- Email: [email protected]
Verification and Validation of GPU-based TomoTherapy Dose Calculation Engine
I. J. Radiation Oncology d Biology d Physics
S732
Volume 78, Number 3, Supplement, 2010
voxel of the region of interest on the US image. For urethra and rectal wall, a rigid-body registration was used based on visicoils and prostate contours. The urethra was not clearly visible on CT and rectal wall on US image was deformed because of the probe making deformable registration difficult. By applying these deformation/transformation, HDR dose distribution on US image was mapped to CT images coordinate system and accumulated with EBRT dose distribution. A linear-quadratic model was applied to calculate the total equivalent dose in 2 Gy fractions (EQD2) with a/b = 1.5 for prostate and 3 for prostatic urethra and rectal wall. Cumulative dose-volume parameters such as V80, V100, V125 and V150 were evaluated for urethra and rectal wall. Cumulative EQD2 to 99%, 95% and 50% volume were determined for prostate. Results: EQD2 for HDR fractionation with an a/b = 1.5 and 3 was 72 Gy and 56.7 Gy respectively. The mean prostate volume for EB and BT fractions were 55.4 ± 13.2 cc and 45.7 ± 9.6 cc. The mean difference between EB and BT prostate volume was 6.0 ± 10.3 cc. The average cumulative EQD2 to 99%, 95% and 50% of prostate volume based on rigid and deformable registration was 84.0 ± 12.9, 94.1 ± 15.6, 131.1 ± 7 Gy and 116.6 ± 12.6, 123.1 ± 7.5, 141.9 ± 6.0 Gy respectively. The mean V80, V100, V125 and V150 based on cumulative doses for urethra were 98.6 ± 3.1, 92.9 ± 6.9, 3.2 ± 7.7, 1.2 ± 3.0%. The corresponding V80 and V100 for rectal wall were 12.8 ± 12.1 and 1.7 ± 3.2%. Conclusions: Large deformations exist in soft tissue anatomy between BT and EB fractions that have a significant impact on cumulative dose distribution. Mean variations of up to 28% were seen for cumulative doses in prostate based on rigid and deformable registration. A post-implant CT without the US probe is required to accurately delineate and deform normal structures such as rectal wall, urethra and bladder. Author Disclosure: N. Tyagi, None; E. Sebastian, None; J. Liang, None; D. Yan, None; M. Ghilezan, None; A. Martinez, None.
3179
Effect of Pre-treatment Imaging Dose on the Radiosensitivity of Tumor Cells
I. U. Ahmad, V. Singh-Gupta, C. Yunker, J. Burmeister, M. C. Joiner, G. G. Hillman Wayne State University/Karmanos Cancer Institute, Detroit, MI Purpose/Objective(s): Imaging of patient anatomy is routinely performed immediately prior to radiation therapy (RT) to confirm the localization of external beam RT delivery to the tumor. Prior to each fraction, CT imaging using kV or MV beams is often used to provide soft tissue differentiation. However, this delivers significantly higher imaging doses than planar imaging. The dose from this CT scan typically ranges from 1-4 cGy depending on the modality used. The time lapse between the CT scan and RT treatment also varies. The goal of this study was to determine whether these low doses of radiation used for image guidance prior to therapeutic doses affects RT efficacy for prostate carcinoma and non-small cell lung carcinoma cells and to determine the influence of the time gap between image guidance and treatment. Materials/Methods: We selected human prostate carcinoma (PC3) and non-small cell lung carcinoma (A549) cell lines. All radiation was delivered using a Pantak orthovoltage unit. A dose of 2 cGy was used to mimic the CT scan and was administered at intervals of 2, 5, and 8 minutes prior to therapeutic doses