spatially weighted b-spline deformable image registration for accurate delineation

1
3161 Modeling Electronic Portal Imaging Device for Dose Reconstruction using Monte Carlo Method J. W. Jung 1 , J. O. Kim 2 , I. Yeo 1 1 Cooper University Hospital, Camden, NJ, 2 University of Pittsburgh Cancer Institute, Pittsburgh, PA Purpose/Objective(s): There have been significant developments of verification methods that contribute to adaptive radiation therapy by utilizing real-time patient images and measurement in EPID during treatment. From this measurement dose can be re- constructed in the images. In our past work, we developed a new method of dose reconstruction based on full Monte Carlo calcu- lations and inverse solutions without involving iterations. The current study develops a fast Monte Carlo model of EPID and validates the method in a clinical environment characterized by patient anatomical information and EPID. Materials/Methods: The model is based on density scaling of actual components of the device for fast Monte Carlo dose calcu- lation in a tissue-equivalent medium. While the model offered fast calculation, due to the homogeneous modeling it provided a lim- ited agreement in penumbrae compared with measurements in EPID. A convolution kernel for measurements was determined and used to enhance the agreement. The model with the determined kernel was validated against measurements of open fields in EPID. To perform the reconstruction employing the method developed earlier, dose responses were precalculated in a Rando phantom and EPID; exit doses in EPID for the beams were measured and calculated; the kernel was convolved to the measurement; finally, de- livered dose in the phantom was inversely obtained from the measurement. The reconstructed doses were compared with forwardly calculated doses in the phantom. Five radiation beams optimized for intensity modulated radiation therapy of a pelvic region were used. Although the method is designed for the inverse reconstruction, it additionally provides exit dose in EPID using the precal- culated responses. Results: The model developed in this study reproduced the measured beam profiles (absolute dose) within the maximum 1.3% and 1.6% agreement in cross- and in-plane directions, respectively, for the open fields tried in this study. The calculated dose distribu- tion agreed with the measurements in EPID by showing 93 through 97% pass rates given the criteria of 4.5 mm distance-to-agree- ment (EPID was at 150cm) and 3% dose difference for the five IMRT beams tried in this study. The reconstructed doses showed an agreement with forward calculations by showing 94 through 99% pass rates for the IMRT beams. Conclusions: A fast-calculational model of EPID was developed and validated for our new method of dose reconstruction. *This study was in part supported by Varian Medical Systems, Inc. Author Disclosure: J.W. Jung, None; J.O. Kim, None; I. Yeo, Varian Medical Systems, Inc, B. Research Grant. 3162 Spatially Weighted B-spline Deformable Image Registration for Accurate Delineation S. B. Park 1 , J. I. Monroe 2 , M. Yao 1 , M. Machtay 1 , J. W. Sohn 1 1 Case Western Reserve University, Cleveland, OH, 2 Midsouth Radiation Physics, Paragould, AR Purpose/Objective(s): Deformable image registration based on B-spline can be improved if the automated deformable algorithm takes into account clinically important structures as designated by the user. B-Spline Deformable Image Registration (DIR) with Mutual Information (MI) metric is a popular solution for the DIR with multi-modal capability. B-Spline approximates the defor- mation area using grid points, and expands or compresses elastically between the grid points. Since the grid is uniform, less im- portant structures have the same weight as clinically important structures and thus can produce inaccurate alignments. Spatially weighted B-Spline DIR method (SWBD) addresses this problem. Materials/Methods: By adding a weight function w(x) along the spatial location, SWBD can consider the importance of different image regions. SWBD increases the similarity value when the higher weighted regions exhibit superior alignment. Although the B- Spline deformation uses only the grid resolution as a parameter, we can assign an importance weight to spatial locations within an image. We developed a GUI tool to assign weight values for the various structures. Our GUI tool imports DICOM-RT images and displays the RT structure sets. For proof-of-concept we tested SWBD by evaluating image segmentations using two CT image sets of a head and neck patient. One was the original planning CT for a Radiation Treatment (RT) and the other was a new CT image set. The planning CT has 110 slices of 512Â512 image and the new CT has 120 slices of 512Â512 image. The patient had three clinical treatment volumes (CTV) around the neck. CTV1 and CTV2 were the primary and secondary CTV respectively. CTV3 was the Lymph nodes. The patient gained weight between the two CT scans.. We compared MI and SWBD with the same registration setup. Images were cropped by a rectangular window so the couch and the excessive air region were removed from registration. 19Â19Â19 grids were set inside of the cropped images. A limited memory BFGS-B optimizer was utilized. The Multi-resolution method was applied with three-layer image pyramids. For SWBD, we set higher weights for CTV1 (w = 1.5), CTV2 (w = 1.4), CTV3 (w = 1.3), and Parotids (w = 1.4). The remaining structures defaulted weight of 1.0. Results: CTV1 and CTV2 were well aligned using both MI and SWBD techniques. For MI, the external skin contour near CTV3 was well aligned, however CTV3 was misaligned. The SWBD CTV3 was well aligned since a higher importance weight was as- signed to this segment. The external skin contour was not perfectly aligned due to its lower importance weight as would be ex- pected. SWBD took 670secs to complete. Conclusions: SWBD can provide the better image segmentation for the more important structures with faster convergence. We are pursuing a larger clinical study for validating our image registration method. Author Disclosure: S.B. Park, None; J.I. Monroe, None; M. Yao, None; M. Machtay, None; J.W. Sohn, None. 3163 A Real-time Low-dose Imaging Strategy for Online Patient Repositioning or Respiratory Gating during IMRT or VMAT W. Liu, G. Luxton, T. Mastoridis, S. L. Hancock, L. Xing Stanford University School of Medicine, Stanford, CA Purpose/Objective(s): To develop and implement on newly available Varian Trilogy MX linacs clinic-ready ultra-low imaging dose protocols for online patient repositioning or respiratory gating during IMRT or VMAT. S724 I. J. Radiation Oncology d Biology d Physics Volume 78, Number 3, Supplement, 2010

