- Email: [email protected]
Refining 3D Dose Distribution at a Regional and Voxel Level in Biological IMRT Treatment Planning
I. J. Radiation Oncology d Biology d Physics
S758
Volume 78, Number 3, Supplement, 2010
to the 3D convolution O(MN3) for the CCCS method, where N is the number of voxels in each dimension and M is the number of cone directions. Thus it is expedient to use the proposed method for point dose calculation in IMRT application. Author Disclosure: M. Chen, TomoTherapy Inc, A. Employment; Q. Chen, TomoTherapy Inc., A. Employment; G. Olivera, TomoTherapy Inc., A. Employment; W. Lu, TomoTherapy Inc., A. Employment.
3236
Systematic and Random Dosimetric Impact Studies on 100 Helical Tomotherapy Cancer Cases
W. He, C. Shi, L. Vazquez Q, E. Dzintars, N. Papanikolaou Cancer Therapy and Research Center, San Antonio, TX Purpose/Objective(s): To assess the dosimetric consequence of inter-fractional setup shifts on helical tomotherapy plans with selfdeveloped independent dose calculation software MU-Tomo. Materials/Methods: A 2nd dose validation software for helical TomoTherapy , called MU-Tomo, has been developed to independently validates point dose upon archived patient documents, initial coordinates and planned dose of the point of calculation, and common dosimetric functions. This work has studied a hundred helical tomotherapy patients from five different treatment sites (30 prostate, 26 head and neck, 18 lung, 17 pelvis, and 9 brain patients). The daily setup shifts were quantified and grouped into systematic (mean daily setup shift for each patient) and random shifts (shifts after corresponding systematic shifts subtraction). Both Systematic and random shifts were incorporated into MU-Tomo to evaluate the systematic and random variations of dosimetric consequences, separately. Results: Systematic variations showed dose deviation with the largest one -10.02% compared to the planned dose and overall SD 3%. Mean random variations showed dose deviation with the largest one -5.65% compared to the planned dose and overall SD 1.9%. According to the ANOVA analyses, random dosimetric variations were found significantly different among specific patient, while systematic dosimetric variations were significantly different between head and neck and brain group and body group. No significant differences were discovered among specific patients for systematic variations, and no significant differences were observed within each of the two groups for random variations. Dosimetric consequences are not significantly correlated with treatment fraction number according to the Pearson correlation analysis. By comparing doses without any shift and with the random shift, the overall dosimetric impacts to each patient are small with the mean value -0.0053% and SD of 1.11%, and 99% of the averaged variation results were within 3.5%. Conclusions: For helical tomotherapy modality, the overall dosimetric impact from random variations is small; instead, systematic shifts cause more dosimetric impact. Author Disclosure: W. He, None; C. Shi, None; L. Vazquez Q, None; E. Dzintars, None; N. Papanikolaou, None.
