05) mm and t_z: 0.44(0.31) mm. It is feasible to accurately estimate all six transformation parameters from a 3D CBCT of the head and a single 2D kV radiograph within an intensity-based registration framework that incorporates the physics of transmission images.

To enhance the measurements of radio-opaque cylindrical fiducial markers in low contrast x-ray and fluoroscopic images, a novel nonlinear marker enhancement filter (MEF) has been designed. It was primarily developed to assist in automatic initialization of a tracking procedure for intra-fraction organ motion analysis in fluoroscopic sequences. Conventional procedures were not able to provide sufficient improvement due to the complications of noise, small marker size, cylindrical shape and multiple orientations, intensity variations of the background, and the presence of overlaying anatomical measurements in this application. The proposed MEF design is based on the principles of linear scale space. It includes measures that assess the probability of each pixel to belong to a marker measurement, morphological operations, and a novel contrast enhancement function for standardization of the filter output. The MEF was tested on fluoroscopic images of two phantoms and three prostate patients, and was shown to perform better or comparable to the existing filters in terms of marker enhancement and background suppression, while performing significantly better in marker shape preservation.

MRI-alone treatment planning and adaptive MRI-based prostate radiation therapy are two promising techniques that could significantly increase the accuracy of the curative dose delivery processes while reducing the total radiation dose. State-of-the-art methods rely on the registration of a patient MRI with a MR-CT atlas for the estimation of pseudo-CT [5]. This atlas itself is generally created by registering many CT and MRI pairs. Most registration methods are not symmetric, but the order of the images influences the result [8]. The computed transformation is therefore biased, introducing unwanted variability. This work examines how much a symmetric algorithm improves the registration.

Methods: A robust symmetric registration algorithm is proposed that simultaneously optimises a half space transform and its inverse. During the registration process, the two input volumetric images are transformed to a common position in space, therefore minimising any computational bias. An asymmetrical implementation of the same algorithm was used for comparison purposes.

Results: Whole pelvis MRI and CT scans from 15 prostate patients were registered, as in the creation of MR-CT atlases. In each case, two registrations were performed, with different input image orders, and the transformation error quantified. Mean residuals of 0.63±0.26 mm (translation) and (8.7±7.3) × 10^−-3 rad (rotation) were found for the asymmetrical implementation with corresponding values of 0.038±0.039 mm and (1.6 ± 1.3) × 10^−-3 rad for the proposed symmetric algorithm, a substantial improvement.

Conclusions: The increased registration precision will enhance the generation of pseudo-CT from MRI for atlas based MR planning methods.

This paper investigates the feasibility and accuracy of tracking the motion of an intruding organ-at-risk (OAR) at the edges of a treatment field using a local optical flow analysis of electronic portal images. An intruding OAR was simulated by modifying the portal images obtained by irradiating a programmable phantom's lung tumour. A rectangular treatment aperture was assumed and the edges of the beam's eye view (BEV) were partitioned into clusters/grids according to the width of the multi-leaf collimators (MLC). The optical flow velocities were calculated and the motion accuracy in these clusters was analysed. A velocity error of 0.4 ± 1.4 mm/s with a linearity of 1.04 for tracking an object intruding at 10mm/s (max) was obtained.

Free-breathing respiratory gated SBRT of surgically inoperable lung cancer has been clinically commissioned. This study was to establish the tumour tracking accuracy under clinical conditions based on an implanted fiducial marker. A Visicoil^TM marker embedded in tissue-equivalent material mounted in a phantom (ET Gating Phantom^TM Brainlab) driven by a patient's breathing data was treated with the ExacTrac^TM system. This one-dimensional moving marker represented a tumour motion in superior-inferior (S-I) direction measured through 4DCT study of the same patient. Both Gafchromic^TM films and the stereoscopic kV images were used for tracking the position of the marker. For tumour motion at magnitudes of 10, 20 and 29 mm and treated with corresponding gate widths of 50%, 33% and 20% of free breathing amplitude, the implanted marker was able to be tracked with a deviation ≤1.53 mm to its planned position.

The purpose of this study was to develop and experimentally verify a patient specific model for simulating the interplay effect in a DMLC based IMRT delivery. A computational model was developed using MATLAB program to incorporate the interplay effect in a 2D beams eye view fluence of dynamic IMRT fields. To simulate interplay effect, the model requires two inputs: IMRT field (DMLC file with dose rate and MU) and the patient specific respiratory motion. The interplay between the DMLC leaf motion and target was simulated for three lung patients. The target trajectory data was acquired using RPM system during the treatment simulation. The model was verified experimentally for the same patients using Imatrix 2D array device placed over QUASAR motion platform in CL2100 linac. The simulated fluences and measured fluences were compared with the TPS generated static fluence (no motion) using an in-house developed gamma evaluation program (2%/2mm). The simulated results were well within agreement with the measured. Comparison of the simulated and measured fluences with the TPS static fluence resulted 55.3% & 58.5% pixels passed the gamma criteria. A patient specific model was developed and validated for simulating the interplay effect in the dynamic IMRT delivery. This model can be clinically used to quantify the dosimetric uncertainty due to the interplay effect prior to the treatment delivery.

We demonstrate acceleration on graphic processing units (GPU) of automatic identification of robust particle therapy beam setups, minimizing negative dosimetric effects of Bragg peak displacement caused by treatment-time patient positioning errors. Our particle therapy research toolkit, RobuR, was extended with OpenCL support and used to implement calculation on GPU of the Port Homogeneity Index, a metric scoring irradiation port robustness through analysis of tissue density patterns prior to dose optimization and computation. Results were benchmarked against an independent native CPU implementation. Numerical results were in agreement between the GPU implementation and native CPU implementation. For 10 skull base cases, the GPU-accelerated implementation was employed to select beam setups for proton and carbon ion treatment plans, which proved to be dosimetrically robust, when recomputed in presence of various simulated positioning errors. From the point of view of performance, average running time on the GPU decreased by at least one order of magnitude compared to the CPU, rendering the GPU-accelerated analysis a feasible step in a clinical treatment planning interactive session. In conclusion, selection of robust particle therapy beam setups can be effectively accelerated on a GPU and become an unintrusive part of the particle therapy treatment planning workflow. Additionally, the speed gain opens new usage scenarios, like interactive analysis manipulation (e.g. constraining of some setup) and re-execution. Finally, through OpenCL portable parallelism, the new implementation is suitable also for CPU-only use, taking advantage of multiple cores, and can potentially exploit types of accelerators other than GPUs.

Beam angle selection (BAS) including fluence optimization (FO) is among the most extensive computational tasks in radiotherapy. Precomputed dose influence data (DID) of all considered beam orientations (up to 100 GB for complex cases) has to be handled in the main memory and repeated FOs are required for different beam ensembles. In this paper, the authors describe concepts accelerating FO for BAS algorithms using off-the-shelf multiprocessor workstations. The FO runtime is not dominated by the arithmetic load of the CPUs but by the transportation of DID from the RAM to the CPUs. On multiprocessor workstations, however, the speed of data transportation from the main memory to the CPUs is non-uniform across the RAM; every CPU has a dedicated memory location (node) with minimum access time. We apply a thread node binding strategy to ensure that CPUs only access DID from their preferred node. Ideal load balancing for arbitrary beam ensembles is guaranteed by distributing the DID of every candidate beam equally to all nodes. Furthermore we use a custom sorting scheme of the DID to minimize the overall data transportation. The framework is implemented on an AMD Opteron workstation. One FO iteration comprising dose, objective function, and gradient calculation takes between 0.010 s (9 beams, skull, 0.23 GB DID) and 0.070 s (9 beams, abdomen, 1.50 GB DID). Our overall FO time is < 1 s for small cases, larger cases take ~ 4 s. BAS runs including FOs for 1000 different beam ensembles take ~ 15–70 min, depending on the treatment site. This enables an efficient clinical evaluation of different BAS algorithms.

Purpose: Only very few treatment planning systems (TPS) are capable of handling heavy ions. Commercial heavy ion TPS are costly and normally restrict the possibility to implement new functionalities. PyTRiP provides Python bindings and a platform-independent graphical user interface (GUI) for the heavy ion treatment program TRiP, and adds seamless support of DICOM files. We aim to provide a front-end for TRiP which does not require any special computer skills.

Methods: PyTRiP is written in Python combined with C for fast computing. Routines for DICOM file import/export to TRiPs native file format are implemented. The GUI comes as an executable with all its dependencies including PyTRiP making it easy to install on Windows, Mac and Linux.

Results: PyTRiP is a comprehensive toolbox for handling TRiP. Treatment plans are handled using an object oriented structure. Bindings to TRiP (which only runs on Linux, either locally or on a remote server) are performed through a single function call. GUI users can intuitively create treatment plans without much knowledge about the TRiP user interface. Advanced users still have full access to all TRiP functionality. The user interface comes with a comprehensive plotting tool, which can visualize 2D contours, volume histograms, as well as dose- and linear energy transfer (LET) distributions.