of radiation to simulate time between CT scan and RT. We tested RT doses of 2 Gy, 4 Gy, and 6 Gy for PC3 cells and 2 Gy, 4 Gy, and 8 Gy for A549 cells. Control groups included non-irradiated cells and cells irradiated only with RT doses of 2-8 Gy. Treated cells were tested in triplicates in a clonogenic assay. The plating efficiency (PE) was determined for each test sample and the surviving fraction (SF) was calculated by normalizing each test PE to the control PE for non-irradiated cells. Cells were also assessed for DNA damage by measuring double strand breaks (dsb) with a gH2AX immunofluorescence assay. Results: In the clonogenic assay, the mean SF for PC3 cells was 0.495 ± 0.039 [mean (n = 3) ± SD] for 2 Gy RT; 0.270 ± 0.031 for 4 Gy RT; and 0.132 ± 0.017 for 6 Gy RT. The mean SF for A549 cells was 0.536 ± 0.036 for 2 Gy RT; 0.284 ± 0.030 for 4 Gy RT; and 0.103 ± 0.006 for 8 Gy RT. For both cell lines, radiation with the pre-treatment dose of 2 cGy prior to 2 Gy, 4 Gy, 6 Gy, and 8 Gy RT did not significantly change the mean SF. Time intervals of 2, 5 and 8 minutes between 2 cGy and RT at each dose tested (28 Gy) had no effect on the SF with no increase or decrease in cell survival compared to cells irradiated with 2-8 Gy RT only. The number of dsb foci and frequency of gH2AX positive cells increased with RT dose. However, no significant differences were observed in the intensity and number of dsb between cells receiving 2 cGy prior to RT and cells treated with RT alone. Conclusions: These studies suggest that low dose CT scans used for daily pre-treatment positioning prior to RT delivery do not affect the efficacy of subsequent therapeutic doses of RT for these prostate carcinoma and non-small cell lung carcinoma cell lines. Author Disclosure: I.U. Ahmad, None; V. Singh-Gupta, None; C. Yunker, None; J. Burmeister, None; M.C. Joiner, None; G.G. Hillman, None.
3180
Verification and Validation of GPU-based TomoTherapy Dose Calculation Engine
Q. Chen, M. Chen, D. Henderson, Y. Chen, G. Olivera, W. Lu Tomotherapy Inc, Madison, WI Purpose/Objective(s): Dose calculation is essential in radiation therapy treatment planning. TomoTherapy uses accurate Collapsed Cone Convolution/superposition (CCCS) algorithm as its dose calculation engine. As the CCCS calculation is computationally demanding, computer cluster with 28 cpu cores or 56 cpu cores is provided for the optimization and dose calculation. The recent generation of graphic processing units (GPUs) possesses great computing power due to its massively parallel architecture. We have developed an ultrafast GPU-based TomotherapyÒ dose calculation engine capable of performing near real-time dose calculation with a single PC. There are many innovations on the algorithm to make it adapt to the unique architecture of the GPU. This study shows the extensive verification and validation work performed for this new TomoTherapy dose calculation engine. Materials/Methods: Twenty TomoDirect, 12 TomoHelical, and a few other special plans (including StatRT, 3DCRT, etc.) was planned and delivered on the Tomo Cheese phantom. For each delivery, a Kodak EDR2 film is used to record the 2D dose distribution and 2 Extradin A1SL ion chambers are used to measure point dose at 2 locations. The GPU dose engine runs on a single NVIDIA GTX295 card and the original commercial version runs on a CPU cluster with 56 cores cluster. The dose calculated with the GPU dose engine is compared to the ion chamber and film measurements as well as to that calculated with the original TomotherapyÒ dose calculation engine.
