- Email: [email protected]
177 oral Integrating geometrical uncertainties in inverse treatment planning for IMRT to avoid a priori definition of planning margins for the tumor and the organs at risk
Proffered papers
while retaining linearity as an alternative approach to improve dose homogeneity in the target volumes, and to attempt to spare as many critical structures as possible. The goal of this work is to develop a very rapid global optimization approach that finds high quality dose distributions. Implementation of this model has demonstrated excellent results. We found globally optimal solutions for a head-and-neck 7 beam plan in less than three minutes of computational time on a single processor personal computer. These plans demonstrate excellent target coverage (>95 %), target homogeneity (<10% overdosing and <7% underdosing), and organ sparing. 174
oral
Optimisation of IMRT based on coverage probabilities derived from systematic setup error distributions C. Baum M. Birkner, A. Alber, F. Nuesslin Universitbtsklinik fur Radioonkologie, Abt. fearMedizinische Physik, TfJbingen, Germany Geometric accuracy of fractionated external beam radiotherapy is limited by internal organ motion and patient setup. Systematic geometric error can lead to massive underdosage in the clinical target volume(CTV) and overdosage in organs at risk(OAR). We present a method which incorporates probabilitiy distribution of systematic setup error into the optimisation of intensity modulated radiotherapy(IMRT) instead of a volume margin. A systematic setup error occurs if the probability is non-zero that some point outside planning CTV ought to belong to CTV in treatment room coordinate system. This interpretation leads to a alternative definition of the planning target volume(PTV): a point x in the treatment room coordinate system belongs to PTV if the probability that x belongs to CTV is greater than e.g. 2.5%. We use this idea of coverage probability density in optimisation and extent the PTV definition: every point has a certain probability of belonging to CTV. The distribution is derived from a patient population as a Gauss]an distribution of systematic errors with average zero and standard deviation S . The application of setup control protocols creates a narrower distribution for the corrected residual error after a few fractions. We applied our method to a group of prostate patients and compared four planning strategies with different PTV definitions. From portal imaging, we determined standard deviations S = 1.5 mm(lateral), 1.7 mm (AP) and 1.3 mm (cranio caudal) for uncorrected translational systematic setup error. For planning strategy Sl and S2, we applied the margin recipe 2.5 S + 0.7 s (van Herk et al.,Int. J. Rad. Oncology Biol. Phys.,Vol. 47/4,2000) and used CTV + 5mm margin for uncorrected(S1) and CTV + reduced margin for corrected residual error(S2) as PTV. We also convolved CTV with uncorrected systematic probability distribution (strategy S3) and CTV with corrected residual error probability distribution (strategy $4). The clinical results of S3 and S4 concerning rectum and CTV are both better than the results of S2 and much better than the results of Sl. Strategy 4 is the optimum strategy. With S4 we achieved the best rectal sparing, with Sl the lowest one. For S3, the number of geographical misses is increased in comparison to Sl. As a final result, we conclude that systematic error correction is neccessary and that a further PTV margin is not needed for our coverage probability approach. 175
oral
Deterministic and stochastic optimisation of physical and biological objective functions by means of the inverse Monte Carlo based TPS IKO M. Hartmann 1,2, L. Bogner 1 l Klinik und Poliklinik fuer Strahlentherapie, Universitaet Regensburg, Regensburg, Germany 2praxisgemeinschaft fuer Strahlentherapie PD. Dr. Staaff Dr. Bund, Bremen, Germany Introduction: Dose calculation inaccuracy due to approximations of conventional inverse TPS's doseengines like pencil beam or collapsed cone are compensated by the optimization algorithm and finally reflected in incorrect fluence modulation. A Monte Carlo dose engine with an adequate model of the beam delivery system can essentially improve the accuracy and thus as well the convergence error of an inverse planning algorithm for IMRT. Furthermore the usage of more and more complex objective functions with dose volume constraints or biological objectives can result in solution spaces which are not convex and where optimization with deterministic methods can get trapped in local minima. Material and methods: We developed the Monte Carlo based inverse planning system IKO (Inverse Kernel Optimization). The so-called inverse ker-
Wednesday, 17 September 2003
$69
