- Email: [email protected]
Evaluation of Fiducial Marker Migration and Respiratory-induced Motion for Image Guided Radiotherapy in Accelerated Partial Breast Irradiation
I. J. Radiation Oncology d Biology d Physics
S626
Volume 75, Number 3, Supplement, 2009
object is over 89%. Average distance between the registered and the source surface is about 2.1mm with a maximum error under 5 mm. In particular, critical structures and volumes, such as the CTV, spinal cord, and radiosensitive parotid glands, are precisely tracked from CT to CBCT, which is of clinical importance for IGRT cumulative dose calculation and online plan adjustment. Conclusions: The proposed registration method significantly enhances registration speed without loss of accuracy. Furthermore, the meshless model integrated framework has a robust registration performance and has the potential to be utilized for IGRT applications. Author Disclosure: T. Chen, None; S. Kim, None; J. Zhou, None; G. Rajagopal, None; S. Goyal, None; S. Jabbour, None; B. Haffty, None; N. Yue, None.
2991
Towards On-line Treatment Verification using EPID Cine Images for Hypofractionated Lung IMRT
1
X. Tang , T. Lin2, A. Sandhu3, S. Jiang3, J. Lian1, S. Chang1, E. Chaney1 University of North Carolina, Chapel Hill, NC, 2Beijing University, Beijing, China, 3University of California San Diego, La Jolla, CA
1
Purpose/Objective(s): To develop a computational algorithm based on an artificial neural network (ANN) that allows treatment verification using an EPID in cine mode for hypofractionated lung IMRT. Materials/Methods: We developed a novel ANN based technique that can extract tumor information from cine EPID images of each segment in an IMRT field and verify that we are treating the planned treatment area. We simulated training images for ANN using DRRs, by shifting DRRs relative to the beam aperture of the segment. We also carefully related each segment to the entire IMRT field in terms of tumor location. With a pre-defined threshold p%, we associated category 1 to the training image if more than p% of the tumor projection in the beam eye view was in the corresponding entire treatment field of the training image, and category -1 otherwise. The trained network could therefore analyze the cine EPID images of IMRT segments obtained during the treatment and classify them into the corresponding category 1 or -1. Results: We are the first who can analyze cine EPID images of hypofractionated lung IMRT without implanted fiducial markers. The challenging problem solved in this work is how to extract tumor information from cine EPID images of each segment in an IMRT field. Three patients, each treated with 4 or 5 fractions, were included in our preliminary study. A window size of 10 by 40 pixels (9.77 by 39.06 mm) was used on each segment to generate training images. The frequency of cine EPID images was 0.625 Hz. A radiation oncologist read the cine EPID images and classified them into category 1 or -1; this served as our ground truth. The ANN was applied to the training images to build the neural network. We set p = 95 for this study if more than 95% of the tumor was inside the treatment field that the current segment corresponds to, the training image was said in category 1. For each IMRT segment, one neural network needed to be built. Averaging over all patients and all segments, the trained network successfully classified 97.3% of the cine EPID images. Conclusions: We proposed a novel ANN based IMRT treatment verification technique using cine EPID images. This work provides an important clinical tool to assure patient safety without additional cost. Margins can be reduced accordingly, and when the tumor moves out of the irradiation segment, the treatment beam can be interrupted so that unnecessary radiation won’t be delivered to normal tissues. Author Disclosure: X. Tang, None; T. Lin, None; A. Sandhu, None; S. Jiang, None; J. Lian, None; S. Chang, None; E. Chaney, None.