Conclusion: We developed a powerful toolbox for ion therapy research using TRiP as backend. The corresponding GUI allows to easily and intuitively create, calculate and visualize treatment plans. TRiP is thereby more accessible and simpler to use.

Background and purpose: Treatment of the contralateral neck after previous ipsilateral intensity modulated radiation therapy (IMRT) for head and neck cancer is a challenging problem. We have developed a technique that limits the cumulative dose to the spinal cord and brainstem while maximizing coverage of a planning target volume (PTV) in the contralateral neck. Our case involves a patient with right tonsil carcinoma who was given ipsilateral IMRT with 70Gy in 35 fractions (Plan A). A left neck recurrence was detected 14 months later. The patient underwent a neck dissection followed by postoperative left neck radiation to a dose of 66 Gy in 33 fractions (Plan B).

Materials and Methods: The spinal cord-brainstem margin (SCBM) was defined as the spinal cord and brainstem with a 1.0 cm margin. Plan A was recalculated on the postoperative CT scan but the fluence outside of SCBM was deleted. A further modification of Plan A resulted in a base plan that was summed with Plan B to evaluate the cumulative dose received by the spinal cord and brainstem. Plan B alone was used to evaluate for coverage of the contralateral neck PTV.

Results: The maximum cumulative doses to the spinal cord with 0.5cm margin and brainstem with 0.5cm margin were 51.96 Gy and 45.60 Gy respectively. For Plan B, 100% of the prescribed dose covered 95% of PTVb1.

Conclusion: The use of a modified ipsilateral IMRT plan as a base plan is an effective way to limit the cumulative dose to the spinal cord and brainstem while enabling coverage of a PTV in the contralateral neck.

This study examines the Poisson tumour control probability (TCP) γ₃₇ and D₃₇ parameters of a uniformly irradiated numerical tumour model using changes in tumour burden as a surrogate for treatment response information. An optimum dose D_i for a tumour sub-volume element V_i is described that maximizes TCP as a function of fixed tumour integral dose ξ. TCP was calculated for spatially-varying clonogen density for a total 10⁸ cells and radiosensitivity α with mean radiosensitivity in the range 0.4 – 1.0 Gy⁻¹. A bivariate normal distribution is used to describe the radiosensitivity α and the linear term of the linear-quadratic (LQ) cell kill governed the changes in the regional tumour burden within sub-volumes V_i. The optimum dose distribution, D_i, for V_i is obtained as a function of fixed tumour integral dose ξ. For a uniform dose delivery and for TCP = 37%, γ₃₇ and D₃₇ are described by the effective radiosensitivity α_eff and the effective clonogen number N_0,eff, respectively. α_eff is equivalent to differential dose changes in the number of clonogenic cells (tumour burden). The γ₃₇ values were found to be inversely correlated with variance of the probability density function of the α distribution. For the biologically optimum dose distribution, γ₃₇ was found to converge to the theoretical maximum limit and D₃₇ was found to reduce relative to that obtained for the uniform dose case. The TCP parameters γ₃₇ and D₃₇ could thus be useful in optimising individual radiation treatment doses even when tumour heterogeneity is taken into account.

Our group have been developing methods for MRI-alone prostate cancer radiation therapy treatment planning. To assist with clinical validation of the workflow we are investigating a cloud platform solution for research purposes. Benefits of cloud computing can include increased scalability, performance and extensibility while reducing total cost of ownership. In this paper we demonstrate the generation of DICOM-RT directories containing an automatic average atlas based electron density image and fast pelvic organ contouring from whole pelvis MR scans.

Second cancer risk in patients, in particular in children, who were treated with radiotherapy is an important side effect. It should be minimized by selecting an appropriate treatment plan for the patient. The objectives of this study were to integrate a risk model for radiation induced cancer into a treatment planning system which allows to judge different treatment plans with regard to second cancer induction and to quantify the potential reduction in predicted risk. A model for radiation induced cancer including fractionation effects which is valid for doses in the radiotherapy range was integrated into a treatment planning system. From the three-dimensional (3D) dose distribution the 3D-risk equivalent dose (RED) was calculated on an organ specific basis. In addition to RED further risk coefficients like OED (organ equivalent dose), EAR (excess absolute risk) and LAR (lifetime attributable risk) are computed. A risk model for radiation induced cancer was successfully integrated in a treatment planning system. Several risk coefficients can be viewed and used to obtain critical situations were a plan can be optimised. Risk-volume-histograms and organ specific risks were calculated for different treatment plans and were used in combination with NTCP estimates for plan evaluation. It is concluded that the integration of second cancer risk estimates in a commercial treatment planning system is feasible. It can be used in addition to NTCP modelling for optimising treatment plans which result in the lowest possible second cancer risk for a patient.

Dose calculation methods in radiotherapy treatment planning require the radiological depth information of the voxels that represent the patient volume to correct for tissue inhomogeneities. This information is acquired by time consuming ray-tracing-based calculations. For treatment planning scenarios with changing geometries and real-time constraints this is a severe bottleneck. We implemented an algorithm for the graphics processing unit (GPU) which implements a ray-matrix approach to reduce the number of rays to trace. Furthermore, we investigated the impact of different strategies of accessing memory in kernel implementations as well as strategies for rapid data transfers between main memory and memory of the graphics device. Our study included the overlapping of computations and memory transfers to reduce the overall runtime using Hyper-Q. We tested our approach on a prostate case (9 beams, coplanar). The measured execution times for a complete ray-tracing range from 28 msec for the computations on the GPU to 99 msec when considering data transfers to and from the graphics device. Our GPU-based algorithm performed the ray-tracing in real-time. The strategies efficiently reduce the time consumption of memory accesses and data transfer overhead. The achieved runtimes demonstrate the viability of this approach and allow improved real-time performance for dose calculation methods in clinical routine.

This study used automated data processing techniques to calculate a set of novel treatment plan accuracy metrics, and investigate their usefulness as predictors of quality assurance (QA) success and failure. A small sample of 151 beams from 23 prostate and cranial IMRT treatment plans were used in this study. These plans had been evaluated before treatment using measurements with a diode array system. The TADA software suite was adapted to allow automatic batch calculation of several proposed plan accuracy metrics, including mean field area, small-aperture, off-axis and closed-leaf factors. All of these results were compared to the gamma pass rates from the QA measurements and correlations were investigated. The mean field area factor provided a threshold field size (5 cm², equivalent to a 2.2 × 2.2 cm² square field), below which all beams failed the QA tests. The small aperture score provided a useful predictor of plan failure, when averaged over all beams, despite being weakly correlated with gamma pass rates for individual beams. By contrast, the closed leaf and off-axis factors provided information about the geometric arrangement of the beam segments but were not useful for distinguishing between plans that passed and failed QA. This study has provided some simple tests for plan accuracy, which may help minimise time spent on QA assessments of treatments that are unlikely to pass.

The interactive dose shaping (IDS) planning paradigm aims to perform interactive local dose adaptations of an IMRT plan without compromising already established valuable dose features in real-time. In this work we introduce an interactive 3D isodose manipulation tool which enables local modifications of a dose distribution intuitively by direct manipulation of an isodose surface.

We developed an in-house IMRT TPS framework employing an IDS engine as well as a 3D GUI for dose manipulation and visualization. In our software an initial dose distribution can be interactively modified through an isodose surface manipulation tool by intuitively clicking on an isodose surface. To guide the user interaction, the position of the modification is indicated by a sphere while the mouse cursor hovers the isodose surface. The sphere's radius controls the locality of the modification. The tool induces a dose modification as a direct change of dose in one or more voxels, which is incrementally obtained by fluence adjustments. A subsequent recovery step identifies voxels with violated dose features and aims to recover their original dose.

We showed a proof of concept study for the proposed tool by adapting the dose distribution of a prostate case (9 beams, coplanar). Single dose modifications take less than 2 seconds on an actual desktop PC.

Purpose: A Bayesian Decision Network will be demonstrated to provide clinical decision support for adaptive lung response-driven treatment management based on evidence that physiologic metrics may correlate better with individual patient response than traditional (population-based) dose and volume-based metrics. Further, there is evidence that information obtained during the course of radiation therapy may further improve response predictions.

Methods: Clinical factors were gathered for 58 patients including planned mean lung dose, and the bio-markers IL-8 and TGF-β1 obtained prior to treatment and two weeks into treatment along with complication outcomes for these patients. A Bayesian Decision Network was constructed using Netica 5.0.2 from Norsys linking these clinical factors to obtain a prediction of radiation induced lung disese (RILD) complication. A decision node was added to the network to provide a plan adaption recommendation based on the trade-off between the RILD prediction and complexity of replanning. A utility node provides the weighting cost between the competing factors.

Results: The decision node predictions were optimized against the data for the 58 cases. With this decision network solution, one can consider the decision result for a new patient with specific findings to obtain a recommendation to adaptively modify the originally planned treatment course.