3237
4D IMRT Inverse Planning of Scarsity-enhanced Compressed Sensing using Fluence Map Reordering Method
T. Kim1,2, T. Suh2, L. Xing1 1
Stanford University, Stanford, CA, 2The Catholic University of Korea, Seoul, Republic of Korea
Purpose/Objective(s): Various fluence map optimization algorithms for 3D IMRT inverse planning have been actively studied. Unlike a beamlet-based algorithm, which generates overly complex fluence maps, a compressed sensing (CS) technique using a regularization-based algorithm maintains the conformity of the target dose distribution while reducing complexity to a clinically-acceptable level. Because the CS technique enforces the piecewise-constant condition in fluence maps, sparsity may be utilized in the fluence maps, which results in a reduction in the complexity. For improvement of 4D IMRT inverse planning, we suggest a sparsity-enhanced CS technique using a fluence map reordering method. In this study, we demonstrate that this proposed method provides a significant benefit in target dose uniformity in clinical 4D IMRT inverse plan. Materials/Methods: In this study, we included two patient cases for 3D IMRT inverse planning: a head and neck patient and a prostate patent. For 4D IMRT inverse planning, we used a simulated lung malignancy case with 2 different target positions. To utilize sparsity of the fluence maps, we employed a total-variation (TV) regularization-based CS algorithm and an intensity-reordering process in the TV regularization using a reordering map obtained from the previous IMRT inverse plan. The proposed algorithm was tested in Matlab, using the MOSEK optimization software package (http://www.mosek.com). We evaluated the proposed algorithm in dose volume histograms (DVH) and dose distribution maps of the 4D IMRT inverse planning with/without the reordering process. Results: The CS techniques using a TV constraint reduced the complexity of the fluence maps, while preserving the edges. This is in stark comparison to the beamlet-based algorithm, which resulted in complex beamlet profiles which became practically undeliverable with a small number of segments. The proposed CS method with TV regularization including a fluence map reordering method provided a significant improvement in DVHs. This proposed method is shown to cooperate with additional temporal information in 4D IMRT inverse planning. Conclusions: This study demonstrated the significant improvement in target dose uniformity and in sparing organs at risk by using the sparsity-enhanced CS 4D IMRT inverse planning using a fluence map reordering method, which providing clinically applicable plan. Author Disclosure: T. Kim, None; T. Suh, None; L. Xing, None.
3238
Refining 3D Dose Distribution at a Regional and Voxel Level in Biological IMRT Treatment Planning
P. Lougovski , L. Xing Stanford University School of Medicine, Stanford, CA Purpose/Objective(s): Biological treatment planning based on the equivalent uniform dose (EUD) demonstrates reduced treatment toxicity and better tumor control. However, EUD based approaches are lacking tools for controlling and refining resultant dose distributions on a voxel level. Here we introduce a method enabling regional dose distribution manipulations for biological IMRT treatment planning.
Proceedings of the 52nd Annual ASTRO Meeting Materials/Methods: Two dose distributions with the same EUD are EUD equivalent. However, one might be more clinically acceptable than the other. Embedding tools that find clinically more acceptable solutions and allow dose refining into biologic optimization is thus very important. We propose to use a hybrid biologic/physical dose optimization approach. We first identify region(s) where dose refinement is desired. For them a quadratic dose-volume objective function with a homogeneous prescription is formulated. The remaining structures are included into the planning via EUD dose constraints. If the resultant optimal dose distribution does not fulfill planner’s clinical dose-volume criteria then the prescription is adjusted and the problem is re-optimized. The adjustment mechanism accounts for the intrinsic dosimetric inequality between voxels which is ignored by the EUD and the objective function. Results: A clinical prostate case was used to test our method. The PTV was selected as a primary target for dose refinement. The rectum and the bladder were incorporated into the optimization via EUD constraints. By adjusting voxel prescriptions in the PTV iteratively, a dose improvement was obtained for the PTV and dose reduction was achieved in the critical structures comparing to a conventional dose-volume based objective plan with the same PTV coverage. For instance, we demonstrate up to 25% reduction in the mean dose to the rectum. Conclusions: We show that regional dose distribution control and refinement can be achieved for biological optimization. It is relevant in the context of 3D dose sculpting based on the EUD. Author Disclosure: P. Lougovski, None; L. Xing, None.