2992
Evaluation of Fiducial Marker Migration and Respiratory-induced Motion for Image Guided Radiotherapy in Accelerated Partial Breast Irradiation
C. K. Park, G. Zhang, K. M. Forster, E. E. R. Harris Moffitt Cancer Center, Tampa, FL Purpose/Objective(s): Image guided radiation therapy (IGRT) may be beneficial to improve set up accuracy and reduce treatment margins, therefore improve efficacy and decrease long term toxicity of accelerated partial breast irradiation (APBI). This study was conducted to quantify intrafraction respiratory motion, variations in respiratory patterns, and fiducial marker migration. These data are used to investigate whether intraparenchymal fiducials can be used for IGRT in breast cancer treatment. Materials/Methods: On a prospective IRB-approved protocol, fiducials were placed in fifteen patients who were treated with 3D conformal APBI. Three or four intraparenchymal gold fiducial markers were placed in each patient at the periphery of the surgical bed intraoperatively. Free breathing 4D CT image sets were obtained pre-treatment and post-treatment. Each fiducial marker was contoured at end-inspiration and end-expiration on both pre- and post-treatment 4D CT image sets. Intrafraction motion due to respiration was assessed by comparing the position of the fiducials’ center of mass between end-inspiration and end-expiration. A variation in respiratory pattern was assessed by comparing motion of the fiducials’ center of mass between pre- and post-treatment. Fiducial migration was determined by comparing the relative positions of each marker to one another from the pre-treatment and post-treatment 4D CT image sets. Results: The pre- and post-treatment CT image sets were acquired on average 19.64±6.52 days apart with a range of 10-32 days. The average intrafraction motion as determined from all 30 4D CT image sets was 1.5 mm (Std Dev = 1.0 mm) with the range of 0 to 4.2 mm for all individual fiducials. The variation in respiration as measured by the average change in the fiducials’ center of mass was 1.1 mm (Std dev 1 mm) with a range of 0 mm to 2.5 mm. Fiducial migration as determined by the average variation in relative seed position from the first image set to the second set was 1.6 mm (Std Dev 1.4 mm) with a range of 0 to 5.2 mm, and only two relative fiducial pairs changed by more than 3 mm. Conclusions: The preliminary results of using gold fiducials for breast IGRT are promising. Fiducial position was very stable during treatment. There was very little intrafraction motion associated with respiration, or changes to due variations in respiratory pattern pre- and post-treatment. Using IGRT, there is the potential to reduce PTV margins in APBI. Better tumor bed localization and reduced margin size will decrease the volume of normal tissue treated, which may translate into improved local control by ensuring
Proceedings of the 51st Annual ASTRO Meeting accurate coverage of the target volumes throughout the treatment course and improvement in cosmesis and other long term toxicities. Author Disclosure: C.K. Park, None; G. Zhang, None; K.M. Forster, None; E.E.R. Harris, None.
2993
PET Tumor Segmentation: Multi-observer Validation of a Gradient-Based Method using a NSCLC PET Phantom
A. D. Nelson1, M. Werner-Wasik2, W. Choi3, Y. Arai4, P. F. Faulhaber5,6, N. Ohri2, J. W. Piper1,7, K. D. Brockway1, A. S. Nelson1 MIMvista Corp., Cleveland, OH, 2Thomas Jefferson University Hospital, Philadelphia, PA, 3Beth Israel Medical Center, New York, NY, 4University of Pittsburgh Medical Center Health Systems, Pittsburgh, PA, 5University Hospitals Case Medical Center, Cleveland, OH, 6Case Western Reserve University, Cleveland, OH, 7Computer Science Dept., Wake Forest University, Winston-Salem, NC
1
Purpose/Objective(s): Consistent and accurate methods for PET tumor segmentation are needed in radiation therapy with the growing role of PET in target definition, prognosis, and therapy response assessment. Previously we demonstrated the greater accuracy of a PET gradient-based segmentation method, PET Edge (GRAD) as compared to constant threshold method (THRESH) using spherical phantoms. Our goal in the present study is to compare the accuracy and consistency of GRAD vs. THRESH vs. manual contouring (MC) across multiple observers utilizing published Monte Carlo simulated PET scans of the thorax with "tumors" simulating common anatomic locations of lung cancer lesions. Materials/Methods: We obtained 25 realistic digital PET phantoms of the thorax containing 31 simulated tumors of varying size, shape and location. Tumors ranged in size from 7 ml to 264 ml with a median size of 73 ml. Five observers segmented each tumor with GRAD, THRESH, and MC. THRESH was performed using thresholds of 25-50% of maximum counts at 5% increments. Tumor volumes obtained by each method were compared to the known phantom tumor