Proceedings of the 52nd Annual ASTRO Meeting Results: The GPU algorithm on one GTX295 is 8-16 times faster than original dose engine running on a blade cluster with 56 cpu cores (2.33GHz Xeon5148). This translates to a speedup of 400-800 over a single Xeon5148 core. The new GPU-based CCCS dose engine produces doses that are generally within 1.5 Gamma (1%, 1 mm) of original dose engine. For TomoHelical cases we tested, the maximum Gamma we observe is 1.24, the majority cases has maximum Gamma less than 1. For TomoDirect cases, only 4 out of 20 cases have less than 0.001% of their total voxels with Gamma value greater than 1.5. The discrepancy between the ion chamber measurement and the new GPU dose engine is small. For plans with field width of 2.5 cm and 5.0 cm, the discrepancies are within 1% with mean absolute error of 0.41% and 0.54% respectively. For plans with field width of 1.0 cm, the discrepancy is still within the passing criterion of 3%. All the cases passed the criterion for film measurement (3%, 3 mm for 95% of pixels) as well. Conclusions: We performed extensive verification and validation work on ultrafast GPU-based TomoTherapy dose engine. The new engine is able to perform much faster dose calculation with a single PC. The new engine will make a lot of time sensitive applications such as on-line ART and real- time optimization feasible. Author Disclosure: Q. Chen, TomoTherapy Inc, A. Employment; M. Chen, TomoTherapy Inc, A. Employment; D. Henderson, TomoTherapy Inc, A. Employment; Y. Chen, TomoTherapy Inc, A. Employment; G. Olivera, TomoTherapy Inc, A. Employment; W. Lu, TomoTheraphy, A. Employment.
3181
Interfraction Geometric Variations of the Mandible and Its Dosimetric Impact during Intensity Modulated Radiotherapy for Head and Neck Cancer
B. Hu, K. Kainz, D. Wang, X. A. Li Medical College of Wisconsin, Milwaukee, WI Purpose/Objective(s): The mandible may move independently relative to the rest of the head and neck anatomy. Using daily CT images, we study the geometric variability of the mandible position and its dosimetric impact upon IMRT for the head and neck cancer. Materials/Methods: For 10 patients treated for disease proximal to the mandible using helical tomotherapy or step-and-shoot IMRT, daily pre-treatment MVCTs (Hi-Art, TomoTherapy) or CT-on-rails KVCTs (CTVision, Siemens) were acquired. The prescribed dose to the PTV ranged from 60 to 70 Gy among the plans. The mandible was delineated on each daily-CT image postregistration with the planning CT. The coordinates of the midpoint between the bilateral mental foramina (MBMF) and the center of mass (COM) of the mandible contour were calculated and recorded for each daily CT image. The daily delivered dose distribution and dose volume histograms (DVHs) were calculated using TomoTherapy Planned Adaptive Software or CMS XiO. The mandible dose volume parameters of V70 (volume receiving 70 Gy), V60, V50, and maximum point dose for the daily CT images were compared to those from the planning CT. Results: Among all patients, the overall geometric variation of the mandible (combined over all three dimensions) was 1.15 ± 0.58 cm (mean ± SD) with a range of 0.07-2.72 cm for the MBMF, and 1.19 ± 0.73 cm with a range of 0.01-3.02 cm for the COM. The mandible dose distribution shows significant variation due to mandible position variation. For earlier fractions the daily mandible DVHs are consistent with the plan DVH, but they increase among later fractions. Although there is no significant difference in mandible maximum point dose between the delivered and the planned values, the overall V70, V60, and V50 are significantly different from the planned values. For one tomotherapy patient, total variation in the translational motion of the mandible was 0.51 ± 0.19 cm (range, 0.07-0.81 cm) for the MBMF and 0.29 ± 0.19 cm (range, 0.01-0.66 cm) for the COM; the V70, V60, and V50 showed differences as high as 3.6% ± 2.8%, 6.5% ± 15.7%, and 8.4% ± 5.9%, relative to the planned values of 4.0%, 12.7%, and 16.6%, respectively. For one CT-on-rails patient, total variation for the MBMF was 0.85 ± 0.25 cm (range, 0.21-1.49 cm) and for the COM was 0.69 ± 0.30 cm (range, 0.14-1.74cm); the V70, V60, and V50 showed differences as low as -0.3% ± 0.3%, -5.2% ± 1.7%, and -10.6% ± 3.6%, relative to the plan values of 0.5%, 9.0%, and 28.3%, respectively. Conclusions: Interfraction geometric variations of the mandible were observed during daily image guided radiotherapy for head and neck cancer. Since the target volumes are proximal to the mandible, such variations can result in significant dosimetric impact and may be accounted for by an adaptive planning strategy. Author Disclosure: B. Hu, None; K. Kainz, None; D. Wang, None; X.A. Li, None.
