D Tewatia

Hippocampal-Sparing Whole-Brain Radiotherapy: A “How-To” Technique Using Helical Tomotherapy and Linear Accelerator–Based Intensity-Modulated Radiotherapy

PURPOSE Sparing the hippocampus during cranial irradiation poses important technical challenges with respect to contouring and treatment planning. Herein we report our preliminary experience with whole-brain radiotherapy using hippocampal sparing for patients with brain metastases. METHODS AND MATERIALS Five anonymous patients previously treated with whole-brain radiotherapy with hippocampal sparing were reviewed. The hippocampus was contoured, and hippocampal avoidance regions were created using a 5-mm volumetric expansion around the hippocampus. Helical tomotherapy and linear accelerator (LINAC)-based intensity-modulated radiotherapy (IMRT) treatment plans were generated for a prescription dose of 30 Gy in 10 fractions. RESULTS On average, the hippocampal avoidance volume was 3.3 cm(3), occupying 2.1% of the whole-brain planned target volume. Helical tomotherapy spared the hippocampus, with a median dose of 5.5 Gy and maximum dose of 12.8 Gy. LINAC-based IMRT spared the hippocampus, with a median dose of 7.8 Gy and maximum dose of 15.3 Gy. On a per-fraction basis, mean dose to the hippocampus (normalized to 2-Gy fractions) was reduced by 87% to 0.49 Gy(2) using helical tomotherapy and by 81% to 0.73 Gy(2) using LINAC-based IMRT. Target coverage and homogeneity was acceptable with both IMRT modalities, with differences largely attributed to more rapid dose fall-off with helical tomotherapy. CONCLUSION Modern IMRT techniques allow for sparing of the hippocampus with acceptable target coverage and homogeneity. Based on compelling preclinical evidence, a Phase II cooperative group trial has been developed to test the postulated neurocognitive benefit.

Clinical implementation of target tracking by breathing synchronized delivery

Target-tracking techniques can be categorized based on the mechanism of the feedback loop. In real time tracking, breathing-delivery phase correlation is provided to the treatment delivery hardware. Clinical implementation of target tracking in real time requires major hardware modifications. In breathing synchronized delivery (BSD), the patient is guided to breathe in accordance with target motion derived from four-dimensional computed tomography (4D-CT). Violations of mechanical limitations of hardware are to be avoided at the treatment planning stage. Hardware modifications are not required. In this article, using sliding window IMRT delivery as an example, we have described step-by-step the implementation of target tracking by the BSD technique: (1) A breathing guide is developed from patients normal breathing pattern. The patient tries to reproduce this guiding cycle by following the display in the goggles; (2) 4D-CT scans are acquired at all the phases of the breathing cycle; (3) The average tumor trajectory is obtained by deformable image registration of 4D-CT datasets and is smoothed by Fourier filtering; (4) Conventional IMRT planning is performed using the images at reference phase (full exhalation phase) and a leaf sequence based on optimized fluence map is generated; (5) Assuming the patient breathes with a reproducible breathing pattern and the machine maintains a constant dose rate, the treatment process is correlated with the breathing phase; (6) The instantaneous average tumor displacement is overlaid on the dMLC position at corresponding phase; and (7) DMLC leaf speed and acceleration are evaluated to ensure treatment delivery. A custom-built mobile phantom driven by a computer-controlled stepper motor was used in the dosimetry verification. A stepper motor was programmed such that the phantom moved according to the linear component of tumor motion used in BSD treatment planning. A conventional plan was delivered on the phantom with and without motion. The BSD plan was also delivered on the phantom that moved with the prescheduled pattern and synchronized with the delivery of each beam. Film dosimetry showed underdose and overdose in the superior and inferior regions of the target, respectively, if the tumor motion is not compensated during the delivery. BSD delivery resulted in a dose distribution very similar to the planned treatments.