volumes. Accuracy was measured by the mean absolute % difference for the volume (Vdiff%) for each group using all methods. Results: Across all observers, GRAD was the most accurate segmentation technique with Vdiff% of 11.3% (12.2% SD). The next most accurate, 25% THRESH, with Vdiff% of 15.1% (16.7% SD), and MC, with Vdiff 18.9% (17.3% SD). Both 25% THRESH and MC were significantly less accurate than GRAD (p-value\ 0.01). GRAD also had the smallest amount of systematic bias with volume differences of -0.45% (16.7% SD) compared with -7.0% (21.4% SD) and -15.7% (20.3% SD) for 25% THRESH and MC, respectively (p-value \0.01). Finally, GRAD segmentations demonstrated greater volumetric consistency between observers than either MC or 25% THRESH in 25/31 and 21/31 lesions, respectively. This difference was significant (p-value\0.05) for 11/25 and 13/21 lesions, respectively. Conclusions: GRAD resulted in more accurate tumor volume delineation across all observers than either MC or THRESH and demonstrated greater consistency between observers. GRAD has the potential to play an important role in Radiation Therapy through more accurate and consistent PET tumor segmentation. Author Disclosure: A.D. Nelson, MIMvista Corp, A. Employment; MIMvista Corp, E. Ownership Interest; M. Werner-Wasik, None; W. Choi, None; Y. Arai, None; P.F. Faulhaber, MIMvista Corp, F. Consultant/Advisory Board; N. Ohri, None; J.W. Piper, MIMvista Corp, A. Employment; MIMvista Corp, E. Ownership Interest; K.D. Brockway, MIMvista Corp, A. Employment; MIMvista Corp, E. Ownership Interest; A.S. Nelson, MIMvista Corp, A. Employment; MIMvista Corp, E. Ownership Interest.
2994
A Novel Optimization Based Leaf Sequencing Algorithm with Explicit Underdose and Overdose Penalties in 4D Radiotherapy
D. Ruan1, A. Sawant1, B. C. Cho1, P. R. Poulsen1,2, P. J. Keall1 1
Stanford University, Stanford, CA, 2Aarhus University Hospital, Aarhus, Denmark
Purpose/Objective(s): Intrafractional adaptivity in 4D radiotherapy requires the MLC to deliver beam patterns that are modified by real-time target motion. In general, the modified beam pattern differs from the original plan, and the MLC leaf-sequence needs to be recomputed. The purpose of this study was to develop and investigate an optimization based leaf sequencing method that penalizes underdose and overdose explicitly, and allows arbitrary target deformation and planning geometry. Materials/Methods: We formulated the leaf sequencing problem in a rigorous constrained optimization framework. Taking the user specified underdose/overdose cost as input, the optimization engine seeks the leaf sequence parameters that minimize the cumulative dose error, i.e., the integrated underdose/overdose cost over the region of interest. The optimal solution was derived based on the Karush-Kuhn-Tucker condition, and the separability of the cost across various leaf tracks was utilized to improve the efficiency of our algorithm. Our method allows for arbitrary target beam topology, making it applicable to rotational and nonrigid deformation of the target. As it is common clinically to prescribe the plan in terms of leaf placement, and consider only translational motion, we have further devised an operation mode that utilizes the track-wise convex geometry, leading to significant computation acceleration. The proposed leaf sequencing algorithm was integrated into an existing 4D tracking platform at Stanford University. To evaluate the efficacy of the proposed leaf fitting algorithm, we recorded the discrepancy between the ideal deformed beam aperture and the fitted leaf pattern, and computed the accuracy with respect to the delivery to a static target, using a 3%-3mm gamma index criterion. Step and shoot (S-IMRT) and sliding window (D-IMRT) plans were tested with patient-derived 3D motion trajectories. Dosimetric measures were obtained with a 4D phantom for further verification. Results: Numerical simulation with various lung tumor motion traces and planning geometry demonstrated that our method of leaf sequencing with tracking reduces the gamma test failure rate by 2- to 3-fold, compared to static delivery. A typical dosimetric test showed a 48% gamma test failure rate with our method, compared to 75% for static delivery. Conclusions: We have developed a new optimization based leaf sequencing algorithm that explicitly penalizes underdose/overdose error, and validated its performance with numerical simulation and dosimetric measurement. The method’s flexibility in casespecific underdose/overdose tradeoff assignment, general topology variations and arbitrary target definition makes it a valuable tool, and is expected to have far-reaching impact in modern 4D radiotherapy. Author Disclosure: D. Ruan, None; A. Sawant, None; B.C. Cho, None; P.R. Poulsen, Varian Medical System, B. Research Grant; P.J. Keall, NIH/NCI R01-93626, B. Research Grant.
