Markerless Tracking of Lung Tumors on Continuous kV Images

S842

3606 Margin Assessment and Optimization in Prostate Irradiation by Quantitative Analysis of Daily Megavoltage Computed Tomography T. Wang,1,2 Y. Lee,3 L. Chou,1 J. Chen,1 and Y. Liu1,2; 1Cancer Center, Taipei Veterans General Hospital, Taipei, Taiwan, 2School of Medicine, National Yang-Ming University, Taipei, Taiwan, 3Center for Condensed Matter Science, National Taiwan University, Taipei, Taiwan Purpose/Objective(s): One major challenge in radiation therapy is to determine the appropriate planning target volume (PTV) margin. The ability to secure tumor coverage while reducing normal tissue toxicity is especially critical in treating prostate cancer in the modern dose-escalation era. The goal of this study is to develop a margin assessment and optimization method that takes into consideration the organ deformation, translation and rotation, and the coverage of CTV and the overlapping of organ at risk (OAR). Materials/Methods: We obtained daily mega-voltage computed tomography (MVCT) data from 20 prostate cancer patients who underwent definitive radiation therapy. Each patient received one planning computed tomography simulation scan (CTsim) and 30-38 MVCT. For all scans, one physician delineated the prostate, the seminal vesicles, the rectum and the bladder. Three off-line matching strategies (skin markers, bone, and soft tissue) were performed for the MVCT-CTsim registration, from which translation and rotation registration parameters were collected. After performing the match, contours were exported to Matlab for analysis. By combining contours of all fractions, a 3D probability density rendering for the four organs was produced for each patient and then merged with CTsim. Two margin recipes were created. The first was an isotropic margin with 2 to 10 mm expansion of either the prostate alone or prostate with seminal vesicles and the second was an anisotropic margin with modification of anterior and posterior borders. Finally, both margin recipes were tested for both CTV coverage and OAR overlapping percentage. Results: The mean isotropic margin to achieve 95% of CTV coverage in skin markers, bone, and soft tissue registration were 9.7, 5.9, and 5.8 mm, while the overlapping of rectum/bladder was 5.1/16.3, 5/6.8, and 3.7/7.1 % of total volume. In anisotropic margin, a reduced anterior/posterior margin of 3.9/9.6, 2.4/3.7, and 2/3.7 mm in skin markers, bone, and soft tissue registration can achieve comparable CTV coverage and better OARs sparing than the isotropic margin recipe. Conclusions: Daily 3D IGRT enable analysis of the inter-fractional change of organs in the presence of deformation, translation and rotation. Utilizing this data, we introduced a new margin recipe that optimizes both CTV coverage and OAR sparing. For the patient sample of this study, anisotropic margin with reduced anterior/posterior borders generates optimal PTV. By building a larger database using this technique, we can find a margin recipe that is most suitable for each patient type. Author Disclosure: T. Wang: None. Y. Lee: None. L. Chou: None. J. Chen: None. Y. Liu: None.

3607 Development of Image Registration for Novel Megavoltage Topogram for Patient Localization in Tomotherapy J. Scholey,1 X. Qi,2 D. Low,3 and B. White1; 1The University of Pennsylvania, Philadelphia, PA, 2Dept. of Radiation Oncology, UCLA, Los Angeles, CA, 3University of California Los Angeles, Los Angeles, CA Purpose/Objective(s): Orthogonal scout images (topograms) can be a fast and accurate alternative to existing megavoltage computed tomography (MVCT) for patient alignment on TomoTherapy HiART system. Patient localization with the on-board MVCT delivers non-negligible dose and has poor temporal resolution compared to kilovoltage computed tomography (kVCT). A potential limiting factor in translating MV topograms into clinical use has been the low quality of the images. The goal of this study was to improve the image quality of MV topograms for better patient localization. Materials/Methods: Digitally reconstructed radiographs (DRR) of anthropomorphic head and pelvis phantoms were synthesized from kVCT

International Journal of Radiation Oncology  Biology  Physics Scientific Abstract 3607; Table shifts using MV topograms

Accuracy of recovered daily patient positing

AP (mm)

RL (mm)

Head

0.690.93

0.730.69

Pelvis

0.440.50

0.460.54

SI (mm) 0.670.83 1.331.11 0.690.75 0.830.77

(AP Topo) (LAT Topo) (AP Topo) (LAT Topo)