Advances in radiation therapy dosimetry

During the last decade, there has been an explosion of new radiation therapy planning and delivery tools. We went through a rapid transition from conventional three-dimensional (3D) conformal radiation therapy to intensity-modulated radiation therapy (IMRT) treatments, and additional new techniques for motion-adaptive radiation therapy are being introduced. These advances push the frontiers in our effort to provide better patient care; and with the addition of IMRT, temporal dimensions are major challenges for the radiotherapy patient dosimetry and delivery verification. Advanced techniques are less tolerant to poor implementation than are standard techniques. Mis-administrations are more difficult to detect and can possibly lead to poor outcomes for some patients. Instead of presenting a manual on quality assurance for radiation therapy, this manuscript provides an overview of dosimetry verification tools and a focused discussion on breath holding, respiratory gating and the applications of four-dimensional computed tomography in motion management. Some of the major challenges in the above areas are discussed.

Rapid Automated Target Segmentation and Tracking on 4D Data without Initial Contours

Purpose. To achieve rapid automated delineation of gross target volume (GTV) and to quantify changes in volume/position of the target for radiotherapy planning using four-dimensional (4D) CT. Methods and Materials. Novel morphological processing and successive localization (MPSL) algorithms were designed and implemented for achieving autosegmentation. Contours automatically generated using MPSL method were compared with contours generated using state-of-the-art deformable registration methods (using Elastix© and MIMVista software). Metrics such as the Dice similarity coefficient, sensitivity, and positive predictive value (PPV) were analyzed. The target motion tracked using the centroid of the GTV estimated using MPSL method was compared with motion tracked using deformable registration methods. Results. MPSL algorithm segmented the GTV in 4DCT images in 27.0 ± 11.1 seconds per phase (512 × 512 resolution) as compared to 142.3 ± 11.3 seconds per phase for deformable registration based methods in 9 cases. Dice coefficients between MPSL generated GTV contours and manual contours (considered as ground-truth) were 0.865 ± 0.037. In comparison, the Dice coefficients between ground-truth and contours generated using deformable registration based methods were 0.909 ± 0.051. Conclusions. The MPSL method achieved similar segmentation accuracy as compared to state-of-the-art deformable registration based segmentation methods, but with significant reduction in time required for GTV segmentation.

TU‐D‐AUD‐05: Comparison for Analytical Anisotropic Algorithm and Adaptive Collapse Cone Convolution Algorithm for Small Field Dosimetry

Purpose: The objective of this study is to compare the calculation accuracy of Analytical Anisotropic Algorithm (AAA) to that of Adaptive Collapse Cone Convolution (CCC) algorithm for small beam sizes where the electron disequilibrium becomes significant. Method and Materials: The comparison of CCC and AAA calculations were performed for various field sizes (2×2, 4×4, 6×6 and 10×10) at 6‐ and 10‐MV photon energies on different phantom situations (homogeneous water phantom, cork/water heterogeneous phantoms, step phantom and IMRTlung phantom). Treatment plans for open beams, MLC shaped beams and IMRT plans were created and calculated by each algorithm. Absolute dose comparisons were made based on measurements, calculations and Monte Carlo Simulations. The XV radiographic films, GAFCHROMIC_EBT radiochromic films, PTW Ion Chamber and Sun Nuclear Map checker were used for measurements. Results and Conclusions: Point dose comparison along the central axis of beams shows that in homogeneous phantom AAA predicts dose within 2%, which is compatible to CCC. For the heterogeneous phantom with vertical density gradient, AAA predicts an up to 5% difference and compared to CCC with less than 2.5% difference. Depth Dose curves also showed that AAA overestimates the dose after passing through low density region. But inside the low density region, AAA gives a compatible prediction to CCC for very small fields. For those points far from heterogeneity, AAA results show a relatively good estimation. Profiles at different depths in the phantoms with density gradient along horizontal direction show that AAA does not model lateral scatter adequately which leads to discrepancies of up to ± 7% in the region of ±1cm lateral from the heterogeneous interface, compared to less than 2% for CCC. Planar dose comparison for IMRT plans shows AAA calculated dose fluence matches with film measured data.