Conclusions: A Bayesian approach allows handling and propagating probabilistic data in a logical and principled manner. Decision networks provide the further ability to provide utility-based trade-offs, reflecting non-medical but practical cost/benefit analysis. The network demonstrated illustrates the basic concept, but many other factors may affect these decisions and work on building better models are being designed and tested.

Acknowledgement: Supported by NIH-P01-CA59827

Purpose: Using a database of prior treated patients, it is possible to predict the dose to critical structures for future patients. Automatic treatment planning speeds the planning process by generating a good initial plan from predicted dose values. Methods: A SQL relational database of previously approved treatment plans is populated via an automated export from Pinnacle³. This script outputs dose and machine information and selected Regions of Interests as well as its associated Dose-Volume Histogram (DVH) and Overlap Volume Histograms (OVHs) with respect to the target structures. Toxicity information is exported from Mosaiq and added to the database for each patient. The SQL query is designed to ask the system for the lowest achievable dose for a specified region of interest (ROI) for each patient with a given volume of that ROI being as close or closer to the target than the current patient. Results: The additional time needed to calculate OVHs is approximately 1.5 minutes for a typical patient. Database lookup of planning objectives takes approximately 4 seconds. The combined additional time is less than that of a typical single plan optimization (2.5 mins). Conclusions: An automatic treatment planning interface has been successfully used by dosimetrists to quickly produce a number of SBRT pancreas treatment plans. The database can be used to compare dose to individual structures with the toxicity experienced and predict toxicities before planning for future patients.

True quality control (QC) of the planning process requires quantitative assessments of treatment plan quality itself, and QC in IMRT has been stymied by intra-patient anatomical variability and inherently complex three-dimensional dose distributions. In this work we describe the development of an automated system to reduce clinical IMRT planning variability and improve plan quality using mathematical models that predict achievable OAR DVHs based on individual patient anatomy. These models rely on the correlation of expected dose to the minimum distance from a voxel to the PTV surface, whereby a three-parameter probability distribution function (PDF) was used to model iso-distance OAR subvolume dose distributions. DVH models were obtained by fitting the evolution of the PDF with distance. Initial validation on clinical cohorts of 40 prostate and 24 head-and-neck plans demonstrated highly accurate model-based predictions for achievable DVHs in rectum, bladder, and parotid glands. By quantifying the integrated difference between candidate DVHs and predicted DVHs, the models correctly identified plans with under-spared OARs, validated by replanning all cases and correlating any realized improvements against the predicted gains. Clinical implementation of these predictive models was demonstrated in the PINNACLE treatment planning system by use of existing margin expansion utilities and the scripting functionality inherent to the system. To maintain independence from specific planning software, a system was developed in MATLAB to directly process DICOM-RT data. Both model training and patient-specific analyses were demonstrated with significant computational accelerations from parallelization.

The aim of this paper was to introduce one of the options of the locally developed DDcon.exe which gives the possibility to transfer the dose distribution from CyberKnife (Accuray) treatment planning system (CK TPS) to Varian treatment planning system (Eclipse TPS, Varian). DICOM format is known as a universal format for medical data. The dose distribution is stored as RTdose file in DICOM format, so there should be a possibility to transfer it between different treatment planning systems. Trying to transfer RTdose file from CK TPS to Eclipse TPS the error message occurs. That's because the RTdose file in CK TPS is connected with Structure_Set_Sequence against Eclipse TPS where it's connected with RT_Plan_Sequence. To make it transferable RTdose file from CK TPS have to be 'disconnected' from Structure_Set_Sequence and 'connected' with RT_Plan_Sequence. This is possible thanks DDcon software which creates new RTdose file by changing proper DICOM tags in original RTdose file. New homemade software gives us an opportunity to transfer dose distribution from CyberKnife TPS to TPS Eclipse. This method opens new possibilities to combine or compare different treatment techniques in Varian TPS.

Cartesian co-ordinates, traditionally used for radiotherapy margins, calculated at 6 points, may not adequately represent changes in inter-observer contour variation as necessary to define a delineation margin. As a first step, this study compared the standard deviation (SD) in contour delineation using Polar and Cartesian co-ordinates for whole breast. Whole breast Clinical Target Volumes (CTV) were delineated by eight observers for 9 patients. The SD of contour position was determined for Polar co-ordinates at 1° increments for 5 slices and averaged across all patients. The mean centre of mass (COM) was used as the origin for the right breast, for the left the COM was shifted 1cm superiorly to avoid clipping. The SD was determined for Cartesian co-ordinates for medial-lateral and anterior-posterior positions. At slice Z=0cm considering Polar co-ordinates, the SD peaked medially reaching 3.55cm at 15° for the right breast, and 1.44cm at 171° for the left. The SD of the remaining slices maintained a similar distribution, with variation in the peak occurring within 10° of the Z=0cm positions. By comparison, for Cartesian co-ordinates at slice Z=0cm, the largest SD in the medial-lateral and anterior-posterior directions was 0.54/0.57cm and 1.03/0.67cm respectively for right/left breasts. The SD for inter-observer variation for whole breast varies with anatomical position. The maximum SD determined with Polar co-ordinates was greater than with Cartesian coordinates. A delineation margin may thus need to vary with angle over the entire structure and Cartesian co-ordinates may not be the best approach for margin determination for whole breast.

Current inverse optimization-based treatment planning for radiotherapy requires a set of complex DVH objectives to be simultaneously minimized. This process, known as multi-objective optimization, is challenging due to non-convexity in individual objectives and insufficient knowledge in the tradeoffs among the objective set. As such, clinical practice involves numerous iterations of human intervention that is costly and often inconsistent. In this work, we propose to address treatment planning with convex imputing, a new-data mining technique that explores the existence of a latent convex objective whose optimizer reflects the DVH and dose-shaping properties of previously optimized cases. Using ten clinical prostate cases as the basis for comparison, we imputed a simple least-squares problem from the optimized solutions of the prostate cases, and show that the imputed plans are more consistent than their clinical counterparts in achieving planning goals.

The purpose of this review is to highlight the critical issues of radiobiological models, particularly as they apply to clinical radiation therapy. Developing models of radiation responses has a long history that continues to the present time. Many different models have been proposed, but in the field of radiation oncology, the linear-quadratic (LQ) model has had the most impact on the design of treatment protocols. Questions have been raised as to the value of the LQ model given that the biological assumption underlying it has been challenged by molecular analyses of cell and tissue responses to radiation. There are also questions as to use of the LQ model for hypofractionation, especially for high dose treatments using a single fraction. While the LQ model might over-estimate the effects of large radiation dose fractions, there is insufficient information to fully justify the adoption of alternative models. However, there is increasing evidence in the literature that non-targeted and other indirect effects of radiation sometimes produce substantial deviations from LQ-like dose-response curves. As preclinical and clinical hypofractionation studies accumulate, new or refined dose-response models that incorporate high-dose/fraction non-targeted and indirect effects may be required, but for now the LQ model remains a simple, useful tool to guide the design of treatment protocols.

IMRT for head and neck patients requires clinicians to delineate clinical target volumes (CTV) on a planning-CT (>2hrs/patient). When patients require a replan-CT, CTVs must be re-delineated. This work assesses the performance of atlas-based autosegmentation (ABAS), which uses deformable image registration between planning and replan-CTs to auto-segment CTVs on the replan-CT, based on the planning contours. Fifteen patients with planning-CT and replan-CTs were selected. One clinician delineated CTVs on the planning-CTs and up to three clinicians delineated CTVs on the replan-CTs. Replan-CT volumes were auto-segmented using ABAS using the manual CTVs from the planning-CT as an atlas. ABAS CTVs were edited manually to make them clinically acceptable. Clinicians were timed to estimate savings using ABAS. CTVs were compared using dice similarity coefficient (DSC) and mean distance to agreement (MDA). Mean inter-observer variability (DSC>0.79 and MDA<2.1mm) was found to be greater than intra-observer variability (DSC>0.91 and MDA<1.5mm). Comparing ABAS to manual CTVs gave DSC=0.86 and MDA=2.07mm. Once edited, ABAS volumes agreed more closely with the manual CTVs (DSC=0.87 and MDA=1.87mm). The mean clinician time required to produce CTVs reduced from 169min to 57min when using ABAS. ABAS segments volumes with accuracy close to inter-observer variability however the volumes require some editing before clinical use. Using ABAS reduces contouring time by a factor of three.