Upload: jw

Post on 26-Jun-2016

215 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Spatially Weighted B-spline Deformable Image Registration for Accurate Delineation

S724 I. J. Radiation Oncology d Biology d Physics Volume 78, Number 3, Supplement, 2010

3161 Modeling Electronic Portal Imaging Device for Dose Reconstruction using Monte Carlo Method

J. W. Jung1, J. O. Kim2, I. Yeo1

1Cooper University Hospital, Camden, NJ, 2University of Pittsburgh Cancer Institute, Pittsburgh, PA

Purpose/Objective(s): There have been significant developments of verification methods that contribute to adaptive radiationtherapy by utilizing real-time patient images and measurement in EPID during treatment. From this measurement dose can be re-constructed in the images. In our past work, we developed a new method of dose reconstruction based on full Monte Carlo calcu-lations and inverse solutions without involving iterations. The current study develops a fast Monte Carlo model of EPID andvalidates the method in a clinical environment characterized by patient anatomical information and EPID.

Materials/Methods: The model is based on density scaling of actual components of the device for fast Monte Carlo dose calcu-lation in a tissue-equivalent medium. While the model offered fast calculation, due to the homogeneous modeling it provided a lim-ited agreement in penumbrae compared with measurements in EPID. A convolution kernel for measurements was determined andused to enhance the agreement. The model with the determined kernel was validated against measurements of open fields in EPID.To perform the reconstruction employing the method developed earlier, dose responses were precalculated in a Rando phantom andEPID; exit doses in EPID for the beams were measured and calculated; the kernel was convolved to the measurement; finally, de-livered dose in the phantom was inversely obtained from the measurement. The reconstructed doses were compared with forwardlycalculated doses in the phantom. Five radiation beams optimized for intensity modulated radiation therapy of a pelvic region wereused. Although the method is designed for the inverse reconstruction, it additionally provides exit dose in EPID using the precal-culated responses.

