Volume 33, Issue 6, June 2006
Index of content:
- Therapy Scientific Session: Room 224 A
- IMRT Verication and QA I
33(2006); http://dx.doi.org/10.1118/1.2241436View Description Hide Description
Purpose: To develop a system for real‐time auditing and archiving of MLC treatment data, both static and IMRT, through the passive detection and reconstruction of packet‐level network traffic between the linac control computers and the central database server. Method and Materials: The three computers under observation are the control computers for a Varian 21EX (120‐leaf dynamic MLC) and a Varian 2100‐2 (52‐leaf static MLC) and the central database server running Varis 7 clinical informaticssoftware. The observing system consists of a Linux‐based PC connected via Ethernet to modified network segments of the two linac control computers, running the Ethereal and Snort open‐source network analysis tools in addition to database and web services. After detailed observation of the computers' data transfer protocols and MLC data formats, the system was configured to passively monitor Ethernet traffic during clinical operations and selectively record only those packets containing specific treatment session, field, and MLC details to a database. Finally, a custom PHP‐MySQL script was written to reassemble, convert, and present this binary data in human‐readable, web‐accessible format. Results: The system successfully reconstructs and stores the network‐detectedMLC configurations for all fields in a treatment fraction, seconds after the patient file is opened at either treatment computer. Its application to IMRT QA is both real‐time MLC auditing (e.g. verify that the MLC data received by the server matches that of the previous fraction, 10–15 minutes before treatment delivery) and as an off‐line archive of transmitted MLC configuration data. Conclusion: This system demonstrates that the examination of network traffic can provide a useful and innovative tool for radiotherapy QA. Its informatics‐centered approach to IMRT QA contrasts with and complements the numerous delivery‐centered techniques, which focus on dosimetric verification of MLC sequences that are known to have been transmitted to the linaccomputer without error.
MO‐D‐224A‐02: Clinical Experience in Using EPID for Quantitative Verification of IMRT Dose Distributions33(2006); http://dx.doi.org/10.1118/1.2241437View Description Hide Description
Purpose: The purpose of this work is to investigate the advantages of using an electronic portal imaging device(EPID) as part of our clinical IMRT QA procedure. We present our experience with the quantitative verification of the planar dose distribution for patient‐specific IMRTtreatment plans. Method and Materials: Our treatment machines which are used for IMRT delivery are equipped with amorphous siliconEPIDs. The planning system used for routine IMRTtreatment planning calculates planar dose at a predefined source to plane distance (SPD) as part of the EPID‐based QA process. For quantitative analysis, dosimetric calibration of the EPID was required, in addition to the standard imagercalibrations of the PortalVision imaging device. Basic dosimetric characteristics of the system were initially evaluated using fields with known dosimetry. Absolute dose calibration produces results comparable to ion chamber measurements. The patient IMRT QA fields were delivered prior to treatment and recorded. We have selected 100 cases from our routine IMRT case load for analysis using this procedure. Results: The verification of dose distributions was performed using portal dosimetry in Vision, after beam delivery. The overlay of the acquired and the calculated planar doses as isodose lines provided a useful qualitative evaluation. For further quantitative analysis, the gamma relative evaluation (GRE) was used on each field for every patient. The GRE scores were normally greater than 0.99 for prostate patients and greater than 0.98 for Head and Neck cases or large pelvic fields. Conclusion: Planar dose verification is an important part of the IMRT QA procedure. With the aid of portal dosimetry using EPID, the QA of an IMRT plan can be performed in 30 minutes, saving considerable time compared to film dosimetry. With proper dosimetric calibration, quantitative analysis further ensures excellent quality assurance of IMRT planning and delivery.