under Tomotherapy geometry. Lateral (LAT) and anterior-posterior (AP) aligned topograms were acquired with a couch speed of 1 cm/s. The phantoms were rigidly translated in all spatial directions with known offsets in increments of 5 mm, 10 mm, and 15 mm to simulate daily positioning errors. The contrast of the MV topograms was automatically adjusted based on the image intensity characteristics. A low-pass frequency filter in Fourier space and a Weiner filter were implemented to reduce the stochastic noise caused by scattered radiation to the detector array. An intensity-based image registration algorithm was used to register the MV topograms to a corresponding kVDRR by minimizing the mean square error between corresponding pixel intensity. The registration accuracy was assessed by comparing the normalized cross correlation coefficients (NCC) between the registered topograms and the kVDRR. The applied phantom offsets were determined by registering the MV topograms with the kVDRR and recovering the spatial translation of the MV topograms. Results: The NCC coefficients for the head phantom were 0.960.01 (AP) and 0.950.01 (LAT) for filtered topogram registration and 0.760.01 (AP) and 0.860.01 (LAT) for unfiltered topogram registration. The NCC coefficients for the pelvis phantom were 0.940.02 (AP) and 0.950.01 (LAT) for filtered topogram registration and 0.890.01 (AP) and 0.740.01 (LAT) for unfiltered topogram registration. The automatic registration technique provided sub-millimeter accuracy. The results can be seen in Table. Conclusions: The automatic registration of the filtered MV topograms to a corresponding kVDRR and the recovered simulated daily positioning errors were accurate to the order of a millimeter. These results suggest the clinical use of MV topograms as a promising alternative to MVCT patient alignment. Author Disclosure: J. Scholey: None. X. Qi: None. D. Low: None. B. White: None.

3608 Markerless Tracking of Lung Tumors on Continuous kV Images J.R. van Sornsen de Koste, M. Dahele, S. Senan, B. Slotman, and W. Verbakel; VU University Medical Center, Amsterdam, Netherlands Purpose/Objective(s): To retrospectively evaluate markerless tracking of small lung tumors using kV projection images from cone beam CT and benchmark it using known tumor motion. Materials/Methods: Non-clinical research tumor tracking software (TT) used normalized cross correlation matching to track the tumor on individual kV projection images from a CBCT and compared the position with a reference dataset of digitally reconstructed kV projection images derived from one phase of the planning 4DCT. TT matches in 2 coordinates and determines the 3rd coordinate by triangulation using previous match results. For 2 patients (3 fractions each), the TT-assessed clinical tumor motion was compared with the tumor motion scored on the 4DCT and clinical TT-tracked tumor motion was compared with the motion of an external marker box (RPM). To benchmark TT it was tested against known tumor motion. To simulate a clinical kV projection series, 360 projection images (1 image/1 ) were digitally reconstructed for the 10 phases of 4DCT. From these, a kV projection series was composed to simulate a breathing cycle every 20 . The reference dataset was 360 projection images from the 4DCT expiration phase. TT was performed using these datasets. Results were correlated with displacement of the tumor’s center of mass on 4DCT. Results: Tumor A (6.6cm3): 4DCT motion X Z 3.5 mm, Y Z 11.3 mm and Z Z 5.5 mm. 360 TT on the clinical kV projection series showed mean

Volume 90  Number 1S  Supplement 2014 motion X Z 6.3 (range: 5.9-7.1) mm, Y Z 10.4 (9.6-11.2) mm, and Z Z 7.1 (6.6-7.7) mm. Root Mean Square (RMS) error between TT-tracked tumor motion and RPM was <0.9mm in all directions. Tumor B (4cm3): TT could not track the position of the tumor at all angles as half-fan CBCT captured the tumor in only half of the images and high density and central airway structures obscured the target. Therefore, analysis was only performed for 100 (300-40 ) kV-source rotation. 4DCT motion was X Z 1.8, Y Z 7.5 and Z Z 8.6 mm. 100 TT on clinical kV projection series showed mean motion X Z 3.7 (2.5-5.8) mm, Y Z 7.3 (5.3-9.2) mm, and Z Z 7.8 (5.5-10.6) mm. RMS error between TT and RPM was <1.1mm. Benchmarking tumor A: 360 TT showed mean motion X Z 3.5, Y Z 11.6 and Z Z 5.2 mm with standard deviation (SD) <0.7mm. Correlation coefficients of TT-tracked tumor motion with center of mass displacement on 4DCT were 0.64, 0.95 and 0.84 (X, Y, Z). For tumor B: 2 sets of TT over 100 (300-40 /120-220 ) showed mean motion X Z 1.9, Y Z 7.7 and Z Z 8.4mm (SD<0.8mm) and correlation coefficients of 0.63, 0.94 and 0.91 (X, Y, Z). Conclusions: TT software can track small lung tumors when they are visible in kV projections. Tumor motion during CBCT can vary from the planning 4DCT. TT and RPM motion were closely associated. Benchmarking TT against known tumor displacements showed good agreement. Author Disclosure: J.R. van Sornsen de Koste: I. Travel Expenses; has received travel support of Varian Medical Systems. M. Dahele: I. Travel Expenses; has received travel support/honoraria from Varian Medical Systems and travel support from Brainlab. S. Senan: I. Travel Expenses; has received travel support and honoraria from Varian Medical Systems. B. Slotman: I. Travel Expenses; has received travel support/honoraria from Varian Medical Systems. W. Verbakel: I. Travel Expenses; has received travel/honoraria support from Varian Medical Systems.