3182
Dose-Based Real-time Treatment Monitoring for IMRT
Y. Jiang, X. Gu, C. Men, X. Jia, R. Li, J. H. Lewis, S. B. Jiang University of California, San Diego, La Jolla, CA Purpose/Objective(s): To reconstruct in real-time the dose distribution of intensity-modulated radiation therapy (IMRT) delivered to a patient. It provides a dose-based real-time treatment monitoring during treatment delivery and thus improves the treatment quality and safety. Materials/Methods: First, we use a GPU-based FSPB algorithm to calculate the dose deposition coefficients Dij for all IMRT fields. Dij is the dose contribution of the ith beamlets to the jth voxel on the patient image, and it is pre-computed and stored in the memory. During the treatment delivery, MLC leaf positions will be read in real-time from either the MLC controller or the cine EPID images. Then the delivered dose distribution is generated using the Dij matrix and the leaf positions. In this work, a typical IMRT HN patient with seven split-fields was selected for testing purpose. It results in 14 individual subfields each with separated MLC DynaLog file. We simulate this real-time process using the MLC DynaLog files in retrospective. We extract the leaf positions and fractional MU every 50 ms from the MLC DynaLog files and convert them to a beam mask Bi using an in-house software written in C. The accumulated dose distribution delivered up to a time point is then computed by convolving Dij with Bi. Results: It takes on average less than 1 s to pre-calculate Dij for each beam. And it takes less than 0.05s to update the accumulative dose distribution on patient’s geometry with Dij matrix and beam mask Bi from MLC DynaLog files. It meets the real-time requirement for clinic environment.
S733
S732
Volume 78, Number 3, Supplement, 2010
voxel of the region of interest on the US image. For urethra and rectal wall, a rigid-body registration was used based on visicoils and prostate contours. The urethra was not clearly visible on CT and rectal wall on US image was deformed because of the probe making deformable registration difficult. By applying these deformation/transformation, HDR dose distribution on US image was mapped to CT images coordinate system and accumulated with EBRT dose distribution. A linear-quadratic model was applied to calculate the total equivalent dose in 2 Gy fractions (EQD2) with a/b = 1.5 for prostate and 3 for prostatic urethra and rectal wall. Cumulative dose-volume parameters such as V80, V100, V125 and V150 were evaluated for urethra and rectal wall. Cumulative EQD2 to 99%, 95% and 50% volume were determined for prostate. Results: EQD2 for HDR fractionation with an a/b = 1.5 and 3 was 72 Gy and 56.7 Gy respectively. The mean prostate volume for EB and BT fractions were 55.4 ± 13.2 cc and 45.7 ± 9.6 cc. The mean difference between EB and BT prostate volume was 6.0 ± 10.3 cc. The average cumulative EQD2 to 99%, 95% and 50% of prostate volume based on rigid and deformable registration was 84.0 ± 12.9, 94.1 ± 15.6, 131.1 ± 7 Gy and 116.6 ± 12.6, 123.1 ± 7.5, 141.9 ± 6.0 Gy respectively. The mean V80, V100, V125 and V150 based on cumulative doses for urethra were 98.6 ± 3.1, 92.9 ± 6.9, 3.2 ± 7.7, 1.2 ± 3.0%. The corresponding V80 and V100 for rectal wall were 12.8 ± 12.1 and 1.7 ± 3.2%. Conclusions: Large deformations exist in soft tissue anatomy between BT and EB fractions that have a significant impact on cumulative dose distribution. Mean variations of up to 28% were seen for cumulative doses in prostate based on rigid and deformable registration. A post-implant CT without the US probe is required to accurately delineate and deform normal structures such as rectal wall, urethra and bladder. Author Disclosure: N. Tyagi, None; E. Sebastian, None; J. Liang, None; D. Yan, None; M. Ghilezan, None; A. Martinez, None.