Advances in preclinical radiotherapy systems have provided the technical foundations for delivering highly heterogeneous dose distributions for unique radiobiological experiments, but methods to deliver arbitrary dose distributions are in their infancy. This study developed a method to optimize and automatically deliver planar dose distributions on a recently developed preclinical radiotherapy platform. The method was based on empirically determined dose kernel distributions from radiochromic film measurements. These kernels were used to determine optimal animal stage positions and beam weights to deliver a desired dose distribution at a given depth using a sequential quadratic programming optimization algorithm. The method was validated by end-to-end delivery of two dosimetric challenges designed to quantify targeting and dosimetric accuracy. The results revelead an overall targeting accuracy of 112 μm and a dosimetric delivery error, calculated along four line profiles in radiochromic film measurements, of 6.8%. Mean absolute delivery error across a linear dose gradient between 0 and 1 Gy over 7.5 mm was 0.03 Gy. These results confirm the optimization framework is an effective platform for delivery of millimetre scale heterogeneous dose distributions with sub-millimetre accuracy.

Purpose: Over a decade ago, the 'Van Herk margin recipe paper' introduced plan evaluation through DVH statistics based on population distributions of systematic and random errors. We extended this work for structures with correlated uncertainties (e.g. lymph nodes or parotid glands), and considered treatment plans containing multiple (overlapping) dose distributions (e.g. conventional lymph node and hypo-fractionated tumor doses) for which different image guidance protocols may lead to correlated errors.

Methods: A command-line software tool 'PlanJury' was developed which reads 3D dose and structure data exported from a treatment planning system. Uncertainties are specified by standard deviations and correlation coefficients. Parameters control the DVH statistics to be computed: e.g. the probability of reaching a DVH constraint, or the dose absorbed at given confidence in a (combined) volume. Code was written in C++ and parallelized using OpenMP. Testing geometries were constructed using idealized spherical volumes and dose distributions.

Results: Negligible stochastic noise could be attained within two minutes computation time for a single target. The confidence to properly cover both of two targets was 90% for two synchronously moving targets, but decreased by 7% if the targets moved independently. For two partially covered organs at risk the confidence of at least one organ below the mean dose threshold was 40% for synchronous motion, 36% for uncorrelated motion, but only 20% for either of the organs separately. Two abutting dose distributions ensuring 91% confidence of proper target dose for correlated motions led to 28% lower confidence for uncorrelated motions as relative displacements between the doses resulted in cold spots near the target.

Conclusions: Probabilistic plan evaluation can efficiently be performed for complicated treatment planning situations, thus providing important plan quality information unavailable in conventional PTV based evaluations.

Scripting in radiotherapy treatment planning systems not only simplifies routine planning tasks but can also be used for clinical research. Treatment planning scripting can only be utilized in a system that has a built-in scripting interface. Among the commercially available treatment planning systems, Pinnacle (Philips) and Raystation (Raysearch Lab.) have inherent scripting functionality. CMS XiO (Elekta) is a widely used treatment planning system in radiotherapy centres around the world, but it does not have an interface that allows the user to script radiotherapy plans. In this study an external scripting interface, PyCMSXiO, was developed for XiO using the Python programming language. The interface was implemented as a python package/library using a modern object-oriented programming methodology. The package was organized as a hierarchy of different classes (objects). Each class (object) corresponds to a plan object such as the beam of a clinical radiotherapy plan. The interface of classes was implemented as object functions. Scripting in XiO using PyCMSXiO is comparable with Pinnacle scripting. This scripting package has been used in several research projects including commissioning of a beam model, independent three-dimensional dose verification for IMRT plans and a setup-uncertainty study. Ease of use and high-level functions provided in the package achieve a useful research tool. It was released as an open-source tool that may benefit the medical physics community.

SharePlan is a treatment planning system developed by Raysearch Laboratories AB to enable creation of a linear accelerator intensity modulated radiotherapy (IMRT) plan as a backup for a Tomotherapy plan. A 6MV Elekta Synergy Linear accelerator photon beam was modelled in SharePlan. The beam model was validated using Matrix Evolution, a 2D ion chamber array, for two head-neck and three prostate plans using 3%/3mm Gamma criteria. For 39 IMRT beams, the minimum and maximum Gamma pass rates are 95.4% and 98.7%. SharePlan is able to generate backup IMRT plans which are deliverable on a traditional linear accelerator and accurate in terms of clinical criteria. During use of SharePlan, however, an out-of-memory error frequently occurred and SharePlan was forced to be closed. This error occurred occasionally at any of these steps: loading the Tomotherapy plan into SharePlan, generating the IMRT plan, selecting the optimal plan, approving the plan and setting up a QA plan. The out-of-memory error was caused by memory leakage in one or more of the C/C++ functions implemented in SharePlan fluence engine, dose engine or optimizer, as acknowledged by the manufacturer. Because of the interruption caused by out-of-memory errors, SharePlan has not been implemented in our clinic although accuracy has been verified. A new software program is now being provided to our centre to replace SharePlan.

The last few years have seen a number of technical and clinical advances which give rise to a need for innovations in dose optimization and delivery strategies. Technically, a new generation of digital linac has become available which offers features such as programmable motion between station parameters and high dose-rate Flattening Filter Free (FFF) beams. Current inverse planning methods are designed for traditional machines and cannot accommodate these features of new generation linacs without compromising either dose conformality and/or delivery efficiency. Furthermore, SBRT is becoming increasingly important, which elevates the need for more efficient delivery, improved dose distribution. Here we will give an overview of our recent work in SPORT designed to harness the digital linacs and highlight the essential components of SPORT. We will summarize the pros and cons of traditional beamlet-based optimization (BBO) and direct aperture optimization (DAO) and introduce a new type of algorithm, compressed sensing (CS)-based inverse planning, that is capable of automatically removing the redundant segments during optimization and providing a plan with high deliverability in the presence of a large number of station control points (potentially non-coplanar, non-isocentric, and even multi-isocenters). We show that CS-approach takes the interplay between planning and delivery into account and allows us to balance the dose optimality and delivery efficiency in a controlled way and, providing a viable framework to address various unmet demands of the new generation linacs. A few specific implementation strategies of SPORT in the forms of fixed-gantry and rotational arc delivery are also presented.

Conventional intensity modulated radiation therapy (IMRT) treatment planning is based on the traditional concept of iterative optimization using an objective function specified by dose volume histogram constraints for pre-segmented VOIs. This indirect approach suffers from unavoidable shortcomings: i) The control of local dose features is limited to segmented VOIs. ii) Any objective function is a mathematical measure of the plan quality, i.e., is not able to define the clinically optimal treatment plan. iii) Adapting an existing plan to changed patient anatomy as detected by IGRT procedures is difficult.

To overcome these shortcomings, we introduce the method of Interactive Dose Shaping (IDS) as a new paradigm for IMRT treatment planning. IDS allows for a direct and interactive manipulation of local dose features in real-time. The key element driving the IDS process is a two-step Dose Modification and Recovery (DMR) strategy: A local dose modification is initiated by the user which translates into modified fluence patterns. This also affects existing desired dose features elsewhere which is compensated by a heuristic recovery process.

The IDS paradigm was implemented together with a CPU-based ultra-fast dose calculation and a 3D GUI for dose manipulation and visualization. A local dose feature can be implemented via the DMR strategy within 1-2 seconds. By imposing a series of local dose features, equal plan qualities could be achieved compared to conventional planning for prostate and head and neck cases within 1-2 minutes.

The idea of Interactive Dose Shaping for treatment planning has been introduced and first applications of this concept have been realized.

Purpose: To develop and validate a pre-treatment EPI dosimetry method on Varian TrueBeam linacs using continuous imaging, with reconstruction in a 3D cylindrical phantom geometry.

Methods: Delivery of VMAT plans with continuous imaging is currently possible only in Research Mode on TrueBeam linacs, with images acquired in a proprietary format. An earlier technique was adapted to take advantage of technical improvements in EPID delivery, and was tested under various acquisition conditions. The dosimetry of VMAT plans was evaluated at isocentre and within patient volumes that had been transferred to the virtual phantom.

Results: Approximately 60 portal image projections per arc were found to be adequate for 3D reconstruction in phantom volumes of 28cm diameter. Twelve prostate, CNS and Head & Neck deliveries were evaluated in Research mode relative to the corresponding Eclipse (v.10) treatment plans, and to measurements on an ArcCheck device in Treatment mode. Mean dose differences at isocentre were within 2% for the three-way comparison, and in PTV volumes were within 1% (s.d. 1%). However, some discrepancies were observed in ArcCheck results that may be related to the small dimensions of certain VMAT apertures.

Conclusions: EPI dosimetry with 3D dose reconstruction is an accurate, comprehensive and efficient pre-treatment validation technique for VMAT delivery. Although currently limited to a research mode on TrueBeam, it has the potential to be implemented for clinical use.

This study presents the development of a software tool 'Treat Check' to analyse the dynamic log files from an Elekta – Synergy accelerator. The software generates formatted output in the form of a plot presenting errors in various treatment delivery parameters such as gantry angle, Multi Leaf Collimator (MLC) leaf position, jaw position and Monitor Units (MU) for each of the control-points (CP) of the treatment beam. The plots are automatically saved in Portable Document Format (pdf). The software also has the functionality to introduce these treatment delivery errors into the original plan in the Pinnacle (Philips) treatment planning system (TPS) in order to assess the clinical impact of treatment delivery errors on delivered dose.