33(2006); http://dx.doi.org/10.1118/1.2241438View Description Hide Description
Purpose: This work aims to investigate the sensitivity of IMRT QA done by means of planar dosimetry to MLC positional inaccuracies. Also we propose an accurate method for measurement of MLC positioning errors using the 2D diode array. Methods: A method to measure the MLC position errors by using 2D‐arrary of 455 diodes (MapCheck, Sun Nuclear) is developed. Our method utilizes the fact that each diode's signal will be most sensitive to the MLC position error when its center coincides with the edge of a MLC leaf where a small deviation from the accurate position produces sharp increase or decrease in the diode output. We designed various multi‐segmented test patterns based on this principle that can evaluate the MLC deviations with sub‐mm accuracy at multiple MLC banks simultaneously. By using this information as a deviation histogram, we created deliberately erroneous IMRT fields with varying standard deviations of MLC leaf position error. MapCHECK planar dose analysis is performed and the sensitivity of the IMRT QA procedure to MLC position inaccuracy using MapCHECK device is evaluated. Results: Our results indicate that for SIEMENS Primus and MD machines, the MLC positional errors show a standard deviation of about +/− 0.7mm. Right after the MLCcalibration, this deviation might reduce to 0.55mm. It has been found that a single leaf position can vary by as much as 1mm between two consecutive measurements. Fields with less number of segments that are generated by Direct Aperture Optimization are found to be less sensitive to these errors by measurements with deliberately modified fields with random MLC inaccuracies. Conclusion: A method to accurately quantify MLC position is proposed and used to obtain a distribution of leaf position errors for many leaves at multiple banks. The sensitivity of planar dosimeter to the MLC positioning errors is investigated.
33(2006); http://dx.doi.org/10.1118/1.2241439View Description Hide Description
Purpose:Radiotherapy patients treated for H&N cancer often lose weight and have shrinkage of their tumors causing drastic anatomical changes. This can result in changes in dose distribution with respect to PTV coverage and OARs. Monitoring these changes is difficult and presents QA problems for IMRTtreatments. In this work we develop a method to monitor H&N thickness changes and correlate with changes in dose distribution. Method and Materials: Wax was applied to the neck region of an anthropomorphic phantom in 3‐1cm layers. Contours depicting tumor and critical structures were delineated. An IMRT plan was generated to delivery 70Gy to the PTV. The resultant sequence was virtually delivered to the phantom with layers of wax removed. PTV coverage, hot spot, and PRV doses were recorded. A characteristic response curve was generated using the amorphous siliconEPID on a Varian 21EX linear accelerator and slabs of solid water. The phantom was then consecutively imaged removing 1cm layers of wax. All images were acquired for the same number of MU. Results: Decreasing the thickness of the neck region bilaterally by 3cm resulted in an increase in 95% PTV coverage from 70Gy to 72.8Gy. The maximum dose in the PTV increased from 80.6Gy to 86Gy. The PTV volume receiving 110% of the prescribed dose increased from 13.4% to 66.6%. The dose to 0.01cc of the spinal cord and brainstem PRVs increased from 45.8Gy to 47.9Gy and 49.7Gy to 51.2Gy, respectively. Using the EPID we were able to predict changes in the lateral dimension of the phantom to within 4mm. Conclusion: Our results indicate that anatomical changes during treatment may lead to unacceptable dose distributions. By using lateral EPIDimages we can monitor the thickness changes (path length) in the H&N region. These changes may be used to determine when re‐planning is necessary.
33(2006); http://dx.doi.org/10.1118/1.2241440View Description Hide Description
Purpose: The purpose of this work is to develop and validate an algorithm based on Monte Carlo(MC) simulations to compute the in‐vivo dose given to patients during conventional treatments and IMRT from portal images.Method and Materials: The exit fluence from primary particles is obtained from the portal image after correction for scatter radiation. The scattered radiation at the portal imager and the spectral energy distribution of the primary particles are estimated from MC simulations at the treatment planning stage. The exit fluence and the spectral energy distribution of the primary particles are then used to ray‐trace the particles from the portal imager towards the source through the CT geometry of the patient. Particle weights reflecting the probability of a particle being transmitted are computed during this step. A dedicated MC code is finally used to transport back these particles from the source through the CT geometry of the patient to obtain an in‐vivo dose. Only Compton interactions are considered and secondary electron transport is implemented by depositing the dose uniformly on the surface of a sphere of radius corresponding to the electron projected range. This code also produces a predicted portal image which is used as a verification tool to ensure that the dose reconstruction is reliable. The dose reconstruction algorithm was compared against MCtreatment planning (MCTP) predictions and against measurements. Results: The reconstructed doses and the MCTP predictions in homogeneous and heterogeneous phantoms agree within 1% for simple open fields except in the buildup region and in the penumbra where the agreement is within 9%. Comparison with film‐measured dose distribution for IMRT fields yielded agreement within 3 mm, 3 %. Conclusion: A novel dose reconstruction algorithm based on MC simulations has been developed and validated in homogeneous and heterogeneous phantoms for conventional and IMRT fields.