nels (IK), which are calculated with very high precision by the implemented Monte Carlo code XVMC [Fippel, M. Med Phys 26 (1999) 1466-75] including the linac head model VEFM [Fippel, M. et al. Med Phys 30, 3 (2003) 301-11], enable a very fast dose calculation with high precision (inverse Kernel concept, see [Bogner et al. ESTRO 2003]). IKO contains various optimization algorithms. In addition to two different kind of gradient methods (quasi-Newton and sequential quadratic programming) a stochastical search engine (simulated annealing) is available. The system comprises a physical objective function with dose-volume constraints and a biological one based on the EUD-concept. Due to an integrated DICOM-RT tool to handle data transfer, the optimized fluence profiles are segmented externally by IMFAST (Siemens). Results and Conclusion: Using physical and biological objectives, the application of IKO is demonstrated by means of h&n cases, optimized with deterministic and stochastic methods. It can be shown, that inverse planning based on Monte Carlo dose calculation is feasible in an acceptable computation time for clinical routine. The implementation of different search engines and objective functions has the potential to compare various optimization methods in an easy and fast way. The simulated annealing algorithm offers the possibility to investigate the solution space of complex objective functions [poster Hartmann et al. ESTRO 2003]. 176
oral
Evaluation of the effect of geometrical uncertainties on dose distributions created with IMRT L.J. Bos, M. van Herk, B.J. Mijnheer, J. V. Lebesque, E.M.F. Damon Netherlands Cancer Institute/Antoni v Leeuwenhoek, Radiotherapy, Amsterdam, The Netherlands Purp0se: To assess the effect of geometrical uncertainties on tumour coverage and dose to the rectal wall for prostate irradiation techniques, using IMRT. Methods: For five patients, four different IMRT techniques were assessed, characterised by an increasing level of dose conformity. CTV=GTV=prostate+SV. The reference technique consisted of irradiating PTVl (CTV+ 10 mm in all directions) to 78 Gy. A coned-down boost was used irradiating PTVl to 68 Gy and boost PTV2 (CTV+ 0 mm towards the rectum, and 5 mm in other directions) with 10 Gy. Also, a simultaneous integrated boost (SlB) was created by forward and inverse planning (PTVl to 68 Gy and simultaneously PTV2 to 78 Gy). The 64.6 Gy and 74.1 Gy ]sodose surfaces (95%) encompassed PTVl and PTV2, respectively. The sensitivity of the plans for geometrical uncertainties (organ motion and set-up) was evaluated by means of in-house developed software, using uncertainty parameters typical for our institution. First, the dose distribution was blurred to mimic the effect of random errors. Systematic errors were modelled using a Monte Carlo approach, testing 5000 possible systematic errors. For each systematic error the dose distribution was evaluated for CTV and rectal wall. After simulation, probability distributions of the equivalent uniform dose (EUD) were derived; e.g., EUDp=90% =78 Gy means that there is a 90% probability that the EUD is at least 78 Gy. EUD was calculated with SF2=0.5 and n=0.25 for CTV and rectal wall, respectively. Results: The coned-down boost did not reduce the EUDp=90% of the CTV significantly, 0.2% on average, compared to the reference plan. The EUDp=I 0% of the rectal wall was reduced by 1.4% on average. For the SlB plans there was a trade-off between the EUD probability distributions for the CTV and rectal wall. The EUDp=90% of the CTV was reduced by 2.4% (SIB forward) and 3.0% (SIB inverse), on average, whilst EUDp=10% of the rectal wall was reduced by 8.5% (SlB forward) and 10.6% (SIB inverse), on average. For all patients, and each plan, the EUDp=90% of the CTV was at least 95% of the prescribed dose (74.1 Gy). The SIB plans resulted in the lowest EUDp=10% of the rectal wall. Conclusion: Prostate plans for IMRT, using a non-uniform margin from CTV to PTV for the boost volume, can lead to a significant reduction in rectal wail dose compared to a plan with a uniform margin. Meanwhile, a high probability of adequate CTV coverage is maintained. The effect is largest for the SlB. 177
oral
integrating geometrical uncertainties in inverse treatment planning for IMRT to avoid a priori definition of planning margins for the tumor and the organs at r i s k B.J.M. He]]men, J+C.J. De Boer, J.C. Stroom Erasmus MC - Daniel den Hoed Cancer Center, Radiation Oncology, Rotterdam, The Netherlands The aim of this work is to develop an algorithm for inverse treatment planning for IMRT that does not rely on a priori defined planning margins to
$70 Wednesday, 17 September 2003