Development of a software tool to ease the Intensity Modulated Radiation Therapy (IMRT) pre-treatment Quality Assurance process is presented in this study. The delivery of IMRT involves equipment from multiple vendors. The limitations of the equipment involved in this chain will impact on the best choice of equipment. This often results in the user needing to use multiple pieces of equipment before determining the most appropriate choices to optimise the QA work flow. This is a time consuming process and potentially delays the start of patient treatment. Software was developed in-house to assist the decision making process, validating deliverability of beam delivery parameters and selecting appropriate detector systems and configuration for QA of IMRT plans. The software has been demonstrated to be accurate and improves efficiency of IMRT pre-treatment QA.

The performance of an automatic image registration algorithm was compared on image sets collected with two commercial CBCT systems, and the relationship with imaging dose was explored. CBCT images of a CIRS Virtually Human Male Pelvis phantom (VHMP) were collected on Varian TrueBeam/OBI and Elekta Synergy/XVI linear accelerators, across a range of mAs settings. Each CBCT image was registered 100 times, with random initial offsets introduced. Image registration was performed using the grey value correlation ratio algorithm in the Elekta XVI software, to a mask of the prostate volume with 5 mm expansion. Residual registration errors were calculated after correcting for the initial introduced phantom set-up error. Registration performance with the OBI images was similar to that of XVI. There was a clear dependence on imaging dose for the XVI images with residual errors increasing below 4mGy. It was not possible to acquire images with doses lower than ~5mGy with the OBI system and no evidence of reduced performance was observed at this dose. Registration failures (maximum target registration error > 3.6 mm on the surface of a 30mm sphere) occurred in 5% to 9% of registrations except for the lowest dose XVI scan (31%). The uncertainty in automatic image registration with both OBI and XVI images was found to be adequate for clinical use within a normal range of acquisition settings.

Two clinical intensity modulated radiotherapy plans were selected. Eleven plan variations were created with systematic errors introduced: Multi-Leaf Collimator (MLC) positional errors with all leaf pairs shifted in the same or the opposite direction, and collimator rotation offsets. Plans were measured using an Electronic Portal Imaging Device (EPID) and an ionisation chamber array. The plans were evaluated using gamma analysis with different criteria. The gamma pass rates remained around 95% or higher for most cases with MLC positional errors of 1 mm and 2 mm with 3%/3mm criteria. The ability of both devices to detect delivery errors was similar.

Purpose: The Gamma Index defines an asymmetric metric between the evaluated image and the reference image. It provides a quantitative comparison that can be used to indicate sample-wised pass/fail on the agreement of the two images. The Gamma passing/failing rate has become an important clinical evaluation tool. However, the presence of noise in the evaluated and/or reference images may change the Gamma Index, hence the passing/failing rate, and further, clinical decisions. In this work, we systematically studied the impact of the image noise on the Gamma Index calculation. Methods: We used both analytic formulation and numerical calculations in our study. The numerical calculations included simulations and clinical images. Three different noise scenarios were studied in simulations: noise in reference images only, in evaluated images only, and in both. Both white and spatially correlated noises of various magnitudes were simulated. For clinical images of various noise levels, the Gamma Index of measurement against calculation, calculation against measurement, and measurement against measurement, were evaluated. Results: Numerical calculations for both the simulation and clinical data agreed with the analytic formulations, and the clinical data agreed with the simulations. For the Gamma Index of measurement against calculation, its distribution has an increased mean and an increased standard deviation as the noise increases. On the contrary, for the Gamma index of calculation against measurement, its distribution has a decreased mean and stabilized standard deviation as the noise increases. White noise has greater impact on the Gamma Index than spatially correlated noise. Conclusions: The noise has significant impact on the Gamma Index calculation and the impact is asymmetric. The Gamma Index should be reported along with the noise levels in both reference and evaluated images. Reporting of the Gamma Index with switched roles of the images as reference and evaluated images or some composite metrics would be a good practice.

Tumours are known to be heterogeneous, yet typical treatment plans consider them as a single unit. This may influence treatment outcomes. However, treatment cannot be customised to intra-tumour variation without a method to establish outcomes at an intra-tumour scale. This work proposes a method to both assess and measure outcomes locally within tumours. Methods: Four patients were scanned at two post-surgery time points using contrast enhanced MRI and 3,4-dihydroxy-6-[18F]-fluoro-L-phenylalanine (18F-DOPA) PET. The shell of active tumour tissue is divided into a set of small subregions at both time points. Local outcome is measured from changes in subregion volume over time. The utility of the proposed approach is evaluated by measuring the correlation between PET uptake and documented growth. Correlation with overall survival time was also examined. Results: Local outcomes were heterogeneous and evidence of a positive correlation between local 18F-DOPA uptake and local progression was observed. Conclusions: Given that intra-tumour outcomes are heterogeneous the consistently positive correlation between FDOPA uptake and progression, local analysis of tumours could prove useful for treatment planning.

A decision-making framework for image-guided radiotherapy (IGRT) is being developed using a Bayesian Network (BN) to graphically describe, and probabilistically quantify, the many interacting factors that are involved in this complex clinical process. Outputs of the BN will provide decision-support for radiation therapists to assist them to make correct inferences relating to the likelihood of treatment delivery accuracy for a given image-guided set-up correction. The framework is being developed as a dynamic object-oriented BN, allowing for complex modelling with specific subregions, as well as representation of the sequential decision-making and belief updating associated with IGRT. A prototype graphic structure for the BN was developed by analysing IGRT practices at a local radiotherapy department and incorporating results obtained from a literature review. Clinical stakeholders reviewed the BN to validate its structure. The BN consists of a sub-network for evaluating the accuracy of IGRT practices and technology. The directed acyclic graph (DAG) contains nodes and directional arcs representing the causal relationship between the many interacting factors such as tumour site and its associated critical organs, technology and technique, and inter-user variability. The BN was extended to support on-line and off-line decision-making with respect to treatment plan compliance. Following conceptualisation of the framework, the BN will be quantified. It is anticipated that the finalised decision-making framework will provide a foundation to develop better decision-support strategies and automated correction algorithms for IGRT.

Purpose: During a typical 5-7 week treatment of external beam radiotherapy, there are potential differences between planned patient's anatomy and positioning, such as patient weight loss, or treatment setup. The discrepancies between planned and delivered doses resulting from these differences could be significant, especially in IMRT where dose distributions tightly conforms to target volumes while avoiding organs-at-risk. We developed an automatic system to monitor delivered dose using daily imaging. Methods: For each treatment, a merged image is generated by registering the daily pre-treatment setup image and planning CT using treatment position information extracted from the Tomotherapy archive. The treatment dose is then computed on this merged image using our in-house convolution-superposition based dose calculator implemented on GPU. The deformation field between merged and planning CT is computed using the Morphon algorithm. The planning structures and treatment doses are subsequently warped for analysis and dose accumulation. All results are saved in DICOM format with private tags and organized in a database. Due to the overwhelming amount of information generated, a customizable tolerance system is used to flag potential treatment errors or significant anatomical changes. A web-based system and a DICOM-RT viewer were developed for reporting and reviewing the results. Results: More than 30 patients were analysed retrospectively. Our in-house dose calculator passed 97% gamma test evaluated with 2% dose difference and 2mm distance-to-agreement compared with Tomotherapy calculated dose, which is considered sufficient for adaptive radiotherapy purposes. Evaluation of the deformable registration through visual inspection showed acceptable and consistent results, except for cases with large or unrealistic deformation. Our automatic flagging system was able to catch significant patient setup errors or anatomical changes. Conclusions: We developed an automatic dose verification system that quantifies treatment doses, and provides necessary information for adaptive planning without impeding clinical workflows.

Cone-beam computed tomography (CBCT) has enormous potential to improve the accuracy of treatment delivery in image-guided radiotherapy (IGRT). To assist radiotherapists in interpreting these images, we use a Bayesian statistical model to label each voxel according to its tissue type. The rich sources of prior information in IGRT are incorporated into a hidden Markov random field model of the 3D image lattice. Tissue densities in the reference CT scan are estimated using inverse regression and then rescaled to approximate the corresponding CBCT intensity values. The treatment planning contours are combined with published studies of physiological variability to produce a spatial prior distribution for changes in the size, shape and position of the tumour volume and organs at risk. The voxel labels are estimated using iterated conditional modes. The accuracy of the method has been evaluated using 27 CBCT scans of an electron density phantom. The mean voxel-wise misclassification rate was 6.2%, with Dice similarity coefficient of 0.73 for liver, muscle, breast and adipose tissue. By incorporating prior information, we are able to successfully segment CBCT images. This could be a viable approach for automated, online image analysis in radiotherapy.