MO‐D‐224A‐06: Fast Monte Carlo‐Based Computation of ASi‐EPID Dose Images for IMRT Treatment Field Through Phantom33(2006); http://dx.doi.org/10.1118/1.2241441View Description Hide Description
Purpose: During‐treatment IMRTdosimetric verification can be accomplished with exit dose portal dosimetry; however, differential beam hardening and patient scatter radiation results in inaccuracies in invariant kernel‐based calculation methods. The purpose of this study is to develop an accurate, yet efficient Monte‐Carlo (MC) based algorithm to predict during treatmentdosimetric aSi‐EPID images to compare with measured images for plan delivery quality assurance.Method and Materials: To compute EPIDimages, the VMC++ MC algorithm is used to transport particles through the patient geometry. Particles exiting the patient are scored into 19 energy‐differential fluence‐matrices at the EPID surface. Computed EPIDimages are generated by summing the contributions of each fluence‐matrix convolved with MC generated mono‐energetic energy deposition kernels. Kernel‐based method validation was performed for open, MLC‐blocked, intensity test‐pattern and a prostate‐IMRT field with and without a 20 cm thick phantom by comparing with full MC computation of EPIDimages. Additionally, the prostate‐IMRT plan was computed through a pelvic phantom. A cone‐beam CT of the pelvic phantom was used for dose computation particle transport. Comparison metrics include image profiles and gamma‐metric evaluation. Results: For the test fields, kernel‐based methods had >95% of voxels with γ<1 for 1 %, 1mm criteria and >99.1% with a 2%, 2mm criteria with respect to the MC‐calculated fields. For the pelvic phantom, 92.6% of pixels had γ<1 for 1%, 1mm criteria. The systematic discrepancy(∼0.5%) is well below the statistical uncertainty(∼3%). Conclusion: The kernel‐based convolution method is comparable in accuracy with full MC while requiring substantially less computation time than a full MCEPID simulation. Image computation time is independent of MC statistical precision and adds <1min to the MC simulation time. Comparison of measured images with MC‐computed portal images may be a practical method to perform during‐treatment dose validation. Conflict of Interest: Supported in‐part by Varian Medical Systems.
33(2006); http://dx.doi.org/10.1118/1.2241442View Description Hide Description
Purpose: To provide a mechanism for quantitative verification of each bixel in an IMRT fluence map within minutes. Method and Materials: A software tool called Super‐IMPOSE (Intensity Map Pre‐treatment OverSight & Evaluation) was developed to quantitatively verify fluence maps using an Electronic Portal Imager (EPI). This software imports EPI‐measured fluence maps along with TPS‐calculated maps and compares them by bixel. We developed a convolution algorithm to correct the raw fluence map for the effects of scatter within the imager, along with an edge detection and registration algorithm. Deviations between the predicted and measured bixel values are displayed as a matrix of percentage deviations, with maximum, mean, and standard deviation values calculated for each row, each column, and the entire map. Clinical applicability was investigated using IMRT plans delivered using a Varian accelerator with aS500 EPI. A set of 16 fluence maps from 6 different patients was selected, spanning a wide range of size and complexity, and employing both 0.5cm and 1cm bixel sizes. Results: The mean and standard deviation values for the bixel intensity differences were 1.9% and 1.4% for the 1cm bixel maps, and 3.2% and 1.8% for 0.5cm bixel maps. These differences are relatively small compared to the bixel intensity step size of 10%, making it relatively easy to assure that the correct intensity level was delivered to a bixel. Gross errors, such as an incorrect DMLC file or jaw setting, are easily detected by visual inspection or the resulting large mean and standard deviation values. Conclusion: This software tool facilitates accurate and expeditious IMRT delivery QA, allowing verification of all maps in a matter of minutes. The development of a more accurate convolution kernel will allow further reduction in comparison uncertainties. Clinical use will allow the establishment of a numerical threshold for a map to “pass” verification.