account for geometrical uncertainties like patient set-up errors and internal organ motion. Plan generation is based on the delineated CTV (the PTVconcept is abandoned) and the organs at risk. In the inverse planning process, geometrical uncertainties are accounted for by distributions of potential geometrical errors, both for the CTV and for the organs at risk. Beam fluence profiles are optimized to find the best trade off between high dose delivery to the CTV and protection of the critical tissues, while simultaneously taking into account all geometrical uncertainties. Recently, we have developed methods for fast calculation of population mean dose volume histograms (mDVH) that account both for potential systematic deviations between the treatment plan geometry and the average treatment geometry, and for day-to-day variations in the treatment geometry (Stroom et al, IJROBP 43, 905-919, 1999). We have now integrated the concept of mDVH in an algorithm for inverse planning for IMRT. In an iterative method, a score function is minimized while trying to obey the imposed mDVH constraints for the CTV and the tissues at risk. In the score function, potential systematic errors in the CTV geometry and the geometry of the critical tissues are described by coverage probability distributions. A high weight is assigned to the criterion that on average (accounting for all uncertainties), 99 % of the CTV should receive a dose of at least 95 % of the prescribed tumor dose. This integrated inverse planning method generated high dose regions that are tight in the vicinity of critical tissues, and for compensation, wider in areas around the CTV that are further away from limiting structures. Apart from patient set-up errors, the developed algorithm also handles organ motion, organ deformation, and tumor delineation uncertainties. In summary, with this new method, knowledge of the magnitude and impact of geometrical uncertainties is optimally accounted for in the inverse planning process and compensated for with IMRT. The use of a priori defined planning margins to account for geometrical uncertainties (e.g. the much applied CTV-PTV margin) may be avoided. The method offers an alternative to the non-trivial definition of margins for organs at risk, as proposed in the ICRU-62 report. 178 oral A d a p t i v e radiation therapy for compensating errors in dose d e l i v e r y d u e to u n c e r t a i n t i e s in organ position and shape
j. Uhrdin 1, 2, H. Rehbinder 1, J. LOf1, A. Brahme 2 1RaySearch Laboratories AB, Stockholm, Sweden 2Karolinska Institutet, Medical radiation physics, Stockholm, Sweden The aim of curative radiation therapy optimization is to find a dose distribution that maximizes the probability of tumor cure. Today this can be accomplished with IMRT if the positions and shapes of the target and surrounding tissues are known at delivery. The importance of this knowledge is increased by the optimization due to steep dose gradients. Unfortunately, the internal patient geometry can vary both intra- and inter-fractionally and if this is not taken into account, it may cause serious deviations in delivered dose. In this study, the approach is to counteract these deviations by using an adaptive control algorithm that utilizes CT image data acquired during treatment delivery, for example obtained by cone beam radiotherapeutic CT. The regions of interest are contoured and treatment corrections are computed between fractions. As this may be time-consuming, it is more beneficial to use this off-line adaptive algorithm instead of making real-time corrections before delivery. The idea is to compute the accumulated delivered dose distribution within each region of interest. The computed accumulated dose is used to correct the fluence profiles in order to minimize deviations from the planned dose distribution. The image data is simulated by randomly moving a complex of tumor and organs at risk and assuming rigid motion. Surrounding tissue is subject to elastic deformations caused by the motion of the rigid complex. The simulated CT images are based on a patient geometry with a central tumor and an organ at risk surrounded by elastic soft tissue. The tumor position is assumed to be a stochastic quantity described by a Gaussian distribution. The adaptive algorithm has been evaluated for different standard deviations in tumor position: 1 - 10 mm, and with a stationary systematic error of 5 mm. The simulated patient is treated with 30 fractions and for each fraction a simulated CT image is taken. The final delivered accumulated dose distribution is barely deviating from the planned one. For lower standard deviations this is achieved after only three to five fractions. For higher standard deviations (7-10 mm), it is more difficult, and is not accomplished until the last ten fractions. The adaptive algorithm, when used together with CT in the treatment room, decreases the error in dose delivery and thereby makes it possible to increase the dose to the target.
