- Email: [email protected]
Investigation of Cone-beam CT for Breast Cancer Treatment Planning
Proceedings of the 47th Annual ASTRO Meeting
of the control regions in the reference image through user interaction, in the proposed method, each of the control regions is mapped to the corresponding parts of the reference image by using an automated image registration algorithm. The conventional automated image registration algorithm is then used to complete the image registration process with the auto-determined control points. A normalized correlation function (intra-modality image registration) or a mutual information metric (inter-modality image registration) was used as the metric in both the selection of the control volumes and the final image registration. The deformable registration was modeled by free form deformations based on spline interpolation. The limited memory BroydenFletcher-Goldfarb-Shanno algorithm (L-BFGS) was used to optimize the metric function. The performance of the registrationin-registration approach was examined by registering CT and FLT-PET images of a rectal cancer patient, CT and MRI images of brain tumor patient, and two sets of images of a lung case acquired at two different respiratory phases. For each case, the convergence behavior of the algorithm was studied by registering the two input images with 100 randomly initiated relative positions. The performance of the registration was evaluated by comparing with the results obtained by using direct registration without the use of auto-mapped control volumes. Results: An image registration algorithm with auto-mapped control regions has been proposed for intra- or inter-modality image registration. For each case, the convergence of the algorithm was confirmed by starting the registration calculation from 100 different initial conditions. The brain image registration suggested that the technique can match a CT and MRI images with an accuracy of ⬃1 mm in the case of rigid body registration in less than a minute on a standard PC computer. Application of the technique to the clinical FLT-PET and CT registration showed a similar level of success. We found that the technique is especially valuable for improving the accuracy and calculation speed of deformable image registration. For the registration of CT images acquired at inhale/exhale phases of the lung patient, the BSpline calculation was speeded up by an order of magnitude with notable improvement in the registration quality. Conclusions: The proposed method of determining the control points greatly reduces the complexity involved with the determination of homologous control points and allows us to minimize the subjectivity and uncertainty often occurring in the use of user-defined control points. Patient studies have indicated that the two-step registration technique is reliable and provides a valuable tool to facilitate both rigid and non-rigid image registration problem.
2537
Investigation of Cone-beam CT for Breast Cancer Treatment Planning
G. Kim, K. Horst, P. Maxim, G. Luxton, L. Xing, A.L. Boyer, T. Pawlicki Radiation Oncology, Stanford Univ School of Medicine, Stanford, CA Purpose/Objective: To investigate the applicability of cone-beam computed tomography (CBCT) for breast radiotherapy treatment planning, dose calculation and conventional setup verification. Materials/Methods: An amorphous silicon flat-panel X-ray detector and kV X-ray source mounted on a convention linac together with special software enables CBCT to be acquired on the treatment machine. Our CBCT data was obtained on a Varian 23EX Trilogy Accelerator. The conventional CT data was obtained on a GE Lightspeed ST CT scanner. A left breast cancer patient was planning based-on the conventional CT data set in the Varian Eclipse planning system. Typical opposed oblique fields were planned to deliver 5040cGy in 28 fractions. DRRs for beam placement verification were also created from the conventional CT data set. A CBCT data set was obtained for the same patient in treatment position on the Trilogy Accelerator. The maximum length in the superior-inferior direction for a CBCT is limited to 17cm. The CBCT data was transferred to the Eclipse planning system. The original treatment plan (based on the conventional CT data) was recreated on the CBCT data. Prior to treatment planning on the CBCT data set, a Hounsfield unit to relative density curve was created in Eclipse for the purpose of heterogeneity corrections. The curve was generated using the Catphan cylindrical QA phantom. Additionally, a bulk lung and tissue density correction was also applied to the CBCT data set for planning. Dose distributions were compared on the CBCT images with heterogeneity corrections and also for bulk tissue corrections to conventional CT images. DRRs for conventional 2D setup verification were compared qualitatively between the CBCT and the conventional CT data sets. Results: The plan based on the conventional CT has a maximum PTV dose of 5236cGy, and mean PTV dose of 4747cGy. The maximum lung dose (1cc) is 4457cGy. The maximum dose in the patient is 5236cGy. The plan based on the CBCT has a maximum PTV dose of 5323cGy, mean PTV dose of 4825cGy, maximum lung dose (1cc) is 4286cGy, and maximum dose in the patient is 5322cGy. On the other hand, if a bulk density correction is done to the structures in the CBCT data set then the dose distribution is closer to that planned on the conventional CT data set. For the CBCT plan with bulk density correction for the lung and breast tissue, the maximum PTV dose is 5286cGy, mean PTV dose of 4802cGy, maximum lung dose (1cc) is 4317cGy, and maximum dose in the patient is 5288cGy. The DRR quality from conventional CT and CBCT are comparable for the purpose of conventional port film setup verification (CBCT left in figure). Conclusions: Dose calculation with heterogeneity corrections based on CBCT Hounsfield units is not recommended for breast cancer treatment planning on the Varian system. Treatment planning and dose calculation based on CBCT data sets should be done with bulk density corrections for the segmented soft tissues. Virtual simulation and soft tissue segmentation are still achievable. The CBCT DRRs are comparable to conventional CT DRRs for routine setup verification. The 17cm length of the CBCT reconstruction volume will be limiting for the treatment planner if non-coplanar beams are required.