33(2006); http://dx.doi.org/10.1118/1.2241443View Description Hide Description
Purpose: To develop tools for automatically evaluating IMRTtreatments during delivery using exit detector data on a helical tomotherapy system. Methods and Materials:Treatment delivery sinograms and exit detector data were obtained following 6 treatment delivery sequences. The delivery sequences included 1.) No known error and nothing in the path of the beam, 2.) No known error with the couch in the path of the beam, 3.) No known error with an anthropomorphic pelvis phantom in the path of the beam, 4.) 1% error with nothing in the path of the beam, 5.) 1% error with the couch in the path of the beam, and 6.) 1% error with an anthropomorphic pelvis phantom in the path of the beam. A modeling technique was developed that could learn the attenuation relationships involved with the compressed data, thus distinguishing MLC errors from patient attenuation. A principal component modeling algorithm was developed for this purpose, employing the Hotellling's T2 statistic and the Q statistic. To develop the principle component model, the sinogram data were standardized, subtracted from one another and projected onto the model. A prospective analysis was also performed using the MLC delivery sequence to produce an expected detector signal with and without errors. Results: The Q‐statistic proved to be most useful for identifying errors in MLC openings, but correctly identified outliers and their contributing channel only when the model was trained with dissimilar (error‐less and error containing) data sets. The T‐statistic accounted for different attenuations present. Conclusions: Overall, the algorithms for analyzing compressed and uncompressed data proved to be useful in identifying errors as small as 1% in the MLCsequence.
33(2006); http://dx.doi.org/10.1118/1.2241444View Description Hide Description
Purpose: To validate a method of retrospective dose reconstruction that uses real‐time intra‐treatment patient motion data that is synchronized with MLC leaf positions during IMRT treatments. Method and Materials:IMRT fields from an IRB‐approved prostate protocol were delivered to a water‐equivalent phantom on a programmable translation stage. Kodak XV film was placed at 5 cm depth with an SSD of 95cm and marked to register it with the Monte Carlo (MC) calculation grid. Motion was synchronized with beam delivery by using the target signal to trigger motion. Film measurements were repeated for each beam while the phantom was stationary, and while moving with both idealized and clinically measured motion profiles. MLC leaf positions and fluence state for each beam were obtained from the Varian DynaLog files. MC dose accumulation was performed which incorporated the real‐time phantom motion, DynaLog files and beam state. Films were digitized and compared to the results of the MC calculations. Results: Film measurements in stationary phantoms were measured three times on two machines. The measured dose distributions were compared and showed an average difference of 0.38 +/− 1.53 cGy. The average difference between MC and film measurements of the moving phantom was 0.44 +/− 3.4 cGy and was independent of the motion profile. Measured dose patterns, for both stationary and moving phantoms, were generally well reproduced by MC dose accumulation, including tongue‐and‐groove and motion related features. Doses in moving and static phantoms were compared, for both films and simulations. The measured dose deviations due to motion were well‐characterized by the MC dose accumulation method and not significantly different when a static phantom was compared to MC calculation. Conclusion: Real‐time motion and machine data may be used to reconstruct the dose delivered to the target volume, and may serve as a basis for dynamic refinement of treatments.
- IMRT Optimization
33(2006); http://dx.doi.org/10.1118/1.2241525View Description Hide Description
Purpose: To investigate a new technique for multifield optimization of the biological effect (relative biological effectiveness times dose) for intensity modulated radiotherapy with scanned ion beams and to compare this method to an existing planning system for ions. Method and Materials: Our approach is based on the mixed irradiation formalism of the linear‐quadratic model using dose averaged mean values of alpha and sqrt(beta). We employ a novel objective function to directly optimize the biological effect rather than the physical dose. It is based on constraints in biologically effective dose for targets and organs at risk in close analogy to inverse planning for photons. The required biological input data are reduced to a minimum and are completely independent from the optimization itself. They can be derived from any radiobiological model or even from directly measured data. The new optimization method is fully integrated into the inverse treatment planning tool KonRad. Results: Comparisons with the TRiP98 treatment planning code are shown for spread‐out Bragg peaks as well as for three‐dimensional treatment plans for carbon ions, where all fields are optimized separately. While the agreement between both planning systems is very good, the calculation time is substantially reduced in KonRad. By enabling the multifield optimization in KonRad, the quality of the treatment plans and the sparing of healthy tissues can be clearly improved, which is demonstrated on several examples. Depending on the number of beam spots used, typical optimization times are between 10 and 60 minutes. Conclusion: The proposed system offers complete and fast inverse treatment planning for ions. Simultaneous multifield optimization of the biological effect can considerably enhance the resulting plans since it makes the best use of all possible degrees of freedom. Conflict of Interest: Research sponsored by Siemens Medical Solutions corporation.