3239
Assessing the Level of Modulation of IMRT Fields
1
M. Nauta , E. Villareal Barajas1,2, M. Tambasco1,2 1
University of Calgary, Calgary, AB, Canada, 2Tom Baker Cancer Centre, Calgary, AB, Canada
Purpose/Objective(s): It is well known that greater modulation in intensity modulated radiation therapy (IMRT) fields increases the length of time the radiotherapy beam must operate, and longer beam-on times increase the total leakage and scattered radiation dose to the patient. We investigated the use of fractal dimension as a metric for quantifying the degree of modulation in computer planned IMRT treatment fields. Materials/Methods: We compared the variation, power spectrum, and variogram methods for computing the fractal dimension and found the variogram method to be the most suitable for assessing modulation complexity. We used a commercial treatment planning system (Eclipse, Varian Medical Systems, Palo Alto, CA) to plan 48 prescribed treatment fields using two different constraint settings that produced treatment fields with high and moderate degrees of modulation. We also used an electronic portal imaging device (PortalVision aS1000, Varian Medical Systems, Palo Alto, CA) to measure the radiation dose distribution for 12 of the 48 treatment fields and compared them to the predicted portal dose fields calculated by the planning system. Results: The fractal dimensions of the highly modulated fields ranged from 0.02 to 0.15 greater than the corresponding moderately modulated fields, and the Kolmogorov-Smirnov test demonstrated a statistically significant difference between the two groups (p\ 0.0001). Although the fractal dimensions of the measured fields were all greater than the corresponding planned fields, the amount by which they were greater (0.01 to 0.03) is within the measurement uncertainty for fractal dimension (±0.04). Conclusions: Fractal dimension is not sensitive enough for identifying differences between planned and measured fields; however, it is a useful metric for identifying planned treatment fields that are over-modulated. Therefore, fractal dimension provides one with the opportunity to adjust highly modulated fields at the treatment planning stage before they are sent to the treatment machine for quality control or patient treatment. Author Disclosure: M. Nauta, None; E. Villareal Barajas, None; M. Tambasco, None.
3240
Automated IMRT Treatment Planning
B. A. Pierburg1, R. Kashani2, K. W. Baker1, A. N. Lindsey1, M. B. Watts1, D. Yang2, K. L. Moore2 1
Barnes-Jewish Hospital, St. Louis, MO, 2Washington University School of Medicine, St. Louis, MO
Purpose/Objective(s): Inter-clinician variability can cause large inconsistencies in IMRT plan quality, resulting in suboptimal treatment plans with respect to target coverage, normal tissue sparing, or both. The purpose of this work is to assess the performance of automated IMRT planning (autoplanning) routines for prostate cancer and the more difficult case of bilateral head/neck cancer. Materials/Methods: Autoplanning routines were created in the Philips Pinnacle 9.0 treatment planning system. This implementation relied on Pinnacle ‘scripts,’ which are assemblies of internal commands stored as text files. When invoked on a new patient dataset, the autoplanning scripts create various target and OAR planning structures, set the beams and dose prescription, and load customized IMRT objectives to start the optimization. The target objectives are based on prescription dose, while OAR objectives are determined from a model, developed from prior cases that analyze geometric properties of the target and OARs to predict mean doses. To determine their clinical utility, the autoplanning routines were compared to ten retrospective plans each for both prostate and bilateral head/neck. The autoplanned cases were normalized to clinical cases by equalizing PTV coverage at the prescription dose. Comparison metrics focused on the target dose homogeneity and OAR dose sparing. Results: The prostate script takes \90 secs to run, while the head/neck script requires ?3 min. Depending on the complexity of the case, the initial optimization sequence ranges from 10-20 mins to converge to an initial solution. In the prostate comparison study, one autoplanned solution bettered the clinically approved plan without intervention, while the remaining cases requiring minimal efforts (10 mins-3 hrs) to meet or exceed the clinically approved plans in target homogeneity and rectum/bladder sparing. In the bilateral head/neck comparison study, the autoplanning solutions required more frequent intercession during optimization due to the model’s occasional overestimation of OAR sparing to the more complicated geometric structures in this region. The ?4 hours of refinement was required to achieve clinical standards of PTV dose homogeneity, significantly less than the average manual planning time for this site. The autoplanning solutions met or exceeded the clinical plans in sparing all of the numerous critical organs in this region. Conclusions: Autoplanning shows great potential in setting the basis for IMRT plans which meet the quality of manually developed plans with minimal intervention, even for a complicated site like bilateral head/neck. Large efficiency gains and robust
S759
S758
Volume 78, Number 3, Supplement, 2010
to the 3D convolution O(MN3) for the CCCS method, where N is the number of voxels in each dimension and M is the number of cone directions. Thus it is expedient to use the proposed method for point dose calculation in IMRT application. Author Disclosure: M. Chen, TomoTherapy Inc, A. Employment; Q. Chen, TomoTherapy Inc., A. Employment; G. Olivera, TomoTherapy Inc., A. Employment; W. Lu, TomoTherapy Inc., A. Employment.