Proffered papers
179
oral
Inverse IMRT planning with commercial software: a comparison of the pinnacle and plato planning algorithms H. Mayles 1, L. Cassapi2, P. Mayles 1, A. Scott 1, I. Syndikus 2, 7. Wolff 1 1Clatterbridge Centre for Oncology, Physics, Liverpool, United Kingdom 2Clatterbridge Centre for Oncology, Radiotherapy, Liverpool, United Kingdom Clatterbridge Centre for Oncology has been privileged to be involved in the early testing of the Plato ITP IMRT planning system over several years and are now introducing the Pinnacle system. Practical implementation of the plans was using an Elekta MLC. Each system has its own strengths (and also some weaknesses). Plato uses a fast gradient algorithm to produce its dose distributions and incorporates median filters to permit the reduction of the number segments. The final dose distribution uses the pencil beam algorithm. Pinnacle's calculation is slower but it has a number of features that may make the definition of a class solution more intuitive, such as the possibility of introducing constraints that must be met. The six alternative ways of setting objectives and constraints allow for considerable flexibility in designing protocols that will produce the best plans. The number of segments may also be minimised and small segments can be eliminated. A biological model is also being evaluated on the Pinnacle system. Initial tests show biological optimisation produces acceptable plans with a more simplified set of objectives. A class solution was derived with Plato ITP, seeking to treat the prostate and seminal vesicles and giving a simultaneous boost to the prostate alone, while constraining the dose given to the part of the PTV which overlaps the rectum. This has been in clinical use for over a year. Because of a number of differences in the philosophies of the two algorithms, implementing this class solution using Pinnacle was not simply a matter of applying the same constraints to the target and normal tissue volumes used for the Plato plan. (These differences need to be considered when defining protocols for use in different centres, for example when setting up clinical trials.) A similar distribution was eventually produced with both systems. Because of the differences in the sequencing algorithms, the leaf sequences produced by the two systems are very different, with greater use of the Elekta backup jaws in the Plato implementation. Verification of the planned dose distributions using film and ionisation chambers shows some differences in the implementation of the dose distribution. We will shortly be able to evaluate the implementation of the dose distributions on Varian linacs to provide a broader perspective of dose accuracy.
while retaining linearity as an alternative approach to improve dose homogeneity in the target volumes, and to attempt to spare as many critical structures as possible. The goal of this work is to develop a very rapid global optimization approach that finds high quality dose distributions. Implementation of this model has demonstrated excellent results. We found globally optimal solutions for a head-and-neck 7 beam plan in less than three minutes of computational time on a single processor personal computer. These plans demonstrate excellent target coverage (>95 %), target homogeneity (<10% overdosing and <7% underdosing), and organ sparing. 174
oral
Optimisation of IMRT based on coverage probabilities derived from systematic setup error distributions C. Baum M. Birkner, A. Alber, F. Nuesslin Universitbtsklinik fur Radioonkologie, Abt. fearMedizinische Physik, TfJbingen, Germany Geometric accuracy of fractionated external beam radiotherapy is limited by internal organ motion and patient setup. Systematic geometric error can lead to massive underdosage in the clinical target volume(CTV) and overdosage in organs at risk(OAR). We present a method which incorporates probabilitiy distribution of systematic setup error into the optimisation of intensity modulated radiotherapy(IMRT) instead of a volume margin. A systematic setup error occurs if the probability is non-zero that some point outside planning CTV ought to belong to CTV in treatment room coordinate system. This interpretation leads to a alternative definition of the planning target volume(PTV): a point x in the treatment room coordinate system belongs to PTV if the probability that x belongs to CTV is greater than e.g. 2.5%. We use this idea of coverage probability density in optimisation and extent the PTV definition: every point has a certain probability of belonging to CTV. The distribution is derived from a patient population as a Gauss]an distribution of systematic errors with average zero and standard deviation S . The application of setup control protocols creates a narrower distribution for the corrected residual error after a few fractions. We applied our method to a group of prostate patients and compared