33(2006); http://dx.doi.org/10.1118/1.2241526View Description Hide Description
Purpose:Treatment planning for intensity modulated proton therapy (IMPT) has traditionally used an approach in which the intensity and energy for each beamlet are modulated, which requires high dose‐rate beam scanning capabilities. The purpose of this work is to develop a new proton beam delivery method for IMPT without the need for high dose rate beam scanning. Method and Materials: In this study, an aperture‐based method to deliver a uniform dose to a target volume has been investigated. For a target with a flat back surface, a broad proton beam is collimated with an aperture conformed to the cross‐section of the target at a specific depth. The proton beam has a small energy spread to cover a 0.5–1.0 cm depth range. The mean energy and the weight of each proton beam are varied to produce a uniform dose distribution in the whole target volume. For an irregularly shaped target located in patient body, a compensator is used to provide equal beam path lengths to the back surface of the target. This will create an equivalent back surface for the target and then it can be treated in the same way as for the target with a flat back surface. A Fluka based Monte Carlo package has been used for dose calculation in aperture‐based IMPT treatment planning.Results: We have tested aperture‐based IMPT planning on a variety of patient cases. The results demonstrate that it can produce highly conformal dose distributions using only five to six apertures per beam direction. As compared with scanning beam delivery, our studies demonstrate that aperture‐based beam delivery can result in a significant reduction in both the number of beam segments and the number of monitor units. Conclusions: Aperture‐based IMPT optimization results in highly efficient treatment delivery while maintaining the dosimetric benefits of IMPT.
TU‐C‐224A‐03: 4D‐Image‐Guided Treatment Planning Optimization for Management of Organ Motion in Radiotherapy Planning33(2006); http://dx.doi.org/10.1118/1.2241527View Description Hide Description
The management of breathing‐registered IMRTtreatment planning is explored by direct incorporation of 4D images within the planning process. The movement of the voxels from one CT timeframe to another is “tracked” and modeled. A timestamp for each voxel is used to specify its position throughout the breathing cycle. A single treatment model incorporates planning constraints throughout multiple time periods. Robustness of the algorithm, plan quality, and potential clinical significance are evaluated.
4D‐CT scans of lung/liver cancer patients were acquired with different breathing phases(phases 0–9, 0:full‐inhale, 5:full‐exhale), Three treatment planning strategies are performed and compared. 1)Standard planning with a static PTV based on a single selected phase (control). 2)The Internal Target Volume (ITV) approach, where ITV is defined as the union of CTVs in all breathing phases. 3)Single‐stage‐4D‐image‐guided planning, where within a single treatment optimization model, planning constraints are incorporated on each voxel for each phase throughout the multiple‐phase period. Sophisticated computational optimization techniques are used to solve these models.
For both lung/liver cases, the static‐PTV‐plan results in unacceptable PTV‐underdose. Compared to ITV‐pamns, 4D‐image‐guided‐plans offer good coverage and comparable min‐PTV‐dose; while in lung, it reduces normal‐lung‐mean‐dose by 20%, heart‐mean‐dose by 20%, and esophagus‐max‐dose by 15%; and in liver, it reduces normal‐liver mean‐dose by 15% and other normal‐tissue mean‐dose by 20%, with improved PTV‐conformity of 10%.
4D‐Image‐Guided treatment planningoptimization can provide good PTV‐coverage plans, improve PTV‐underdose, and significantly reduce dose to organs‐at‐risk, especially those organs in the proximity of the tumor. Evidence of morbidity reduction to organs‐at‐risk is observed. The challenge involves the ability to solve a large‐scale treatment planning problem. With sophisticated mathematical optimization modeling and computational strategies, such planning is possible and can be made available for clinical use. Clinical studies are needed to validate the importance of our approach to treatment outcome.
TU‐C‐224A‐04: Simplifying Parameter Adjustment for Prostate IMRT Planning Using Sensitivity Analysis33(2006); http://dx.doi.org/10.1118/1.2241528View Description Hide Description
Purpose: To simplify the trial‐and‐error process of adjusting objective function parameters (e.g. weights, dose limits) in prostate IMRT planning, we present a feasibility study showing that an outer loop optimization over 6 automatically‐identified sensitive parameters can quickly and automatically determine parameters that lead to a plan meeting the clinical requirements. Method and Materials: We apply statistical sensitivity analysis to quantify the effect of each hand‐tunable parameter of the IMRT cost function on each clinical objective, automatically identifying those parameters with the strongest impact. Second, we globally optimize a plan quality score over the six most sensitive parameters in an outer loop to determine acceptable parameters, using a search algorithm based on multiscale random sampling. Results: Experiments on a 36‐patient dataset showed that a clinically acceptable five‐field 8640cGy prostate IMRT plan could be automatically determined in 35 minutes on the average in 87% of the cases. Compared to the plans from the planner's protocol default settings, the mean value of the minimum dose in PTV increased from 67.5% to 79.7%, and the mean value of PTV V95 coverage increased from 82.2% to 94.1%. The mean values of rectal wall V54%, rectal wall V87% and bladder wall V54% are 50.4%, 12.6% and 36.1%, respectively. The outer‐loop‐optimized plans met DVH constraints defining clinical acceptability and were comparable to manually‐determined plans. Confining the parameter search to the sensitive parameter set greatly improves the quality and speed of the outer‐loop optimization. Conclusion: The reduced‐order outer‐loop optimization can facilitate parameter selection for dose‐volume‐based IMRT objectives. It may also be applicable to other types of objective functions, and has the potential to ease the manual burden of IMRT planning in more complex sites (e.g. head and neck).