3179
Effect of Pre-treatment Imaging Dose on the Radiosensitivity of Tumor Cells
I. U. Ahmad, V. Singh-Gupta, C. Yunker, J. Burmeister, M. C. Joiner, G. G. Hillman Wayne State University/Karmanos Cancer Institute, Detroit, MI Purpose/Objective(s): Imaging of patient anatomy is routinely performed immediately prior to radiation therapy (RT) to confirm the localization of external beam RT delivery to the tumor. Prior to each fraction, CT imaging using kV or MV beams is often used to provide soft tissue differentiation. However, this delivers significantly higher imaging doses than planar imaging. The dose from this CT scan typically ranges from 1-4 cGy depending on the modality used. The time lapse between the CT scan and RT treatment also varies. The goal of this study was to determine whether these low doses of radiation used for image guidance prior to therapeutic doses affects RT efficacy for prostate carcinoma and non-small cell lung carcinoma cells and to determine the influence of the time gap between image guidance and treatment. Materials/Methods: We selected human prostate carcinoma (PC3) and non-small cell lung carcinoma (A549) cell lines. All radiation was delivered using a Pantak orthovoltage unit. A dose of 2 cGy was used to mimic the CT scan and was administered at intervals of 2, 5, and 8 minutes prior to therapeutic doses of radiation to simulate time between CT scan and RT. We tested RT doses of 2 Gy, 4 Gy, and 6 Gy for PC3 cells and 2 Gy, 4 Gy, and 8 Gy for A549 cells. Control groups included non-irradiated cells and cells irradiated only with RT doses of 2-8 Gy. Treated cells were tested in triplicates in a clonogenic assay. The plating efficiency (PE) was determined for each test sample and the surviving fraction (SF) was calculated by normalizing each test PE to the control PE for non-irradiated cells. Cells were also assessed for DNA damage by measuring double strand breaks (dsb) with a gH2AX immunofluorescence assay. Results: In the clonogenic assay, the mean SF for PC3 cells was 0.495 ± 0.039 [mean (n = 3) ± SD] for 2 Gy RT; 0.270 ± 0.031 for 4 Gy RT; and 0.132 ± 0.017 for 6 Gy RT. The mean SF for A549 cells was 0.536 ± 0.036 for 2 Gy RT; 0.284 ± 0.030 for 4 Gy RT; and 0.103 ± 0.006 for 8 Gy RT. For both cell lines, radiation with the pre-treatment dose of 2 cGy prior to 2 Gy, 4 Gy, 6 Gy, and 8 Gy RT did not significantly change the mean SF. Time intervals of 2, 5 and 8 minutes between 2 cGy and RT at each dose tested (28 Gy) had no effect on the SF with no increase or decrease in cell survival compared to cells irradiated with 2-8 Gy RT only. The number of dsb foci and frequency of gH2AX positive cells increased with RT dose. However, no significant differences were observed in the intensity and number of dsb between cells receiving 2 cGy prior to RT and cells treated with RT alone. Conclusions: These studies suggest that low dose CT scans used for daily pre-treatment positioning prior to RT delivery do not affect the efficacy of subsequent therapeutic doses of RT for these prostate carcinoma and non-small cell lung carcinoma cell lines. Author Disclosure: I.U. Ahmad, None; V. Singh-Gupta, None; C. Yunker, None; J. Burmeister, None; M.C. Joiner, None; G.G. Hillman, None.