SU‐EE‐A3‐05: Evaluation of Kilo‐Voltage Cone Beam CT Image Quality in Context to Dose Re‐Computation

Purpose: The purpose of this study was to evaluate volumetric kV Cone Beam CT(CBCT)image quality at different scan parameter settings in context to treatment planning tolerances. Method and Materials: Both large and small density phantoms with eight density inserts were scanned by GE LS CT/PET system, as well as the Varians OBITM system in half fan and full fan scanning modes. Scans for CBCTimages were performed at different tube currents (20‐, 40‐ and 80‐mA) and source‐imager distance (SID) (150cm and 160cm) after prior calibration of each mode. Deviation of the Hounsfield Unit (HU) values at different settings compared to conventional kV CTimages were obtained for further evaluation. We also adjusted the CT number in CTimages to simulate CBCT artifacts that was not produced by our experiments, and to see how much degradation of image would violate dosimetric feasibility of CBCT based treatment planning. Treatment plans for single beam or multiple beams were calculated based on CT,CBCT and modified CTimages for various phantoms geometries and patients. Results and Conclusions: Results show that the HU for different anatomies in the body have different amount of change for different scan parameters settings (including current, SID and fan angle used) for CBCTimage acquisition. Larger variations in HU appeared in lung and dense bone regions, compared to those with HU closer to tissue. Maximum variations in HU were found in the images with data truncation. Dose profiles, dose volume histograms, isodose distributions and Gamma values of CBCT based plan with images scanned at full fan mode agree relatively well with CT based plan. Larger dose discrepancy appears in lung or dense bone region. Results from the CT‐modified images based plans show that the dosimetric error becomes significant as the HU variation goes beyond 50.

SU-FF-T-448: Validation of a New Photon Dose Calculation Model-Analytical Anisotropic Algorithm

Purpose: To validate a new photon dose calculation model Analytical Anisotropic Algorithm (AAA) on Eclipse™ treatment planning system (TPS). Comparison of AAA dose calculation was performed with measurements and other two conventional algorithms, Pencil Beam Convolution (PBC) algorithm on Eclipse™ and Collapsed Cone Convolution/Superposition (CCC) algorithm on Pinnacle3.0 TPS. Method and Materials: Four phantoms were CT scanned and the image set was imported into both TPS for dose computation and analysis. The four phantoms were: 1) homogenous tissue equivalent phantom, 2) tissue equivalent phantom with infinite lung heterogeneity, 3) tissue equivalent phantom with finite lung, 4) IMRT dose verification phantom. Measurements were made by exposing the phantom using Varian Linac. Point measurement and film measurements were compared with calculated results from the three algorithms. Dose responses for high and low energy photon beams were investigated for several different depths and PDD curves were compared in the phantom for various field sizes. The IMRT plans were generated by both TPS and were performed on the IMRT phantom to compare fluence maps. Results: AAA dose prediction fits the film measurements well except that there is up to ±6% discrepancy for dose profile perpendicular to the interface of tissue and lung. Point measurements support the AAA algorithm calculations. AAA also accurately predicts the decrease in PDD curves due to the lung inhomogeneity for 6MV energy. For the high energy photon beam and very small field size (2cm*2cm) in lung region, AAA prediction is up to 8% lower than the measurements. Conclusion: AAA algorithm accounts for attenuation corrections and electron transport, and models the deposited dose in the lung with greater accuracy than PBC. It is also faster than CCC algorithm. AAA algorithm can not accurately model the lateral scattering in tissue heterogeneity, but it can still give a reasonably close (within ±6%) prediction.

SU‐FF‐J‐42: Cone Beam CT Based Treatment Planning

Purpose: To evaluate treatment planning based on cone beam CT(CBCT) using latest software on a LINAC 21IX CBCTimaging system. Method and Materials: An anthropomorphic chest phantom having bone, soft tissue, and lung components was used to create and evaluate treatment plans based on conventional CT and CBCTimages. Conventional CTimages of 2.5 mm slice thickness were taken with a GE discovery LS CT/PET system. CBCTimages slices were also reconstructed from flat panel system on a Varian LINAC 21IX. Eclipse treatment planning system was used to compare treatment plans from the conventional CT and CBCTimages. The AAA algorithm was used in the treatment planning system for inhomogeneity correction. Regions of interest around the bone, soft tissue, and lung in both CT and CBCTimages using identical HU threshold values were drawn. Identical targets located in the lung were used in each treatment plan. Analysis of the treatment plans was performed by comparison of geometrical dimensions, total volumes and dose volume histograms of the target and regions of interest. Results: Geometric comparison of actual external spatial dimensions and others in lung and bone were found to be within 1 mm. Volumetric comparison of the regions of interest resulted in a 2.8% difference of the vertebrae, 3.3% of the right lung, and 3.7% of the total external volume. Dosimetric results show similar dose distributions. Dose volume histograms are also comparable. Conclusion: Results demonstrate that treatment planning based on CBCT is feasible. Plans created from CBCTimages are comparable to plans created with conventional CT systems. Conflict of Interest: Partly funded by Prostate Cancer Foundation grant UW 133‐HR30.