four planning strategies with different PTV definitions. From portal imaging, we determined standard deviations S = 1.5 mm(lateral), 1.7 mm (AP) and 1.3 mm (cranio caudal) for uncorrected translational systematic setup error. For planning strategy Sl and S2, we applied the margin recipe 2.5 S + 0.7 s (van Herk et al.,Int. J. Rad. Oncology Biol. Phys.,Vol. 47/4,2000) and used CTV + 5mm margin for uncorrected(S1) and CTV + reduced margin for corrected residual error(S2) as PTV. We also convolved CTV with uncorrected systematic probability distribution (strategy S3) and CTV with corrected residual error probability distribution (strategy $4). The clinical results of S3 and S4 concerning rectum and CTV are both better than the results of S2 and much better than the results of Sl. Strategy 4 is the optimum strategy. With S4 we achieved the best rectal sparing, with Sl the lowest one. For S3, the number of geographical misses is increased in comparison to Sl. As a final result, we conclude that systematic error correction is neccessary and that a further PTV margin is not needed for our coverage probability approach. 175
oral
Deterministic and stochastic optimisation of physical and biological objective functions by means of the inverse Monte Carlo based TPS IKO M. Hartmann 1,2, L. Bogner 1 l Klinik und Poliklinik fuer Strahlentherapie, Universitaet Regensburg, Regensburg, Germany 2praxisgemeinschaft fuer Strahlentherapie PD. Dr. Staaff Dr. Bund, Bremen, Germany Introduction: Dose calculation inaccuracy due to approximations of conventional inverse TPS's doseengines like pencil beam or collapsed cone are compensated by the optimization algorithm and finally reflected in incorrect fluence modulation. A Monte Carlo dose engine with an adequate model of the beam delivery system can essentially improve the accuracy and thus as well the convergence error of an inverse planning algorithm for IMRT. Furthermore the usage of more and more complex objective functions with dose volume constraints or biological objectives can result in solution spaces which are not convex and where optimization with deterministic methods can get trapped in local minima. Material and methods: We developed the Monte Carlo based inverse planning system IKO (Inverse Kernel Optimization). The so-called inverse ker-
Wednesday, 17 September 2003
$69
nels (IK), which are calculated with very high precision by the implemented Monte Carlo code XVMC [Fippel, M. Med Phys 26 (1999) 1466-75] including the linac head model VEFM [Fippel, M. et al. Med Phys 30, 3 (2003) 301-11], enable a very fast dose calculation with high precision (inverse Kernel concept, see [Bogner et al. ESTRO 2003]). IKO contains various optimization algorithms. In addition to two different kind of gradient methods (quasi-Newton and sequential quadratic programming) a stochastical search engine (simulated annealing) is available. The system comprises a physical objective function with dose-volume constraints and a biological one based on the EUD-concept. Due to an integrated DICOM-RT tool to handle data transfer, the optimized fluence profiles are segmented externally by IMFAST (Siemens). Results and Conclusion: Using physical and biological objectives, the application of IKO is demonstrated by means of h&n cases, optimized with deterministic and stochastic methods. It can be shown, that inverse planning based on Monte Carlo dose calculation is feasible in an acceptable computation time for clinical routine. The implementation of different search engines and objective functions has the potential to compare various optimization methods in an easy and fast way. The simulated annealing algorithm offers the possibility to investigate the solution space of complex objective functions [poster Hartmann et al. ESTRO 2003]. 176
oral
Evaluation of the effect of geometrical uncertainties on dose distributions created with IMRT L.J. Bos, M. van Herk, B.J. Mijnheer, J. V. Lebesque, E.M.F. Damon Netherlands Cancer Institute/Antoni v Leeuwenhoek, Radiotherapy, Amsterdam, The Netherlands Purp0se: To assess the effect of geometrical uncertainties on tumour coverage and dose to the rectal wall for prostate irradiation techniques, using IMRT. Methods: For five patients, four different IMRT techniques were assessed, characterised by an increasing level of dose conformity. CTV=GTV=prostate+SV. The reference technique consisted of irradiating PTVl (CTV+ 10 mm in all directions) to 78 Gy. A coned-down boost was used irradiating PTVl to 68 Gy and boost PTV2 (CTV+ 0 mm towards the rectum, and 5 mm in other directions) with 10 Gy. Also, a simultaneous integrated boost (SlB) was created by forward and inverse planning (PTVl to 68 Gy and simultaneously PTV2 to 78 Gy). The 64.6 Gy and 74.1 Gy ]sodose surfaces (95%) encompassed PTVl and PTV2, respectively. The sensitivity of the plans for geometrical uncertainties (organ motion and set-up) was evaluated