Respiration and anatomical variation during radiotherapy (RT) of lung cancer yield dosimetric uncertainties of the delivered dose, possibly affecting the clinical outcome if not corrected for. Adaptive radiotherapy (ART), based on deformable image registration (DIR) and Deep-Inspiration-Breath-Hold (DIBH) gating can potentially improve the accuracy of RT. Purpose: The objective was to investigate the performance of contour propagation on repeated CT and Cone Beam CT (CBCT) images in DIBH compared to images acquired in free breathing (FB), using a recently released DIR software. Method: Three locally advanced non-small cell lung cancer patients were included, each with a planning-, midterm- and final CT (pCT, mCT, fCT) and 7 CBCTs acquired weekly and on the same day as the mCT and fCT. All imaging were performed in both FB and DIBH, using Varian RPM system for respiratory tracking. Delineations of anatomical structures were performed on each image set. The CT images were retrospective rigidly and deformable registered to all obtained images using the Varian Smart Adapt v. 11.0. The registered images were analysed for volume change and Dice Similarity Coefficient (DSC). Result: Geometrical similarities were found between propagated and manually delineated structures, with a slightly favour of FB imaging. Special notice should be taken to registrations where image artefacts or low tissue contrast are present. Conclusion: This study does not support the hypothesis that DIBH images perform better image registration than FB images. However DIR is a feasible tool for ART of lung cancer.

To develop a faster and lower dose topogram based image registration for TomoTherapy as an alternative image guidance tool to volumetric megavoltage computed tomography (MVCT). Topogram procedures were performed for an anthropomorphic thorax phantom on a TomoTherapy HD unit (Accuray Inc., Sunnyvale, CA) using couch speeds from 1-4 cm/s and gantry angles of 0 and 90 degrees, other scanning parameters are: 1 mm imaging jaw, compression factor of 1, 30 seconds scanning duration with all multileaf collimators (MLCs) open. The raw exit detector data was exported after each scan. The topogram was reconstructed from a fan beam source for TomoTherapy beam and detector geometry at a SSD of 85 cm. A reference image, so called Digitally Reconstructed Topogram (DRT) was created by integrating the trajectories through the kVCT simulation with the topogram geometry. Image registration was performed by visually aligning the bony structure in topogram to the DRT. Image resolution was determined by the radius of curvature for the detector array, source to axis distance, source to detector distance, detector spacing, and number of detectors. The localization errors were 1.5, 2.5 mm in medio-lateral and anterior-posterior direction, larger errors in cranial-caudal direction was observed for faster couch speeds (i.e., ≥3cm/s). The topographic imaging time was 30 sec (versus 3-5 minutes for MVCT thorax scan) with imaging dose less than 1% of MVCT scan. Topograms with appropriate couch speed provide reliable patient localization images while significantly reducing pre-treatment imaging time. Topogram can be used as an alternative and/or additional patient alignment tool to MVCT on TomoTherapy.

We propose the Reconstruction Toolkit (RTK, http://www.openrtk.org), an open-source toolkit for fast cone-beam CT reconstruction, based on the Insight Toolkit (ITK) and using GPU code extracted from Plastimatch. RTK is developed by an open consortium (see affiliations) under the non-contaminating Apache 2.0 license. The quality of the platform is daily checked with regression tests in partnership with Kitware, the company supporting ITK. Several features are already available: Elekta, Varian and IBA inputs, multi-threaded Feldkamp-David-Kress reconstruction on CPU and GPU, Parker short scan weighting, multi-threaded CPU and GPU forward projectors, etc. Each feature is either accessible through command line tools or C++ classes that can be included in independent software. A MIDAS community has been opened to share CatPhan datasets of several vendors (Elekta, Varian and IBA). RTK will be used in the upcoming cone-beam CT scanner developed by IBA for proton therapy rooms. Many features are under development: new input format support, iterative reconstruction, hybrid Monte Carlo / deterministic CBCT simulation, etc. RTK has been built to freely share tomographic reconstruction developments between researchers and is open for new contributions.

Aim: To identify potential interoperability problems facing multi-site Radiation Oncology (RO) departments in the Netherlands and solutions for unambiguous multi-system workflows. Specific challenges confronting the RO department of VUmc (RO-VUmc), which is soon to open a satellite department, were characterized. Methods: A nationwide questionnaire survey was conducted to identify possible interoperability problems and solutions. Further detailed information was obtained by in-depth interviews at 3 Dutch RO institutes that already operate in more than one site. Results: The survey had a 100% response rate (n=21). Altogether 95 interoperability problems were described. Most reported problems were on a strategic and semantic level. The majority were DICOM(-RT) and HL7 related (n=65), primarily between treatment planning and verification systems or between departmental and hospital systems. Seven were identified as being relevant for RO-VUmc. Departments have overcome interoperability problems with their own, or with tailor-made vendor solutions. There was little knowledge about or utilization of solutions developed by Integrating the Healthcare Enterprise Radiation Oncology (IHE-RO). Conclusions: Although interoperability problems are still common, solutions have been identified. Awareness of IHE-RO needs to be raised. No major new interoperability problems are predicted as RO-VUmc develops into a multi-site department.

Purpose: To use daily re-calculated dose distributions for optimization of target point corrections (TPCs) in image guided radiation therapy (IGRT). This aims to adapt fractioned intensity modulated radiation therapy (IMRT) to changes in the dose distribution induced by anatomical changes. Methods: Daily control images from an in-room on-rail spiral CT-Scanner of three head-and-neck cancer patients were analyzed. The dose distribution was re-calculated on each control CT after an initial TPC, found by a rigid image registration method. The clinical target volumes (CTVs) were transformed from the planning CT to the rigidly aligned control CTs using a deformable image registration method. If at least 95% of each transformed CTV was covered by the initially planned D95 value, the TPC was considered acceptable. Otherwise the TPC was iteratively altered to maximize the dose coverage of the CTVs. Results: In 14 (out of 59) fractions the criterion was already fulfilled after the initial TPC. In 10 fractions the TPC can be optimized to fulfill the coverage criterion. In 31 fractions the coverage can be increased but the criterion is not fulfilled. In another 4 fractions the coverage cannot be increased by the TPC optimization. Conclusions: The dose coverage criterion allows selection of patients who would benefit from replanning. Using the criterion to include daily re-calculated dose distributions in the TPC reduces the replanning rate in the analysed three patients from 76% to 59% compared to the rigid image registration TPC.

Purpose: Conformal radiation of moving tumours is a challenging task in radiotherapy. Tumour motion induced by respiration can be visualized in fluoroscopic images recorded during patients breathing. Markerless methods making use of registration techniques can be used to estimate tumour motion. However, registration methods might fail when the tumour is hidden by ribs. Using motion of anatomical surrogates, like the diaphragm, is promising to model tumour motion. Methods: A sequence of 116 fluoroscopic images was analyzed and the tumour positions were manually defined by three experts. A block matching (BM) technique is used to calculate the displacement vector relatively to a selected reference image of the first breathing cycle. An enhanced method was developed: Positions, when the tumour is not located behind a rib, are taken as valid estimations of the tumour position. Furthermore, these valid estimations are used to establish a linear model of tumour position and diaphragm motion. For invalid estimations the calculated tumour positions are not taken into consideration, and instead the model is used to determine tumour motion. Results: Enhancing BM with a model of tumour motion from diaphragm motion improves the tracking accuracy when the tumour moves behind a rib. The error (mean ± SD) in longitudinal dimension was 2.0 ± 1.5mm using only BM and 1.0 ± 1.1mm when the enhanced approach was used. Conclusion: The enhanced tracking technique is capable to improve tracking accuracy compared to BM in the case that the tumour is occluded by ribs.

The purpose of this work is to validate an in-house deformable image registration (DIR) algorithm for adaptive radiotherapy for head and neck patients. We aim to use the registrations to estimate the "dose of the day" and assess the need to replan. NiftyReg is an open-source implementation of the B-splines deformable registration algorithm, developed in our institution. We registered a planning CT to a CBCT acquired midway through treatment for 5 HN patients that required replanning. We investigated 16 different parameter settings that previously showed promising results. To assess the registrations, structures delineated in the CT were warped and compared with contours manually drawn by the same clinical expert on the CBCT. This structure set contained vertebral bodies and soft tissue. Dice similarity coefficient (DSC), overlap index (OI), centroid position and distance between structures' surfaces were calculated for every registration, and a set of parameters that produces good results for all datasets was found. We achieve a median value of 0.845 in DSC, 0.889 in OI, error smaller than 2 mm in centroid position and over 90% of the warped surface pixels are distanced less than 2 mm of the manually drawn ones. By using appropriate DIR parameters, we are able to register the planning geometry (pCT) to the daily geometry (CBCT).