Supported by National Cancer Institute grant 5P01CA59017‐13 and the NSF Center for Subsurface Sensing and Imaging Systems, grant EEC‐9986821.
TU‐C‐224A‐05: Dosimetry Comparison of the Newly Implemented Multi‐Criteria Optimization Tool for IMRT Planning33(2006); http://dx.doi.org/10.1118/1.2241529View Description Hide Description
Purpose: To apply a biological model based algorithm for acquiring optimized IMRT planning solutions. This interactive planning tool will help users to select the best available plans in the IMRTsolution space.
Method and Materials:IMRT often is a time consuming iterative optimization process between evaluation of the dose distribution and redefinition of the object function. An IMRT planning optimization tool (Multi‐Criteria Optimization, MCO™) has been introduced for non‐clinical evaluation to acquire the best available solutions. Based on a Pareto's solution concept, this tool could search the solution space and offer users a limited set of deliverable IMRT plans. With this interactive process, users can set the target and critical structures dose constraints with the biological model (EUD) to obtain the best solution. We used Pinnacle system as the benchmark to compare the dosimetric gain from the MCO algorithm, DVH indicated excellent sparing with better PTV coverage is achievable from the MCO process in KonRad system.
Results: Dosimetric findings are summarized as 1) MCO optimizationtesting shows that much better dose distribution can be achieved compared to the current planning results (Fig. 1 and Fig. 2). Due to the confined solution space, the optimal results are easily achievable. 2) MCO with Pareto's approach is durable in the solution searching process. It is interactive with the graphical interface which the dose distribution along with the DVH can be compared simultaneously (Fig. 3). 3) IMRT dose optimization and summary based on the MCO methodology are very conceivable. With pre‐calculated IMRTsolutions, final results help users to select the best available plan from the solution domain in real time (Fig. 4). Conclusion: From this interactive MCO planning tool, we can calculate the best IMRT results in a very reasonable time frame. Human factors for determining an acceptable plan can be dramatically reduced.
TU‐C‐224A‐06: Exploiting the Full Potential of MLC Based Aperture Optimization Through Collimator Rotation33(2006); http://dx.doi.org/10.1118/1.2241530View Description Hide Description
Purpose: To investigate the benefits of MLC rotation in Direct Aperture Optimization (DAO) inverse treatment planning.Method and Materials: An alternative to fluence based inverse planning is to optimize directly the leaf positions and field weights of MLC apertures. Here we introduce a new technique called Rotating Aperture Optimization (RAO) which is based on an extension of DAO. Our technique differs from existing aperture based IMRT techniques in that the MLC is rotated in between each aperture. Treatment plans are generated for 10 nasopharynx recurrence patients with and without MLC rotation for 5 mm and 1cm leaf width MLCs. A comparison study is performed between RAO and DAO in order to assess the benefits of RAO over and above those available with fixed collimator angle DAO. Film verification is also performed to evaluate the accuracy of fixed and rotated collimator aperture delivery. Results: An analysis of the final cost values and DVHs indicate that plans generated with RAO are as good as or better than DAO while maintaining a smaller number of apertures and MU than conventional IMRT. In particular, RAO with the 1cm leaf width MLC is able to produce better plans than DAO with the 1 cm leaf width MLC and plans that are equivalent to DAO with the higher resolution 5mm leaf width MLC. Film verification results show that RAO is less sensitive to tongue and groove effects than DAO. Although delivery time is increased due to the collimator rotation speed this is a mechanical limit that could be easily overcome. Conclusion: Our results indicate that RAO is able to provide superior dose distributions, particularly with larger (1 cm) leaf width MLCs, while maintaining the lower MU and number of apertures afforded by the direct aperture approach.