Results: The model developed in this study reproduced the measured beam profiles (absolute dose) within the maximum 1.3% and1.6% agreement in cross- and in-plane directions, respectively, for the open fields tried in this study. The calculated dose distribu-tion agreed with the measurements in EPID by showing 93 through 97% pass rates given the criteria of 4.5 mm distance-to-agree-ment (EPID was at 150cm) and 3% dose difference for the five IMRT beams tried in this study. The reconstructed doses showed anagreement with forward calculations by showing 94 through 99% pass rates for the IMRT beams.

Conclusions: A fast-calculational model of EPID was developed and validated for our new method of dose reconstruction. *Thisstudy was in part supported by Varian Medical Systems, Inc.

Author Disclosure: J.W. Jung, None; J.O. Kim, None; I. Yeo, Varian Medical Systems, Inc, B. Research Grant.

3162 Spatially Weighted B-spline Deformable Image Registration for Accurate Delineation

S. B. Park1, J. I. Monroe2, M. Yao1, M. Machtay1, J. W. Sohn1

1Case Western Reserve University, Cleveland, OH, 2Midsouth Radiation Physics, Paragould, AR

Purpose/Objective(s): Deformable image registration based on B-spline can be improved if the automated deformable algorithmtakes into account clinically important structures as designated by the user. B-Spline Deformable Image Registration (DIR) withMutual Information (MI) metric is a popular solution for the DIR with multi-modal capability. B-Spline approximates the defor-mation area using grid points, and expands or compresses elastically between the grid points. Since the grid is uniform, less im-portant structures have the same weight as clinically important structures and thus can produce inaccurate alignments. Spatiallyweighted B-Spline DIR method (SWBD) addresses this problem.

Materials/Methods: By adding a weight function w(x) along the spatial location, SWBD can consider the importance of differentimage regions. SWBD increases the similarity value when the higher weighted regions exhibit superior alignment. Although the B-Spline deformation uses only the grid resolution as a parameter, we can assign an importance weight to spatial locations within animage. We developed a GUI tool to assign weight values for the various structures. Our GUI tool imports DICOM-RT images anddisplays the RT structure sets. For proof-of-concept we tested SWBD by evaluating image segmentations using two CT image setsof a head and neck patient. One was the original planning CT for a Radiation Treatment (RT) and the other was a new CT image set.The planning CT has 110 slices of 512�512 image and the new CT has 120 slices of 512�512 image. The patient had three clinicaltreatment volumes (CTV) around the neck. CTV1 and CTV2 were the primary and secondary CTV respectively. CTV3 was theLymph nodes. The patient gained weight between the two CT scans.. We compared MI and SWBD with the same registrationsetup. Images were cropped by a rectangular window so the couch and the excessive air region were removed from registration.19�19�19 grids were set inside of the cropped images. A limited memory BFGS-B optimizer was utilized. The Multi-resolutionmethod was applied with three-layer image pyramids. For SWBD, we set higher weights for CTV1 (w = 1.5), CTV2 (w = 1.4),CTV3 (w = 1.3), and Parotids (w = 1.4). The remaining structures defaulted weight of 1.0.

Results: CTV1 and CTV2 were well aligned using both MI and SWBD techniques. For MI, the external skin contour near CTV3was well aligned, however CTV3 was misaligned. The SWBD CTV3 was well aligned since a higher importance weight was as-signed to this segment. The external skin contour was not perfectly aligned due to its lower importance weight as would be ex-pected. SWBD took 670secs to complete.

Conclusions: SWBD can provide the better image segmentation for the more important structures with faster convergence. We arepursuing a larger clinical study for validating our image registration method.

Author Disclosure: S.B. Park, None; J.I. Monroe, None; M. Yao, None; M. Machtay, None; J.W. Sohn, None.

3163 A Real-time Low-dose Imaging Strategy for Online Patient Repositioning or Respiratory Gating during

IMRT or VMAT

W. Liu, G. Luxton, T. Mastoridis, S. L. Hancock, L. Xing

Stanford University School of Medicine, Stanford, CA

Purpose/Objective(s): To develop and implement on newly available Varian Trilogy MX linacs clinic-ready ultra-low imagingdose protocols for online patient repositioning or respiratory gating during IMRT or VMAT.