S627
S626
Volume 75, Number 3, Supplement, 2009
object is over 89%. Average distance between the registered and the source surface is about 2.1mm with a maximum error under 5 mm. In particular, critical structures and volumes, such as the CTV, spinal cord, and radiosensitive parotid glands, are precisely tracked from CT to CBCT, which is of clinical importance for IGRT cumulative dose calculation and online plan adjustment. Conclusions: The proposed registration method significantly enhances registration speed without loss of accuracy. Furthermore, the meshless model integrated framework has a robust registration performance and has the potential to be utilized for IGRT applications. Author Disclosure: T. Chen, None; S. Kim, None; J. Zhou, None; G. Rajagopal, None; S. Goyal, None; S. Jabbour, None; B. Haffty, None; N. Yue, None.
2991
Towards On-line Treatment Verification using EPID Cine Images for Hypofractionated Lung IMRT
1
X. Tang , T. Lin2, A. Sandhu3, S. Jiang3, J. Lian1, S. Chang1, E. Chaney1 University of North Carolina, Chapel Hill, NC, 2Beijing University, Beijing, China, 3University of California San Diego, La Jolla, CA
1
Purpose/Objective(s): To develop a computational algorithm based on an artificial neural network (ANN) that allows treatment verification using an EPID in cine mode for hypofractionated lung IMRT. Materials/Methods: We developed a novel ANN based technique that can extract tumor information from cine EPID images of each segment in an IMRT field and verify that we are treating the planned treatment area. We simulated training images for ANN using DRRs, by shifting DRRs relative to the beam aperture of the segment. We also carefully related each segment to the entire IMRT field in terms of tumor location. With a pre-defined threshold p%, we associated category 1 to the training image if more than p% of the tumor projection in the beam eye view was in the corresponding entire treatment field of the training image, and category -1 otherwise. The trained network could therefore analyze the cine EPID images of IMRT segments obtained during the treatment and classify them into the corresponding category 1 or -1. Results: We are the first who can analyze cine EPID images of hypofractionated lung IMRT without implanted fiducial markers. The challenging problem solved in this work is how to extract tumor information from cine EPID images of each segment in an IMRT field. Three patients, each treated with 4 or 5 fractions, were included in our preliminary study. A window size of 10 by 40 pixels (9.77 by 39.06 mm) was used on each segment to generate training images. The frequency of cine EPID images was 0.625 Hz. A radiation oncologist read the cine EPID images and classified them into category 1 or -1; this served as our ground truth. The ANN was applied to the training images to build the neural network. We set p = 95 for this study if more than 95% of the tumor was inside the treatment field that the current segment corresponds to, the training image was said in category 1. For each IMRT segment, one neural network needed to be built. Averaging over all patients and all segments, the trained network successfully classified 97.3% of the cine EPID images. Conclusions: We proposed a novel ANN based IMRT treatment verification technique using cine EPID images. This work provides an important clinical tool to assure patient safety without additional cost. Margins can be reduced accordingly, and when the tumor moves out of the irradiation segment, the treatment beam can be interrupted so that unnecessary radiation won’t be delivered to normal tissues. Author Disclosure: X. Tang, None; T. Lin, None; A. Sandhu, None; S. Jiang, None; J. Lian, None; S. Chang, None; E. Chaney, None.