SU‐E‐J‐146: Time Series Prediction of Lung Cancer Patients’ Breathing Pattern Based on Nonlinear Dynamics

PURPOSE Prediction methods for breathing patterns, which are crucial to deal with system latency in treatments of moving lung tumors using state-space methodologies based on non-linear dynamics are contrasted to linear predictive methods. METHOD AND MATERIALS In our previous work we established that breathing patterns can be described as a 5-6 dimensional nonlinear, stationary and deterministic system that exhibits sensitive dependence on initial conditions. In this work, nonlinear prediction methods are used to predict the short-term evolution of the respiratory system for 3 patients. Single step and N-point multi step prediction are performed for sampling rates of 5Hz, 10Hz, and 30Hz. We compare the employed nonlinear prediction methods with respect to prediction accuracy to Infinite Impulse Response (IIR) prediction filters. The simplest form of local prediction is finding similar segments of scalar time series data in a higher dimensional embedding space. Hence, we predict the future value x(t)of N-time steps ahead by simply finding the average of nearest neighbor points to the point x(t) in the past and using them to estimate x(t+N), yielding a local average model (LAM). Local linear models (LLM) which are linear autoregressive models that hold only for a region around the target point formed by the nearest neighbor points is combined with a set of linear regularization techniques to solve ill-posed regression problems are also implemented. RESULTS For all sampling frequencies, both single step and N-point multi step prediction results obtained using LAM and LLM with regularization methods are better than IIR prediction filters for the selected sample patients. CONCLUSIONS The use of non-linear prediction methods for predicting the breathing pattern of lung cancer patients may lead to improved, robust and accurate long-term prediction to account for system latencies.

TH‐A‐220‐06: Four‐Dimensional MRI/CT Based Auto‐Adaptive Segmentation for Real‐Time Radiotherapy in Lung Cancer Treatment

Purpose: To enable real‐time magnetic resonance (MR)/computer tomography(CT)image‐guidedradiotherapy (IGRT) and to reduce treatment planning durations in the radiotherapy of lungcancer through the design and implementation of computationally efficient four‐dimensional (4D) MR/CT based automated segmentation algorithms. Methods: Hyperpolarized helium‐3 (HP3He) and proton‐density 4DMRI lung data was acquired for six subjects. Thoracic 4DCT data often subjects with lungcancer was also acquired. Automated segmentation was performed using a novel Morphological Processing and Successive Localization (MPSL) approach. Three different MPSL segmentation algorithms were developed to segment the regions of tumor, body and lung respectively. MPSL segmentation: A mask that includes the intensity range of the region‐of‐interest was generated. Then morphological processing (and/or reconstruction) was performed to separate the target volume(s) from other regions. Then the different connected regions were labeled using the union‐find algorithm and their areas/volumes were calculated. A limit on the maximum/minimum possible area/volume of the target volume was used as a filter to segment the target volume. Morphological processing/reconstruction were performed again to create the final contours. Results: MPSL was shown to successfully segment the regions‐of‐interest (tumors,lung and body) from the images with both high and low signal‐to‐ noise ratio. With the use of 3D processing and successive localization MPSL was shown to separately classify tumor and diaphragm (that may appear within the 2D contours of the lung). MPSL segmentation was compared with manual segmentation. Average computational time for achieving automated lung segmentation using MPSL on one phase volume (128×128×128) of 4D HP3He MR data was 0.5s. For proton‐density 4DMRI data, the average time for automated segmentation of body and lung on one phase volume (128×128×128) was 2s. Conclusions: With the computation speed of the order of seconds for achieving automated segmentation, MPSL has realistic potential for application in real‐time image‐guidance for adaptive radiotherapy.