Conflict of Interest: Supported in part by Varian Medical Systems.
33(2006); http://dx.doi.org/10.1118/1.2241531View Description Hide Description
Purpose: Intensity modulated radiation treatment planning for difficult cases is typically a time‐consuming manual search for a plan which gives an acceptable tradeoff between tumor coverage and critical structure sparing. We develop a method to calculate the efficient tradeoff surface of a multi‐objective IMRT inverse planning problem. This serves two purposes: to eliminate the time‐consuming manual search process, and to provide the treatment planners with the complete tradeoff information, allowing them to make more informed decisions. Method and Materials: We formulate a linear multi‐objective IMRTtreatment planning problem, the solution of which is a set of Pareto optimal treatment plans. Since each Pareto optimal plan involves a lengthy optimization, it is prudent to represent the complete surface with as few points as possible. Given the current set of Pareto surface plans, we use geometric considerations to formulate the optimization problem which computes the next plan. In this way, plans are added to the Pareto database until the surface is well represented. Results: The algorithm is applied to two clinical cases. For the prostate case, we display a tradeoff between the prostate coverage, femoral head sparing, and rectal sparing. For the skull‐based tumor, we display a tradeoff between tumor coverage, and the maximum doses of the chiasm, pituitary, and brainstem. Conclusion: We provide a method to efficiently generate Pareto surfaces for treatment planning, even when the number of organs to be traded off exceeds two or three. The method is applicable to any convex objective functions, including equivalent uniform dose, as well as the more standard quadratic penalty IMRT formulations. We expect that the clinical benefit of being able to visualize the tradeoff information — e.g. exactly how a decrease in critical structure dose degrades the tumor coverage — during the planning process will inspire a surge of research in this field.
TU‐C‐224A‐08: A Scientific Comparison of Inverse Treatment Plan Quality Using a Convex Non‐Linear Programming Model as a Function of Beam Quality and Beam Number33(2006); http://dx.doi.org/10.1118/1.2241532View Description Hide Description
Purpose: Recent advances in large scale fluence map optimizations for IMRT allow the use of large beam numbers that conform to the target to generate the desired target coverage while at the same time maintaining dose to critical organs below tolerance limits. Additionally, IMRT has effectively removed the need for high energy accelerator beams due to the excellent plan quality achievable with low beam quality. We investigate the diminishing returns in plan quality with increasing beam numbers and compare IMRTtreatment planning of 6MV and 60Co therapy dose models.Method and Materials: A convex non‐linear model was used to compare the plan quality, from dose volume histograms and fluence maps, for three treatment sites (H & N, CNS and prostate) for a 6MV and 60Co dose model. Plans were calculated for 5, 7, 9, 11, 17, 35 and 71 equidistant beam angles and quality assessed on target coverage (R95% > RRx) and organ sparing for each case. Results: Similar target coverage was achieved for 60Co as with 6MV and equivalent organ sparing was also observed for all three sites. Increasing the number of beams provided some improvement in organ sparing while maintaining target coverage conditions. Dose calculation times increased linearly with beam number and FMO calculations increased by up to 900% between 5 and 71 beams. Conclusion: We have demonstrated that IMRT plan quality using a 60Co dose model produces similar dose distributions to 6MV. We also show that plan quality does not show considerable improvement above 11 beams for IMRT and significant increases in the treatment planning times are observed extending the number of treatment beams to 71 beams.
This work supported in part by NSF grant DMI‐0457394 and the State of Florida DOH Grant 04‐NIR03.
33(2006); http://dx.doi.org/10.1118/1.2241533View Description Hide Description
Purpose: To computationally compare the plan quality provided by 3 different intensity modulation proton therapy (IMPT) techniques: 3D modulation; 2.5D modulation; and distal edge tracking (DET) with optimization. Plan quality, problem size, and efficiency were assessed. Method and Materials: The dose calculation was based a finite sized beamlet model, which was 1×1 cm2 at isocenter. The dose from each beamlet was the superposition/convolution of infinitesimal pencil beams falling within the beamlet. Ray tracing to each voxel was performed for each beamlet to find out the radiological depth to each voxel. In the 3D modulation algorithm, a stack of Bragg peaks were placed between the maximal and minimal radiological depths of targets passed by a beamlet with 2.5 mm spacing. The fluence of each Bragg peak was modulated independently by the optimizer. The placement of the Bragg peaks in the 2.5D algorithm was similar to the 3D algorithm, but the Bragg peaks belong to a beamlet were pre‐optimized to make a spread out Bragg (SOBP) whose dose level was equal to the prescription dose of the target it passed. The weights of SOBPs were optimization variables. In the DET algorithm, the Bragg peaks were set at the distal edge of the targets. Results: The 3D algorithm could produce the best plan with least beams, but the data size was large. The DET was most efficient in dose calculation and fluence map optimization, its plan quality was fair. The advantage of the 2.5D algorithm was in its small data size and efficiency in fluency map optimization.Conclusion: An appropriate algorithm should be selected to meet the trade‐off of plan quality versus computational and delivery efficiency.