3180
Verification and Validation of GPU-based TomoTherapy Dose Calculation Engine
Q. Chen, M. Chen, D. Henderson, Y. Chen, G. Olivera, W. Lu Tomotherapy Inc, Madison, WI Purpose/Objective(s): Dose calculation is essential in radiation therapy treatment planning. TomoTherapy uses accurate Collapsed Cone Convolution/superposition (CCCS) algorithm as its dose calculation engine. As the CCCS calculation is computationally demanding, computer cluster with 28 cpu cores or 56 cpu cores is provided for the optimization and dose calculation. The recent generation of graphic processing units (GPUs) possesses great computing power due to its massively parallel architecture. We have developed an ultrafast GPU-based TomotherapyÒ dose calculation engine capable of performing near real-time dose calculation with a single PC. There are many innovations on the algorithm to make it adapt to the unique architecture of the GPU. This study shows the extensive verification and validation work performed for this new TomoTherapy dose calculation engine. Materials/Methods: Twenty TomoDirect, 12 TomoHelical, and a few other special plans (including StatRT, 3DCRT, etc.) was planned and delivered on the Tomo Cheese phantom. For each delivery, a Kodak EDR2 film is used to record the 2D dose distribution and 2 Extradin A1SL ion chambers are used to measure point dose at 2 locations. The GPU dose engine runs on a single NVIDIA GTX295 card and the original commercial version runs on a CPU cluster with 56 cores cluster. The dose calculated with the GPU dose engine is compared to the ion chamber and film measurements as well as to that calculated with the original TomotherapyÒ dose calculation engine.
Proceedings of the 52nd Annual ASTRO Meeting Results: The GPU algorithm on one GTX295 is 8-16 times faster than original dose engine running on a blade cluster with 56 cpu cores (2.33GHz Xeon5148). This translates to a speedup of 400-800 over a single Xeon5148 core. The new GPU-based CCCS dose engine produces doses that are generally within 1.5 Gamma (1%, 1 mm) of original dose engine. For TomoHelical cases we tested, the maximum Gamma we observe is 1.24, the majority cases has maximum Gamma less than 1. For TomoDirect cases, only 4 out of 20 cases have less than 0.001% of their total voxels with Gamma value greater than 1.5. The discrepancy between the ion chamber measurement and the new GPU dose engine is small. For plans with field width of 2.5 cm and 5.0 cm, the discrepancies are within 1% with mean absolute error of 0.41% and 0.54% respectively. For plans with field width of 1.0 cm, the discrepancy is still within the passing criterion of 3%. All the cases passed the criterion for film measurement (3%, 3 mm for 95% of pixels) as well. Conclusions: We performed extensive verification and validation work on ultrafast GPU-based TomoTherapy dose engine. The new engine is able to perform much faster dose calculation with a single PC. The new engine will make a lot of time sensitive applications such as on-line ART and real- time optimization feasible. Author Disclosure: Q. Chen, TomoTherapy Inc, A. Employment; M. Chen, TomoTherapy Inc, A. Employment; D. Henderson, TomoTherapy Inc, A. Employment; Y. Chen, TomoTherapy Inc, A. Employment; G. Olivera, TomoTherapy Inc, A. Employment; W. Lu, TomoTheraphy, A. Employment.