Using an EPID for patient specific IMRT QA is an efficient way to verify patient plans prior to the treatment. Our centres' current EPID dosimetry method is based on a water equivalent depth approach, where EPID images of each IMRT field are converted to dose images and compared to Treatment planning system calculated dose images using a commercial software tool. Two physicists across two sites perform the analysis for an average of 12 new IMRT patients per week. To speed up this process, an in-house program called AutoEPIDIMRTQA was developed. The program automatically performs the following tasks sequentially: reading and converting raw EPID images acquired on either a Siemens Oncor or Elekta Synergy linear accelerator, registering them with planar dose images calculated by Pinnacle (Philips) or CMS XiO (Elekta) treatment planning systems, analyzing the profiles for registered images and calculating the Gamma map. Finally an IMRT QA report is automatically generated. AutoEPIDIMRTQA was validated against commercial software. The analysis time for a typical 9-beam IMRT head-neck patient decreases from 30 minutes to 4 minutes. The total QA time was reduced by 40% using AutoEPIDIMRTQA. Thus we have demonstrated a significant reduction in the time burden for physics staff performing IMRT QA.

Advanced radiotherapy techniques such as volumetric modulated arc therapy (VMAT) require verification of the complex beam delivery including tracking of multileaf collimators (MLC) and monitoring the dose rate. This work explores the feasibility of a prototype Complementary metal-oxide semiconductor Image Sensor (CIS) for tracking these complex treatments by utilising fast, region of interest (ROI) read out functionality. An automatic edge tracking algorithm was used to locate the MLC leaves edges moving at various speeds (from a moving triangle field shape) and imaged with various sensor frame rates. The CIS demonstrates successful edge detection of the dynamic MLC motion within accuracy of 1.0 mm. This demonstrates the feasibility of the sensor to verify treatment delivery involving dynamic MLC up to ~400 frames per second (equivalent to the linac pulse rate), which is superior to any current techniques such as using electronic portal imaging devices (EPID). CIS provides the basis to an essential real-time verification tool, useful in accessing accurate delivery of complex high energy radiation to the tumour and ultimately to achieve better cure rates for cancer patients.

Purpose: In recent years there has been interest in using Computer Simulation within Medical training. The VERT (Virtual Environment for Radiotherapy Training) system is a Flight Simulator for Radiation Oncology professionals, wherein fundamental concepts, techniques and problematic scenarios can be safely investigated.

Methods: The system provides detailed simulations of several Linacs and the ability to display DICOM treatment plans. Patients can be mis-positioned with 'set-up errors' which can be explored visually, dosimetrically and using IGRT. Similarly, a variety of Linac calibration and configuration parameters can be altered manually or randomly via controlled errors in the simulated 3D Linac and its component parts. The implication of these can be investigated by following through a treatment scenario or using QC devices available within a Physics software module.

Results: One resultant exercise is a systematic mis-calibration of 'lateral laser height' by 2mm. The offset in patient alignment is easily identified using IGRT and once corrected by reference to the 'in-room monitor'. The dosimetric implication is demonstrated to be 0.4% by setting a dosimetry phantom by the lasers (and ignoring TSD information). Finally, the need for recalibration can be shown by the Laser Alignment Phantom or by reference to the front pointer.

Conclusions: The VERT system provides a realistic environment for training and enhancing understanding of radiotherapy concepts and techniques. Linac error conditions can be explored in this context and valuable experience gained in a controlled manner in a compressed period of time.

The aims of this work were to establish a program to fit NTCP models to clinical data with multiple toxicity endpoints, to test the method using a realistic test dataset, to compare three methods for estimating confidence intervals for the fitted parameters and to characterise the speed and performance of the program.

Purpose: A system has been developed which recognizes and classifies Digital Imaging and Communication in Medicine contour data with minimal human intervention. It allows researchers to overcome obstacles which tax analysis and mining systems, including inconsistent naming conventions and differences in data age or resolution.

Methods: Lexicographic and geometric analysis is used for recognition. Well-known lexicographic methods implemented include Levenshtein-Damerau, bag-of-characters, Double Metaphone, Soundex, and (word and character)-N-grams. Geometrical implementations include 3D Fourier Descriptors, probability spheres, boolean overlap, simple feature comparison (e.g. eccentricity, volume) and rule-based techniques. Both analyses implement custom, domain-specific modules (e.g. emphasis differentiating left/right organ variants). Contour labels from 60 head and neck patients are used for cross-validation.

Results: Mixed-lexicographical methods show an effective improvement in more than 10% of recognition attempts compared with a pure Levenshtein-Damerau approach when withholding 70% of the lexicon. Domain-specific and geometrical techniques further boost performance.

Conclusions: DICOMautomaton allows users to recognize contours semi-automatically. As usage increases and the lexicon is filled with additional structures, performance improves, increasing the overall utility of the system.

Purpose: To provide clinicians & researchers participating in multi-centre clinical trials with a central repository for large volume dynamic imaging data as well as a set of tools for providing end-to-end testing and image analysis standards of practice. Methods: There are three main pieces to the data archiving and analysis system; the PACS server, the data analysis computer(s) and the high-speed networks that connect them. Each clinical trial is anonymized using a customizable anonymizer and is stored on a PACS only accessible by AE title access control. The remote analysis station consists of a single virtual machine per trial running on a powerful PC supporting multiple simultaneous instances. Imaging data management and analysis is performed within ClearCanvas Workstation® using custom designed plug-ins for kinetic modelling (The DCE-Tool®), quality assurance (The DCE-QA Tool) and RECIST. Results: A framework has been set up currently serving seven clinical trials spanning five hospitals with three more trials to be added over the next six months. After initial rapid image transfer (+ 2 MB/s), all data analysis is done server side making it robust and rapid. This has provided the ability to perform computationally expensive operations such as voxel-wise kinetic modelling on very large data archives (+20 GB/50k images/patient) remotely with minimal end-user hardware. Conclusions: This system is currently in its proof of concept stage but has been used successfully to send and analyze data from remote hospitals. Next steps will involve scaling up the system with a more powerful PACS and multiple high powered analysis machines as well as adding real-time review capabilities.

Large multicentre radiotherapy trials incorporating assessment of multiple outcomes at multiple timepoints can generate extensive datasets. We have investigated graphical techniques for presentation of this data and the associated underlying dose-volume response information, necessary for guiding statistical analyses and translating outcomes to future patient treatments. A relational database was used to archive reviewed plan data for patients accrued to the TROG 03.04 RADAR trial. Viewing software was used to clean and enhance the data. Scripts were developed to export arbitrary dose-histogram data which was combined with clinical toxicity data with a median follow-up of 72 months. Graphical representations of dose-volume response developed include prevalence atlasing, univariate logistic regression and dose-volume-point odds ratios, and continuous cut-point derivation via ROC analysis. These representations indicate variable association of toxicities across structures and time-points.

There are many advantages to performing a clinical trial when implementing a novel radiotherapy technique. The clinical trials framework enables the safety and efficacy of the "experimental arm" to be tested and ensures practical support, rigorous quality control and data monitoring for participating centres. In addition to the clinical and follow-up data collected from patients within the trial, it is also possible to collect 3-D dosimetric information from the corresponding radiotherapy treatment plans. Analysing the combination of dosimetric, clinical and follow-up data enhances the understanding of the relationship between the dose delivered to both the target and normal tissue structures and reported outcomes & toxicity. Aspects of the collection, collation and analysis of data from two UK multicentre Phase III radiotherapy trials are presented here. MRC-RT01 dose-escalation prostate radiotherapy trial ISRCTN47772397 was one of the first UK multi-centre radiotherapy trials to collect 3-D dosimetric data. A number of different analysis methodologies were implemented to investigate the relationship between the dose distribution to the rectum and specific rectal toxicities. More recently data was collected from the PARSPORT trial (Parotid Sparing IMRT vs conventional head and neck radiotherapy) ISRCTN48243537. In addition to the planned analysis, dosimetric analysis was employed to investigate an unexpected finding that acute fatigue was more prevalent in the IMRT arm of the trial. It can be challenging to collect 3-D dosimetric information from multicentre radiotherapy trials. However, analysing the relationship between dosimetric and toxicity data provides invaluable information which can influence the next generation of radiotherapy techniques.

The management of patient information and treatment planning is traditionally an intra-departmental requirement of a radiation oncology service. Epworth Radiation Oncology systems must support the transient nature of Visiting Medical Officers (VMOs). This unique work practice created challenges when implementing the vision of a completely paperless solution that allows for a responsive and efficient service delivery. ARIA^® and Eclipse^TM (Varian Medical Systems, Palo Alto, CA, USA) have been deployed across four dedicated Citrix^® (Citrix Systems, Santa Clara, CA, USA) servers allowing VMOs to access these applications remotely. A range of paperless solutions were developed within ARIA^® to facilitate clinical and organisational management whilst optimising efficient work practices. The IT infrastructure and paperless workflow has enabled VMOs to securely access the Varian^TM (Varian Medical Systems, Palo Alto, CA, USA) oncology software and experience full functionality from any location on multiple devices. This has enhanced access to patient information and improved the responsiveness of the service. Epworth HealthCare has developed a unique solution to enable remote access to a centralised oncology management suite, while maintaining a secure and paperless working environment.