by means of in-house developed software, using uncertainty parameters typical for our institution. First, the dose distribution was blurred to mimic the effect of random errors. Systematic errors were modelled using a Monte Carlo approach, testing 5000 possible systematic errors. For each systematic error the dose distribution was evaluated for CTV and rectal wall. After simulation, probability distributions of the equivalent uniform dose (EUD) were derived; e.g., EUDp=90% =78 Gy means that there is a 90% probability that the EUD is at least 78 Gy. EUD was calculated with SF2=0.5 and n=0.25 for CTV and rectal wall, respectively. Results: The coned-down boost did not reduce the EUDp=90% of the CTV significantly, 0.2% on average, compared to the reference plan. The EUDp=I 0% of the rectal wall was reduced by 1.4% on average. For the SlB plans there was a trade-off between the EUD probability distributions for the CTV and rectal wall. The EUDp=90% of the CTV was reduced by 2.4% (SIB forward) and 3.0% (SIB inverse), on average, whilst EUDp=10% of the rectal wall was reduced by 8.5% (SlB forward) and 10.6% (SIB inverse), on average. For all patients, and each plan, the EUDp=90% of the CTV was at least 95% of the prescribed dose (74.1 Gy). The SIB plans resulted in the lowest EUDp=10% of the rectal wall. Conclusion: Prostate plans for IMRT, using a non-uniform margin from CTV to PTV for the boost volume, can lead to a significant reduction in rectal wail dose compared to a plan with a uniform margin. Meanwhile, a high probability of adequate CTV coverage is maintained. The effect is largest for the SlB. 177
oral
integrating geometrical uncertainties in inverse treatment planning for IMRT to avoid a priori definition of planning margins for the tumor and the organs at r i s k B.J.M. He]]men, J+C.J. De Boer, J.C. Stroom Erasmus MC - Daniel den Hoed Cancer Center, Radiation Oncology, Rotterdam, The Netherlands The aim of this work is to develop an algorithm for inverse treatment planning for IMRT that does not rely on a priori defined planning margins to
$70 Wednesday, 17 September 2003
account for geometrical uncertainties like patient set-up errors and internal organ motion. Plan generation is based on the delineated CTV (the PTVconcept is abandoned) and the organs at risk. In the inverse planning process, geometrical uncertainties are accounted for by distributions of potential geometrical errors, both for the CTV and for the organs at risk. Beam fluence profiles are optimized to find the best trade off between high dose delivery to the CTV and protection of the critical tissues, while simultaneously taking into account all geometrical uncertainties. Recently, we have developed methods for fast calculation of population mean dose volume histograms (mDVH) that account both for potential systematic deviations between the treatment plan geometry and the average treatment geometry, and for day-to-day variations in the treatment geometry (Stroom et al, IJROBP 43, 905-919, 1999). We have now integrated the concept of mDVH in an algorithm for inverse planning for IMRT. In an iterative method, a score function is minimized while trying to obey the imposed mDVH constraints for the CTV and the tissues at risk. In the score function, potential systematic errors in the CTV geometry and the geometry of the critical tissues are described by coverage probability distributions. A high weight is assigned to the criterion that on average (accounting for all uncertainties), 99 % of the CTV should receive a dose of at least 95 % of the prescribed tumor dose. This integrated inverse planning method generated high dose regions that are tight in the vicinity of critical tissues, and for compensation, wider in areas around the CTV that are further away from limiting structures. Apart from patient set-up errors, the developed algorithm also handles organ motion, organ deformation, and tumor delineation uncertainties. In summary, with this new method, knowledge of the magnitude and impact of geometrical uncertainties is optimally accounted for in the inverse planning process and compensated for with IMRT. The use of a priori defined planning margins to account for geometrical uncertainties (e.g. the much applied CTV-PTV margin) may be avoided. The method offers an alternative to the non-trivial definition of margins for organs at risk, as proposed in the ICRU-62 report. 178 oral A d a p t i v e radiation therapy for compensating errors in dose d e l i v e r y d u e to u n c e r t a i n t i e s in organ position and shape