3181
Interfraction Geometric Variations of the Mandible and Its Dosimetric Impact during Intensity Modulated Radiotherapy for Head and Neck Cancer
B. Hu, K. Kainz, D. Wang, X. A. Li Medical College of Wisconsin, Milwaukee, WI Purpose/Objective(s): The mandible may move independently relative to the rest of the head and neck anatomy. Using daily CT images, we study the geometric variability of the mandible position and its dosimetric impact upon IMRT for the head and neck cancer. Materials/Methods: For 10 patients treated for disease proximal to the mandible using helical tomotherapy or step-and-shoot IMRT, daily pre-treatment MVCTs (Hi-Art, TomoTherapy) or CT-on-rails KVCTs (CTVision, Siemens) were acquired. The prescribed dose to the PTV ranged from 60 to 70 Gy among the plans. The mandible was delineated on each daily-CT image postregistration with the planning CT. The coordinates of the midpoint between the bilateral mental foramina (MBMF) and the center of mass (COM) of the mandible contour were calculated and recorded for each daily CT image. The daily delivered dose distribution and dose volume histograms (DVHs) were calculated using TomoTherapy Planned Adaptive Software or CMS XiO. The mandible dose volume parameters of V70 (volume receiving 70 Gy), V60, V50, and maximum point dose for the daily CT images were compared to those from the planning CT. Results: Among all patients, the overall geometric variation of the mandible (combined over all three dimensions) was 1.15 ± 0.58 cm (mean ± SD) with a range of 0.07-2.72 cm for the MBMF, and 1.19 ± 0.73 cm with a range of 0.01-3.02 cm for the COM. The mandible dose distribution shows significant variation due to mandible position variation. For earlier fractions the daily mandible DVHs are consistent with the plan DVH, but they increase among later fractions. Although there is no significant difference in mandible maximum point dose between the delivered and the planned values, the overall V70, V60, and V50 are significantly different from the planned values. For one tomotherapy patient, total variation in the translational motion of the mandible was 0.51 ± 0.19 cm (range, 0.07-0.81 cm) for the MBMF and 0.29 ± 0.19 cm (range, 0.01-0.66 cm) for the COM; the V70, V60, and V50 showed differences as high as 3.6% ± 2.8%, 6.5% ± 15.7%, and 8.4% ± 5.9%, relative to the planned values of 4.0%, 12.7%, and 16.6%, respectively. For one CT-on-rails patient, total variation for the MBMF was 0.85 ± 0.25 cm (range, 0.21-1.49 cm) and for the COM was 0.69 ± 0.30 cm (range, 0.14-1.74cm); the V70, V60, and V50 showed differences as low as -0.3% ± 0.3%, -5.2% ± 1.7%, and -10.6% ± 3.6%, relative to the plan values of 0.5%, 9.0%, and 28.3%, respectively. Conclusions: Interfraction geometric variations of the mandible were observed during daily image guided radiotherapy for head and neck cancer. Since the target volumes are proximal to the mandible, such variations can result in significant dosimetric impact and may be accounted for by an adaptive planning strategy. Author Disclosure: B. Hu, None; K. Kainz, None; D. Wang, None; X.A. Li, None.
3182
Dose-Based Real-time Treatment Monitoring for IMRT
Y. Jiang, X. Gu, C. Men, X. Jia, R. Li, J. H. Lewis, S. B. Jiang University of California, San Diego, La Jolla, CA Purpose/Objective(s): To reconstruct in real-time the dose distribution of intensity-modulated radiation therapy (IMRT) delivered to a patient. It provides a dose-based real-time treatment monitoring during treatment delivery and thus improves the treatment quality and safety. Materials/Methods: First, we use a GPU-based FSPB algorithm to calculate the dose deposition coefficients Dij for all IMRT fields. Dij is the dose contribution of the ith beamlets to the jth voxel on the patient image, and it is pre-computed and stored in the memory. During the treatment delivery, MLC leaf positions will be read in real-time from either the MLC controller or the cine EPID images. Then the delivered dose distribution is generated using the Dij matrix and the leaf positions. In this work, a typical IMRT HN patient with seven split-fields was selected for testing purpose. It results in 14 individual subfields each with separated MLC DynaLog file. We simulate this real-time process using the MLC DynaLog files in retrospective. We extract the leaf positions and fractional MU every 50 ms from the MLC DynaLog files and convert them to a beam mask Bi using an in-house software written in C. The accumulated dose distribution delivered up to a time point is then computed by convolving Dij with Bi. Results: It takes on average less than 1 s to pre-calculate Dij for each beam. And it takes less than 0.05s to update the accumulative dose distribution on patient’s geometry with Dij matrix and beam mask Bi from MLC DynaLog files. It meets the real-time requirement for clinic environment.
S733