3236
Systematic and Random Dosimetric Impact Studies on 100 Helical Tomotherapy Cancer Cases
W. He, C. Shi, L. Vazquez Q, E. Dzintars, N. Papanikolaou Cancer Therapy and Research Center, San Antonio, TX Purpose/Objective(s): To assess the dosimetric consequence of inter-fractional setup shifts on helical tomotherapy plans with selfdeveloped independent dose calculation software MU-Tomo. Materials/Methods: A 2nd dose validation software for helical TomoTherapy , called MU-Tomo, has been developed to independently validates point dose upon archived patient documents, initial coordinates and planned dose of the point of calculation, and common dosimetric functions. This work has studied a hundred helical tomotherapy patients from five different treatment sites (30 prostate, 26 head and neck, 18 lung, 17 pelvis, and 9 brain patients). The daily setup shifts were quantified and grouped into systematic (mean daily setup shift for each patient) and random shifts (shifts after corresponding systematic shifts subtraction). Both Systematic and random shifts were incorporated into MU-Tomo to evaluate the systematic and random variations of dosimetric consequences, separately. Results: Systematic variations showed dose deviation with the largest one -10.02% compared to the planned dose and overall SD 3%. Mean random variations showed dose deviation with the largest one -5.65% compared to the planned dose and overall SD 1.9%. According to the ANOVA analyses, random dosimetric variations were found significantly different among specific patient, while systematic dosimetric variations were significantly different between head and neck and brain group and body group. No significant differences were discovered among specific patients for systematic variations, and no significant differences were observed within each of the two groups for random variations. Dosimetric consequences are not significantly correlated with treatment fraction number according to the Pearson correlation analysis. By comparing doses without any shift and with the random shift, the overall dosimetric impacts to each patient are small with the mean value -0.0053% and SD of 1.11%, and 99% of the averaged variation results were within 3.5%. Conclusions: For helical tomotherapy modality, the overall dosimetric impact from random variations is small; instead, systematic shifts cause more dosimetric impact. Author Disclosure: W. He, None; C. Shi, None; L. Vazquez Q, None; E. Dzintars, None; N. Papanikolaou, None.