Plan review systems often allow dose volume histogram (DVH) recalculation as part of a quality assurance process for trials. A review of the algorithms provided by a number of systems indicated that they are often very similar. One notable point of variation between implementations is in the location and frequency of dose sampling. This study explored the impact such variations can have on DVH based plan evaluation metrics (Normal Tissue Complication Probability (NTCP), min, mean and max dose), for a plan with small structures placed over areas of high dose gradient. Dose grids considered were exported from the original planning system at a range of resolutions. We found that for the CT based resolutions used in all but one plan review systems (CT and CT with guaranteed minimum number of sampling voxels in the x and y direction) results were very similar and changed in a similar manner with changes in the dose grid resolution despite the extreme conditions. Differences became noticeable however when resolution was increased in the axial (z) direction. Evaluation metrics also varied differently with changing dose grid for CT based resolutions compared to dose grid based resolutions. This suggests that if DVHs are being compared between systems that use a different basis for selecting sampling resolution it may become important to confirm that a similar resolution was used during calculation.

Computer based Treatment Planning Systems (TPS) are used worldwide to design and calculate treatment plans for treating radiation therapy patients. TPS are generally well designed and thoroughly tested by their developers and local physicists prior to clinical use. However, the wide-reaching impact of their accuracy warrants ongoing vigilance. This work reviews the findings of the Australian national audit system and provides recommendations for checks of TPS. The Australian Clinical Dosimetry Service (ACDS) has designed and implemented a national system of audits, currently in a three year test phase. The Level III audits verify the accuracy of a beam model of a facility's TPS through a comparison of measurements with calculation at selected points in an anthropomorphic phantom. The plans are prescribed by the ACDS and all measurement equipment is brought in for independent onsite measurements. In this first version of audits, plans are comparatively simple, involving asymmetric fields, wedges and inhomogeneities. The ACDS has performed 14 Level III audits to-date. Six audits returned at least one measurement at Action Level, indicating that the measured dose differed more than 3.3% (but less than 5%) from the planned dose. Two audits failed (difference >5%). One fail was caused by a data transmission error coupled with quality assurance (QA) not being performed. The second fail was investigated and reduced to Action Level with the onsite audit team finding phantom setup at treatment a contributing factor. The Action Level results are attributed to small dose calculation deviations within the TPS, which are investigated and corrected by the facilities. Small deviations exist in clinical TPS which can add up and can combine with output variations to result in unacceptable variations. Ongoing checks and independent audits are recommended.

Purpose: Incompatibility between documentation and clinical workflow causes physician resistance in organized data collection, which in turn complicates the use of data in patient care improvement. To resolve the gap, we developed an iPad compatible in situ browser-based platform that integrates clinical activity with data collection and analysis presentation. The ability to perform in-clinic activities and monitor decision making using the iPad was evaluated.

Methods: A browser-based platform that can exchange and present analysed data from the MOSAIQ database was developed in situ, the iPads were distributed in head and neck clinics to present the browser for clinical activities, data collection and assessment monitoring. Performance of the iPads for in-clinic activities was observed.

Results: All in-clinic documentation activities can be performed without workstation computers. Accessing patient record and previous assessments was significantly faster without having to open the MOSAIQ application. Patient assessments can be completed with the physician facing the patient. Graphical presentation of toxicity progression and patient radiation plans to the patient can be performed in single interface without patient leaving the seating area. Updates in patient treatment status and medical history were presented in real time without having to move paper charts around.

Conclusions: The iPad can be used in clinical activities independent of computer workstations. Improvements in clinical workflow can be critical in reducing physician resistance in data maintenance. Using the iPad in providing real-time quality monitoring is intuitive to both providers and patients.

Purpose: To describe the unique characteristics of models that represent the entire course of radiation therapy at the organism level and to highlight the uses to which such models can be put. Methods: At the level of an organism, traditional model-building runs into severe difficulties. We do not have sufficient knowledge to devise a complete biochemistry-based model. Statistical model-building fails due to the vast number of variables and the inability to control many of them in any meaningful way. Finally, building surrogate models, such as animal-based models, can result in excluding some of the most critical variables. Bayesian probabilistic models (Bayesian networks) provide a useful alternative that have the advantages of being mathematically rigorous, incorporating the knowledge that we do have, and being practical. Results: Bayesian networks representing radiation therapy pathways for prostate cancer and head & neck cancer were used to highlight the important aspects of such models and some techniques of model-building. A more specific model representing the treatment of occult lymph nodes in head & neck cancer were provided as an example of how such a model can inform clinical decisions. A model of the possible role of PET imaging in brain cancer was used to illustrate the means by which clinical trials can be modelled in order to come up with a trial design that will have meaningful outcomes. Conclusions: Probabilistic models are currently the most useful approach to representing the entire therapy outcome process.

We devise a paradigm for representing the DICOM-RT structure sets in a database management system, in such way that secondary calculations of geometric information can be performed quickly from the existing contour definitions. The implementation of this paradigm is achieved using the PostgreSQL database system and the PostGIS extension, a geographic information system commonly used for encoding geographical map data. The proposed paradigm eliminates the overhead of retrieving large data records from the database, as well as the need to implement various numerical and data parsing routines, when additional information related to the geometry of the anatomy is desired.

Purpose: It is well recognized that computer technology has had a major impact on the practice of radiation oncology. This paper addresses the question as to how these computer advances have specifically impacted the accuracy of radiation dose delivery to the patient.Methods: A review was undertaken of all the key steps in the radiation treatment process ranging from machine calibration to patient treatment verification and irradiation. Using a semi-quantitative scale, each stage in the process was analysed from the point of view of gains in treatment accuracy. Results: Our critical review indicated that computerization related to digital medical imaging (ranging from target volume localization, to treatment planning, to image-guided treatment) has had the most significant impact on the accuracy of radiation treatment. Conversely, the premature adoption of intensity-modulated radiation therapy has actually degraded the accuracy of dose delivery compared to 3-D conformal radiation therapy. While computational power has improved dose calibration accuracy through Monte Carlo simulations of dosimeter response parameters, the overall impact in terms of percent improvement is relatively small compared to the improvements accrued from 3-D/4-D imaging. Conclusions: As a result of computer applications, we are better able to see and track the internal anatomy of the patient before, during and after treatment. This has yielded the most significant enhancement to the knowledge of "in vivo" dose distributions in the patient. Furthermore, a much richer set of 3-D/4-D co-registered dose-image data is thus becoming available for retrospective analysis of radiobiological and clinical responses.

Many clinical trials use Electronic Case Report Forms (ECRF), e.g., from OpenClinica. Trial data is augmented if DICOM scans, dose cubes, etc. from the Picture Archiving and Communication System (PACS) are included for data mining. Unfortunately, there is as yet no structured way to collect DICOM objects in trial databases. In this paper, we obtain a tight integration of ECRF and PACS using open source software. Methods: DICOM identifiers for selected images/series/studies are stored in associated ECRF events (e.g., baseline) as follows: 1) JavaScript added to OpenClinica communicates using HTML with a gateway server inside the hospitals firewall; 2) On this gateway, an open source DICOM server runs scripts to query and select the data, returning anonymized identifiers; 3) The scripts then collects, anonymizes, zips and transmits selected data to a central trial server; 4) Here data is stored in a DICOM archive which allows authorized ECRF users to view and download the anonymous images associated with each event. Results: All integration scripts are open source. The PACS administrator configures the anonymization script and decides to use the gateway in passive (receiving) mode or in an active mode going out to the PACS to gather data. Our ECRF centric approach supports automatic data mining by iterating over the cases in the ECRF database, providing the identifiers to load images and the clinical data to correlate with image analysis results. Conclusions: Using open source software and web technology, a tight integration has been achieved between PACS and ECRF.

Purpose: To create a central radiotherapy (RT) plan database for dose analysis and reporting, capable of calculating and presenting statistics on user defined patient groups. The goal is to facilitate multi-center research studies with easy and secure access to RT plans and statistics on protocol compliance. Methods: RT institutions are able to send data to the central database using DICOM communications on a secure computer network. The central system is composed of a number of DICOM servers, an SQL database and in-house developed software services to process the incoming data. A web site within the secure network allows the user to manage their submitted data. Results: The RT plan database has been developed in Microsoft .NET and users are able to send DICOM data between RT centers in Denmark. Dose-volume histogram (DVH) calculations performed by the system are comparable to those of conventional RT software. A permission system was implemented to ensure access control and easy, yet secure, data sharing across centers. The reports contain DVH statistics for structures in user defined patient groups. The system currently contains over 2200 patients in 14 collaborations. Conclusions: A central RT plan repository for use in multi-center trials and quality assurance was created. The system provides an attractive alternative to dummy runs by enabling continuous monitoring of protocol conformity and plan metrics in a trial.