2992
Evaluation of Fiducial Marker Migration and Respiratory-induced Motion for Image Guided Radiotherapy in Accelerated Partial Breast Irradiation
C. K. Park, G. Zhang, K. M. Forster, E. E. R. Harris Moffitt Cancer Center, Tampa, FL Purpose/Objective(s): Image guided radiation therapy (IGRT) may be beneficial to improve set up accuracy and reduce treatment margins, therefore improve efficacy and decrease long term toxicity of accelerated partial breast irradiation (APBI). This study was conducted to quantify intrafraction respiratory motion, variations in respiratory patterns, and fiducial marker migration. These data are used to investigate whether intraparenchymal fiducials can be used for IGRT in breast cancer treatment. Materials/Methods: On a prospective IRB-approved protocol, fiducials were placed in fifteen patients who were treated with 3D conformal APBI. Three or four intraparenchymal gold fiducial markers were placed in each patient at the periphery of the surgical bed intraoperatively. Free breathing 4D CT image sets were obtained pre-treatment and post-treatment. Each fiducial marker was contoured at end-inspiration and end-expiration on both pre- and post-treatment 4D CT image sets. Intrafraction motion due to respiration was assessed by comparing the position of the fiducials’ center of mass between end-inspiration and end-expiration. A variation in respiratory pattern was assessed by comparing motion of the fiducials’ center of mass between pre- and post-treatment. Fiducial migration was determined by comparing the relative positions of each marker to one another from the pre-treatment and post-treatment 4D CT image sets. Results: The pre- and post-treatment CT image sets were acquired on average 19.64±6.52 days apart with a range of 10-32 days. The average intrafraction motion as determined from all 30 4D CT image sets was 1.5 mm (Std Dev = 1.0 mm) with the range of 0 to 4.2 mm for all individual fiducials. The variation in respiration as measured by the average change in the fiducials’ center of mass was 1.1 mm (Std dev 1 mm) with a range of 0 mm to 2.5 mm. Fiducial migration as determined by the average variation in relative seed position from the first image set to the second set was 1.6 mm (Std Dev 1.4 mm) with a range of 0 to 5.2 mm, and only two relative fiducial pairs changed by more than 3 mm. Conclusions: The preliminary results of using gold fiducials for breast IGRT are promising. Fiducial position was very stable during treatment. There was very little intrafraction motion associated with respiration, or changes to due variations in respiratory pattern pre- and post-treatment. Using IGRT, there is the potential to reduce PTV margins in APBI. Better tumor bed localization and reduced margin size will decrease the volume of normal tissue treated, which may translate into improved local control by ensuring
Proceedings of the 51st Annual ASTRO Meeting accurate coverage of the target volumes throughout the treatment course and improvement in cosmesis and other long term toxicities. Author Disclosure: C.K. Park, None; G. Zhang, None; K.M. Forster, None; E.E.R. Harris, None.
2993
PET Tumor Segmentation: Multi-observer Validation of a Gradient-Based Method using a NSCLC PET Phantom
A. D. Nelson1, M. Werner-Wasik2, W. Choi3, Y. Arai4, P. F. Faulhaber5,6, N. Ohri2, J. W. Piper1,7, K. D. Brockway1, A. S. Nelson1 MIMvista Corp., Cleveland, OH, 2Thomas Jefferson University Hospital, Philadelphia, PA, 3Beth Israel Medical Center, New York, NY, 4University of Pittsburgh Medical Center Health Systems, Pittsburgh, PA, 5University Hospitals Case Medical Center, Cleveland, OH, 6Case Western Reserve University, Cleveland, OH, 7Computer Science Dept., Wake Forest University, Winston-Salem, NC
1
Purpose/Objective(s): Consistent and accurate methods for PET tumor segmentation are needed in radiation therapy with the growing role of PET in target definition, prognosis, and therapy response assessment. Previously we demonstrated the greater accuracy of a PET gradient-based segmentation method, PET Edge (GRAD) as compared to constant threshold method (THRESH) using spherical phantoms. Our goal in the present study is to compare the accuracy and consistency of GRAD vs. THRESH vs. manual contouring (MC) across multiple observers utilizing published Monte Carlo simulated PET scans of the thorax with "tumors" simulating common anatomic locations of lung cancer lesions. Materials/Methods: We obtained 25 realistic digital PET phantoms of the thorax containing 31 simulated tumors of varying size, shape and location. Tumors ranged in size from 7 ml to 264 ml with a median size of 73 ml. Five observers segmented each tumor with GRAD, THRESH, and MC. THRESH was performed using thresholds of 25-50% of maximum counts at 5% increments. Tumor volumes obtained by each method were compared to the known phantom tumor