j. Uhrdin 1, 2, H. Rehbinder 1, J. LOf1, A. Brahme 2 1RaySearch Laboratories AB, Stockholm, Sweden 2Karolinska Institutet, Medical radiation physics, Stockholm, Sweden The aim of curative radiation therapy optimization is to find a dose distribution that maximizes the probability of tumor cure. Today this can be accomplished with IMRT if the positions and shapes of the target and surrounding tissues are known at delivery. The importance of this knowledge is increased by the optimization due to steep dose gradients. Unfortunately, the internal patient geometry can vary both intra- and inter-fractionally and if this is not taken into account, it may cause serious deviations in delivered dose. In this study, the approach is to counteract these deviations by using an adaptive control algorithm that utilizes CT image data acquired during treatment delivery, for example obtained by cone beam radiotherapeutic CT. The regions of interest are contoured and treatment corrections are computed between fractions. As this may be time-consuming, it is more beneficial to use this off-line adaptive algorithm instead of making real-time corrections before delivery. The idea is to compute the accumulated delivered dose distribution within each region of interest. The computed accumulated dose is used to correct the fluence profiles in order to minimize deviations from the planned dose distribution. The image data is simulated by randomly moving a complex of tumor and organs at risk and assuming rigid motion. Surrounding tissue is subject to elastic deformations caused by the motion of the rigid complex. The simulated CT images are based on a patient geometry with a central tumor and an organ at risk surrounded by elastic soft tissue. The tumor position is assumed to be a stochastic quantity described by a Gaussian distribution. The adaptive algorithm has been evaluated for different standard deviations in tumor position: 1 - 10 mm, and with a stationary systematic error of 5 mm. The simulated patient is treated with 30 fractions and for each fraction a simulated CT image is taken. The final delivered accumulated dose distribution is barely deviating from the planned one. For lower standard deviations this is achieved after only three to five fractions. For higher standard deviations (7-10 mm), it is more difficult, and is not accomplished until the last ten fractions. The adaptive algorithm, when used together with CT in the treatment room, decreases the error in dose delivery and thereby makes it possible to increase the dose to the target.
Proffered papers
179
oral
Inverse IMRT planning with commercial software: a comparison of the pinnacle and plato planning algorithms H. Mayles 1, L. Cassapi2, P. Mayles 1, A. Scott 1, I. Syndikus 2, 7. Wolff 1 1Clatterbridge Centre for Oncology, Physics, Liverpool, United Kingdom 2Clatterbridge Centre for Oncology, Radiotherapy, Liverpool, United Kingdom Clatterbridge Centre for Oncology has been privileged to be involved in the early testing of the Plato ITP IMRT planning system over several years and are now introducing the Pinnacle system. Practical implementation of the plans was using an Elekta MLC. Each system has its own strengths (and also some weaknesses). Plato uses a fast gradient algorithm to produce its dose distributions and incorporates median filters to permit the reduction of the number segments. The final dose distribution uses the pencil beam algorithm. Pinnacle's calculation is slower but it has a number of features that may make the definition of a class solution more intuitive, such as the possibility of introducing constraints that must be met. The six alternative ways of setting objectives and constraints allow for considerable flexibility in designing protocols that will produce the best plans. The number of segments may also be minimised and small segments can be eliminated. A biological model is also being evaluated on the Pinnacle system. Initial tests show biological optimisation produces acceptable plans with a more simplified set of objectives. A class solution was derived with Plato ITP, seeking to treat the prostate and seminal vesicles and giving a simultaneous boost to the prostate alone, while constraining the dose given to the part of the PTV which overlaps the rectum. This has been in clinical use for over a year. Because of a number of differences in the philosophies of the two algorithms, implementing this class solution using Pinnacle was not simply a matter of applying the same constraints to the target and normal tissue volumes used for the Plato plan. (These differences need to be considered when defining protocols for use in different centres, for example when setting up clinical trials.) A similar distribution was eventually produced with both systems. Because of the differences in the sequencing algorithms, the leaf sequences produced by the two systems are very different, with greater use of the Elekta backup jaws in the Plato implementation. Verification of the planned dose distributions using film and ionisation chambers shows some differences in the implementation of the dose distribution. We will shortly be able to evaluate the implementation of the dose distributions on Varian linacs to provide a broader perspective of dose accuracy.