This work was supported in part by Florida DOH Grant 04‐NIR03.
TU‐C‐224A‐10: A Novel Neighborhood for Local Search and Simulated Annealing Methods in Beam Orientation Optimization in IMRT33(2006); http://dx.doi.org/10.1118/1.2241534View Description Hide Description
Purpose: To select quality coplanar solutions to the beam orientation optimization (BOO) problem in intensity‐modulated radiation therapy(IMRT)treatment planning, and to demonstrate that high‐quality treatment plans can be obtained using fewer beams than typically used in equi‐spaced plans. Method and Materials: We consider the problem of obtaining quality 3‐ and 4‐beam coplanar radiation treatment plans. Two methods to obtaining these solutions are tested: the simulated annealing (SA) algorithm, which provides a global approach to the problem, and the Add/Drop heuristic, which provides a locally optimal solution. In the algorithms, a novel neighborhood is considered wherein a beam's neighborhood consists of a number of beams adjacent to the current beam plus a number of beams adjacent to the parallel‐opposed beam, which we call the “flip neighborhood”. For the SA algorithm, several methods of obtaining neighboring solutions and different cooling schedules are considered. The algorithms were tested on six head‐and‐neck cases using coplanar beams on a 5° grid. The resulting treatment plans were compared to the 5‐ and 7‐beam equi‐spaced plans typically used in practice. Results: While the 3‐beam treatment plans were poor in quality, the 4‐beam treatment plans obtained using both the SA method and the Add/Drop heuristic had comparable or improved quality to the 5‐ and 7‐beam equi‐spaced plans typically used in head‐and‐neck treatment.Conclusion: For head‐and‐neck cases, quality plans with fewer beams than standard 5–7 beam treatment plans can be obtained if BOO is applied. We also show that although the flip neighborhood increases run time for the Add/Drop heuristic (the run‐time of the simulated annealing algorithm is unchanged), it improves the FMO value for both the simulated annealing algorithm and the Add/Drop heuristic.
This work supported in part by NSF DMI‐0457394 and the NSF Alliances for Graduation Education and the Professoriate and Graduate Research Fellowship programs.
- IMRT Verification and QA II
33(2006); http://dx.doi.org/10.1118/1.2241771View Description Hide Description
Purpose: To measure the inter‐ and intra‐user variations in the manual alignment of calculated and measured doses in film‐based IMRT QA. Methods and Materials: Twenty (4 coronal and 16 axial) IMRT film based QA test cases were selected, each detailed by the “QA mode” in the IMPAC information management system. Films were shot in phantoms using a Varian 21EX, and do not contain fiducial marks. The treatment plans were created using the Pinnacle treatment planning system. Four of the films had known MLC problems, and were designed to fail the QA analysis. The films and corresponding calculated doses were placed on the internet for download. Participants were instructed to download the files, perform manual registration using the RIT113 software, save the registration films and return the test package to the investigators. Participants were instructed not to change the regions of interest and to indicate if each case would pass or fail their particular institutional criteria. Returned data was then analyzed for inter‐ and intra‐user variations in the manual alignment. Results: As of abstract submission, six respondents had been analyzed. The respondents had a wide range of passes and failures. Five out of the six respondents correctly identified the four films with known problems. One respondent incorrectly identified Patient #4 as a pass, but did note that the film was overly cropped. On average, the respondents indicated that seven of the films would not pass. Without fiducial marks on the film, each user placed the registration point in unique locations. As a result, each user had a unique QA analysis. Errors in the selection of registration points were directly related to false negatives. Conclusions: In order to minimize inter‐ and intra‐user variation, fiducial marks should be used to register the calculated and measured films in IMRT QA.