volumes. Accuracy was measured by the mean absolute % difference for the volume (Vdiff%) for each group using all methods. Results: Across all observers, GRAD was the most accurate segmentation technique with Vdiff% of 11.3% (12.2% SD). The next most accurate, 25% THRESH, with Vdiff% of 15.1% (16.7% SD), and MC, with Vdiff 18.9% (17.3% SD). Both 25% THRESH and MC were significantly less accurate than GRAD (p-value\ 0.01). GRAD also had the smallest amount of systematic bias with volume differences of -0.45% (16.7% SD) compared with -7.0% (21.4% SD) and -15.7% (20.3% SD) for 25% THRESH and MC, respectively (p-value \0.01). Finally, GRAD segmentations demonstrated greater volumetric consistency between observers than either MC or 25% THRESH in 25/31 and 21/31 lesions, respectively. This difference was significant (p-value\0.05) for 11/25 and 13/21 lesions, respectively. Conclusions: GRAD resulted in more accurate tumor volume delineation across all observers than either MC or THRESH and demonstrated greater consistency between observers. GRAD has the potential to play an important role in Radiation Therapy through more accurate and consistent PET tumor segmentation. Author Disclosure: A.D. Nelson, MIMvista Corp, A. Employment; MIMvista Corp, E. Ownership Interest; M. Werner-Wasik, None; W. Choi, None; Y. Arai, None; P.F. Faulhaber, MIMvista Corp, F. Consultant/Advisory Board; N. Ohri, None; J.W. Piper, MIMvista Corp, A. Employment; MIMvista Corp, E. Ownership Interest; K.D. Brockway, MIMvista Corp, A. Employment; MIMvista Corp, E. Ownership Interest; A.S. Nelson, MIMvista Corp, A. Employment; MIMvista Corp, E. Ownership Interest.
2994
A Novel Optimization Based Leaf Sequencing Algorithm with Explicit Underdose and Overdose Penalties in 4D Radiotherapy
D. Ruan1, A. Sawant1, B. C. Cho1, P. R. Poulsen1,2, P. J. Keall1 1
Stanford University, Stanford, CA, 2Aarhus University Hospital, Aarhus, Denmark
Purpose/Objective(s): Intrafractional adaptivity in 4D radiotherapy requires the MLC to deliver beam patterns that are modified by real-time target motion. In general, the modified beam pattern differs from the original plan, and the MLC leaf-sequence needs to be recomputed. The purpose of this study was to develop and investigate an optimization based leaf sequencing method that penalizes underdose and overdose explicitly, and allows arbitrary target deformation and planning geometry. Materials/Methods: We formulated the leaf sequencing problem in a rigorous constrained optimization framework. Taking the user specified underdose/overdose cost as input, the optimization engine seeks the leaf sequence parameters that minimize the cumulative dose error, i.e., the integrated underdose/overdose cost over the region of interest. The optimal solution was derived based on the Karush-Kuhn-Tucker condition, and the separability of the cost across various leaf tracks was utilized to improve the efficiency of our algorithm. Our method allows for arbitrary target beam topology, making it applicable to rotational and nonrigid deformation of the target. As it is common clinically to prescribe the plan in terms of leaf placement, and consider only translational motion, we have further devised an operation mode that utilizes the track-wise convex geometry, leading to significant computation acceleration. The proposed leaf sequencing algorithm was integrated into an existing 4D tracking platform at Stanford University. To evaluate the efficacy of the proposed leaf fitting algorithm, we recorded the discrepancy between the ideal deformed beam aperture and the fitted leaf pattern, and computed the accuracy with respect to the delivery to a static target, using a 3%-3mm gamma index criterion. Step and shoot (S-IMRT) and sliding window (D-IMRT) plans were tested with patient-derived 3D motion trajectories. Dosimetric measures were obtained with a 4D phantom for further verification. Results: Numerical simulation with various lung tumor motion traces and planning geometry demonstrated that our method of leaf sequencing with tracking reduces the gamma test failure rate by 2- to 3-fold, compared to static delivery. A typical dosimetric test showed a 48% gamma test failure rate with our method, compared to 75% for static delivery. Conclusions: We have developed a new optimization based leaf sequencing algorithm that explicitly penalizes underdose/overdose error, and validated its performance with numerical simulation and dosimetric measurement. The method’s flexibility in casespecific underdose/overdose tradeoff assignment, general topology variations and arbitrary target definition makes it a valuable tool, and is expected to have far-reaching impact in modern 4D radiotherapy. Author Disclosure: D. Ruan, None; A. Sawant, None; B.C. Cho, None; P.R. Poulsen, Varian Medical System, B. Research Grant; P.J. Keall, NIH/NCI R01-93626, B. Research Grant.
S627