3237
4D IMRT Inverse Planning of Scarsity-enhanced Compressed Sensing using Fluence Map Reordering Method
T. Kim1,2, T. Suh2, L. Xing1 1
Stanford University, Stanford, CA, 2The Catholic University of Korea, Seoul, Republic of Korea
Purpose/Objective(s): Various fluence map optimization algorithms for 3D IMRT inverse planning have been actively studied. Unlike a beamlet-based algorithm, which generates overly complex fluence maps, a compressed sensing (CS) technique using a regularization-based algorithm maintains the conformity of the target dose distribution while reducing complexity to a clinically-acceptable level. Because the CS technique enforces the piecewise-constant condition in fluence maps, sparsity may be utilized in the fluence maps, which results in a reduction in the complexity. For improvement of 4D IMRT inverse planning, we suggest a sparsity-enhanced CS technique using a fluence map reordering method. In this study, we demonstrate that this proposed method provides a significant benefit in target dose uniformity in clinical 4D IMRT inverse plan. Materials/Methods: In this study, we included two patient cases for 3D IMRT inverse planning: a head and neck patient and a prostate patent. For 4D IMRT inverse planning, we used a simulated lung malignancy case with 2 different target positions. To utilize sparsity of the fluence maps, we employed a total-variation (TV) regularization-based CS algorithm and an intensity-reordering process in the TV regularization using a reordering map obtained from the previous IMRT inverse plan. The proposed algorithm was tested in Matlab, using the MOSEK optimization software package (http://www.mosek.com). We evaluated the proposed algorithm in dose volume histograms (DVH) and dose distribution maps of the 4D IMRT inverse planning with/without the reordering process. Results: The CS techniques using a TV constraint reduced the complexity of the fluence maps, while preserving the edges. This is in stark comparison to the beamlet-based algorithm, which resulted in complex beamlet profiles which became practically undeliverable with a small number of segments. The proposed CS method with TV regularization including a fluence map reordering method provided a significant improvement in DVHs. This proposed method is shown to cooperate with additional temporal information in 4D IMRT inverse planning. Conclusions: This study demonstrated the significant improvement in target dose uniformity and in sparing organs at risk by using the sparsity-enhanced CS 4D IMRT inverse planning using a fluence map reordering method, which providing clinically applicable plan. Author Disclosure: T. Kim, None; T. Suh, None; L. Xing, None.
3238
Refining 3D Dose Distribution at a Regional and Voxel Level in Biological IMRT Treatment Planning
P. Lougovski , L. Xing Stanford University School of Medicine, Stanford, CA Purpose/Objective(s): Biological treatment planning based on the equivalent uniform dose (EUD) demonstrates reduced treatment toxicity and better tumor control. However, EUD based approaches are lacking tools for controlling and refining resultant dose distributions on a voxel level. Here we introduce a method enabling regional dose distribution manipulations for biological IMRT treatment planning.
Proceedings of the 52nd Annual ASTRO Meeting Materials/Methods: Two dose distributions with the same EUD are EUD equivalent. However, one might be more clinically acceptable than the other. Embedding tools that find clinically more acceptable solutions and allow dose refining into biologic optimization is thus very important. We propose to use a hybrid biologic/physical dose optimization approach. We first identify region(s) where dose refinement is desired. For them a quadratic dose-volume objective function with a homogeneous prescription is formulated. The remaining structures are included into the planning via EUD dose constraints. If the resultant optimal dose distribution does not fulfill planner’s clinical dose-volume criteria then the prescription is adjusted and the problem is re-optimized. The adjustment mechanism accounts for the intrinsic dosimetric inequality between voxels which is ignored by the EUD and the objective function. Results: A clinical prostate case was used to test our method. The PTV was selected as a primary target for dose refinement. The rectum and the bladder were incorporated into the optimization via EUD constraints. By adjusting voxel prescriptions in the PTV iteratively, a dose improvement was obtained for the PTV and dose reduction was achieved in the critical structures comparing to a conventional dose-volume based objective plan with the same PTV coverage. For instance, we demonstrate up to 25% reduction in the mean dose to the rectum. Conclusions: We show that regional dose distribution control and refinement can be achieved for biological optimization. It is relevant in the context of 3D dose sculpting based on the EUD. Author Disclosure: P. Lougovski, None; L. Xing, None.