S553
of the control regions in the reference image through user interaction, in the proposed method, each of the control regions is mapped to the corresponding parts of the reference image by using an automated image registration algorithm. The conventional automated image registration algorithm is then used to complete the image registration process with the auto-determined control points. A normalized correlation function (intra-modality image registration) or a mutual information metric (inter-modality image registration) was used as the metric in both the selection of the control volumes and the final image registration. The deformable registration was modeled by free form deformations based on spline interpolation. The limited memory BroydenFletcher-Goldfarb-Shanno algorithm (L-BFGS) was used to optimize the metric function. The performance of the registrationin-registration approach was examined by registering CT and FLT-PET images of a rectal cancer patient, CT and MRI images of brain tumor patient, and two sets of images of a lung case acquired at two different respiratory phases. For each case, the convergence behavior of the algorithm was studied by registering the two input images with 100 randomly initiated relative positions. The performance of the registration was evaluated by comparing with the results obtained by using direct registration without the use of auto-mapped control volumes. Results: An image registration algorithm with auto-mapped control regions has been proposed for intra- or inter-modality image registration. For each case, the convergence of the algorithm was confirmed by starting the registration calculation from 100 different initial conditions. The brain image registration suggested that the technique can match a CT and MRI images with an accuracy of ⬃1 mm in the case of rigid body registration in less than a minute on a standard PC computer. Application of the technique to the clinical FLT-PET and CT registration showed a similar level of success. We found that the technique is especially valuable for improving the accuracy and calculation speed of deformable image registration. For the registration of CT images acquired at inhale/exhale phases of the lung patient, the BSpline calculation was speeded up by an order of magnitude with notable improvement in the registration quality. Conclusions: The proposed method of determining the control points greatly reduces the complexity involved with the determination of homologous control points and allows us to minimize the subjectivity and uncertainty often occurring in the use of user-defined control points. Patient studies have indicated that the two-step registration technique is reliable and provides a valuable tool to facilitate both rigid and non-rigid image registration problem.
2537
Investigation of Cone-beam CT for Breast Cancer Treatment Planning
G. Kim, K. Horst, P. Maxim, G. Luxton, L. Xing, A.L. Boyer, T. Pawlicki Radiation Oncology, Stanford Univ School of Medicine, Stanford, CA Purpose/Objective: To investigate the applicability of cone-beam computed tomography (CBCT) for breast radiotherapy treatment planning, dose calculation and conventional setup verification. Materials/Methods: An amorphous silicon flat-panel X-ray detector and kV X-ray source mounted on a convention linac together with special software enables CBCT to be acquired on the treatment machine. Our CBCT data was obtained on a Varian 23EX Trilogy Accelerator. The conventional CT data was obtained on a GE Lightspeed ST CT scanner. A left breast cancer patient was planning based-on the conventional CT data set in the Varian Eclipse planning system. Typical opposed oblique fields were planned to deliver 5040cGy in 28 fractions. DRRs for beam placement verification were also created from the conventional CT data set. A CBCT data set was obtained for the same patient in treatment position on the Trilogy Accelerator. The maximum length in the superior-inferior direction for a CBCT is limited to 17cm. The CBCT data was transferred to the Eclipse planning system. The original treatment plan (based on the conventional CT data) was recreated on the CBCT data. Prior to treatment planning on the CBCT data set, a Hounsfield unit to relative density curve was created in Eclipse for the purpose of heterogeneity corrections. The curve was generated using the Catphan cylindrical QA phantom. Additionally, a bulk lung and tissue density correction was also applied to the CBCT data set for planning. Dose distributions were compared on the CBCT images with heterogeneity corrections and also for bulk tissue corrections to conventional CT images. DRRs for conventional 2D setup verification were compared qualitatively between the CBCT and the conventional CT data sets. Results: The plan based on the conventional CT has a maximum PTV dose of 5236cGy, and mean PTV dose of 4747cGy. The maximum lung dose (1cc) is 4457cGy. The maximum dose in the patient is 5236cGy. The plan based on the CBCT has a maximum PTV dose of 5323cGy, mean PTV dose of 4825cGy, maximum lung dose (1cc) is 4286cGy, and maximum dose in the patient is 5322cGy. On the other hand, if a bulk density correction is done to the structures in the CBCT data set then the dose distribution is closer to that planned on the conventional CT data set. For the CBCT plan with bulk density correction for the lung and breast tissue, the maximum PTV dose is 5286cGy, mean PTV dose of 4802cGy, maximum lung dose (1cc) is 4317cGy, and maximum dose in the patient is 5288cGy. The DRR quality from conventional CT and CBCT are comparable for the purpose of conventional port film setup verification (CBCT left in figure). Conclusions: Dose calculation with heterogeneity corrections based on CBCT Hounsfield units is not recommended for breast cancer treatment planning on the Varian system. Treatment planning and dose calculation based on CBCT data sets should be done with bulk density corrections for the segmented soft tissues. Virtual simulation and soft tissue segmentation are still achievable. The CBCT DRRs are comparable to conventional CT DRRs for routine setup verification. The 17cm length of the CBCT reconstruction volume will be limiting for the treatment planner if non-coplanar beams are required.
S553