3239
Assessing the Level of Modulation of IMRT Fields
1
M. Nauta , E. Villareal Barajas1,2, M. Tambasco1,2 1
University of Calgary, Calgary, AB, Canada, 2Tom Baker Cancer Centre, Calgary, AB, Canada
Purpose/Objective(s): It is well known that greater modulation in intensity modulated radiation therapy (IMRT) fields increases the length of time the radiotherapy beam must operate, and longer beam-on times increase the total leakage and scattered radiation dose to the patient. We investigated the use of fractal dimension as a metric for quantifying the degree of modulation in computer planned IMRT treatment fields. Materials/Methods: We compared the variation, power spectrum, and variogram methods for computing the fractal dimension and found the variogram method to be the most suitable for assessing modulation complexity. We used a commercial treatment planning system (Eclipse, Varian Medical Systems, Palo Alto, CA) to plan 48 prescribed treatment fields using two different constraint settings that produced treatment fields with high and moderate degrees of modulation. We also used an electronic portal imaging device (PortalVision aS1000, Varian Medical Systems, Palo Alto, CA) to measure the radiation dose distribution for 12 of the 48 treatment fields and compared them to the predicted portal dose fields calculated by the planning system. Results: The fractal dimensions of the highly modulated fields ranged from 0.02 to 0.15 greater than the corresponding moderately modulated fields, and the Kolmogorov-Smirnov test demonstrated a statistically significant difference between the two groups (p\ 0.0001). Although the fractal dimensions of the measured fields were all greater than the corresponding planned fields, the amount by which they were greater (0.01 to 0.03) is within the measurement uncertainty for fractal dimension (±0.04). Conclusions: Fractal dimension is not sensitive enough for identifying differences between planned and measured fields; however, it is a useful metric for identifying planned treatment fields that are over-modulated. Therefore, fractal dimension provides one with the opportunity to adjust highly modulated fields at the treatment planning stage before they are sent to the treatment machine for quality control or patient treatment. Author Disclosure: M. Nauta, None; E. Villareal Barajas, None; M. Tambasco, None.
3240
Automated IMRT Treatment Planning
B. A. Pierburg1, R. Kashani2, K. W. Baker1, A. N. Lindsey1, M. B. Watts1, D. Yang2, K. L. Moore2 1
Barnes-Jewish Hospital, St. Louis, MO, 2Washington University School of Medicine, St. Louis, MO
Purpose/Objective(s): Inter-clinician variability can cause large inconsistencies in IMRT plan quality, resulting in suboptimal treatment plans with respect to target coverage, normal tissue sparing, or both. The purpose of this work is to assess the performance of automated IMRT planning (autoplanning) routines for prostate cancer and the more difficult case of bilateral head/neck cancer. Materials/Methods: Autoplanning routines were created in the Philips Pinnacle 9.0 treatment planning system. This implementation relied on Pinnacle ‘scripts,’ which are assemblies of internal commands stored as text files. When invoked on a new patient dataset, the autoplanning scripts create various target and OAR planning structures, set the beams and dose prescription, and load customized IMRT objectives to start the optimization. The target objectives are based on prescription dose, while OAR objectives are determined from a model, developed from prior cases that analyze geometric properties of the target and OARs to predict mean doses. To determine their clinical utility, the autoplanning routines were compared to ten retrospective plans each for both prostate and bilateral head/neck. The autoplanned cases were normalized to clinical cases by equalizing PTV coverage at the prescription dose. Comparison metrics focused on the target dose homogeneity and OAR dose sparing. Results: The prostate script takes \90 secs to run, while the head/neck script requires ?3 min. Depending on the complexity of the case, the initial optimization sequence ranges from 10-20 mins to converge to an initial solution. In the prostate comparison study, one autoplanned solution bettered the clinically approved plan without intervention, while the remaining cases requiring minimal efforts (10 mins-3 hrs) to meet or exceed the clinically approved plans in target homogeneity and rectum/bladder sparing. In the bilateral head/neck comparison study, the autoplanning solutions required more frequent intercession during optimization due to the model’s occasional overestimation of OAR sparing to the more complicated geometric structures in this region. The ?4 hours of refinement was required to achieve clinical standards of PTV dose homogeneity, significantly less than the average manual planning time for this site. The autoplanning solutions met or exceeded the clinical plans in sparing all of the numerous critical organs in this region. Conclusions: Autoplanning shows great potential in setting the basis for IMRT plans which meet the quality of manually developed plans with minimal intervention, even for a complicated site like bilateral head/neck. Large efficiency gains and robust
S759