3609 Feasibility of X-Ray Acoustic Computed Tomography as a Tool for Noninvasive Volumetric In Vivo Dosimetry S. Hickling,1 M. Hobson,2 and I. El Naqa2; 1McGill University, Montreal, QC, Canada, 2McGill University Health Centre, Montreal, QC, Canada Purpose/Objective(s): We propose that the novel modality of x-ray acoustic computed tomography (XACT) has the potential to be an effective tool for non-invasive in vivo dosimetry during external beam radiation therapy. XACT is based on the principle that acoustic waves proportional to the dose deposited are created following each radiation pulse. After detecting these acoustic waves with an ultrasound transducer array, an image of the dose distribution can be reconstructed. This work tests the feasibility of using XACT as a real-time dosimeter by performing realistic simulations to determine the expected amplitude and frequency of the induced acoustic waves for a typical clinical prostate case. The dose distribution is then reconstructed and compared to the treatment plan. Materials/Methods: Commercially available treatment planning software was used to obtain the dose distribution for a clinical prostate patient treated using a four-field box technique with 18 MV photon beams. A 2D map of the initial acoustic pressure distribution for a slice in the middle of the target was calculated for each LINAC pulse using the dose and CT data. An open source toolkit for the simulation of acoustic wave fields was then used to generate the expected time-varying signal at 360 transducer locations spaced equally around the patient. The transducers were assumed to be ideal and able to detect all wave frequencies up to a maximum simulated frequency of 1.6 MHz. Acoustic wave attenuation was not accounted for. A time reversal reconstruction algorithm was then used to obtain an XACT image of the dose for each LINAC pulse, after which a composite dose distribution for the entire fraction was derived. Results: The detected acoustic waves had a differential pressure amplitude on the order of 1 Pa to 10 Pa, with the main frequency component of the signal falling below 1 MHz. The reconstructed dose distribution closely resembled the plan, with 89% of pixels passing a 3% / 3 mm 2D gamma test. The largest dose discrepancies occurred at the interfaces of anatomical structures, such as the femur, pelvic bone, and seminal vesicles. This could be caused by the lack of frequency information above 1.6 MHz and the finite number of transducer positions.

Poster Viewing Abstracts S843 Conclusions: The simulated amplitude and frequency of the induced acoustic waves for a typical clinical prostate case indicate that they should be detectable with commercial ultrasound transducers. Further investigation of detector geometry, reconstruction algorithms and image post-processing could improve the agreement of the reconstructed and original dose distributions. An experimental validation of these simulations is ongoing. We conclude that XACT is a promising technique for volumetric non-invasive in vivo dosimetry and merits further research. Author Disclosure: S. Hickling: None. M. Hobson: None. I. El Naqa: None.

3610 Quantification of PTV Margin When Using a Robotic Radiosurgery System to Treat Lung Tumors With Spine Tracking J.A. James, B. Lynch, C. Swanson, B. Wang, and N.E. Dunlap; University of Louisville, Louisville, KY Purpose/Objective(s): The use of fiducial markers or direct tumor visualization allows for tumor tracking and ultimately smaller PTV margins when treating lung tumors, yet many patients are either not amenable to fiducial marker placement or tumors are unable to be visualized on orthogonal x-rays. Spine tracking is an alternative method for localizing the tumor but is limited by the assumption that the location of the lung tumor relative to the spine is constant. The purpose of this study is to quantify the additional PTV margin needed when using spine tracking to ensure the ITV receives the prescription dose during treatment. Materials/Methods: Daily CBCTs, registered based on tumor position, from 63 patients treated with lung SBRT were collected and analyzed. Rigid registrations were re-performed so that the position of the spine on the CBCT was aligned to its position on the planning CT. Shifts from the treatment position to the new position were recorded, and per patient mean shifts and standard deviations were calculated as well as group systematic and random standard deviations. This data was used with van Herk’s margin recipe to determine the additional margin required to adequately treat the patient population if spine tracking were used instead of direct daily tumor imaging. A retrospective dosimetric analysis was also performed on 6 lung patients previously treated on CyberKnife using spine tracking to determine the potential decrease in target coverage due to insufficient margin on the ITV. This analysis was performed by shifting the PTV volume relative to the CyberKnife treatment geometry to simulate a setup error due to tracking the spine as opposed to the tumor. Results: The additional margin calculated by van Herk’s margin recipe to adequately cover the ITV with the 95% isodose surface for 90% of the entire patient population in the vertical, longitudinal, and lateral directions are 6.4, 6.0 and 4.5 mm, respectively. The retrospective analysis showed a decrease in PTV coverage from 95.6% to 93.1% and an increase in new conformity index (nCI) by 2.7% when using the average shift data to simulate setup error. When using the maximum shift data to simulate the worst possible outcome, the PTV coverage decreased to 73.4% and the nCI increased by 26.8%. Conclusions: Standard margins of 5 mm on the ITV for treating lung SBRT patients is insufficient and may result in geographic misses of the tumor when using spine tracking to locate the position of tumor in the lung. Therefore, we recommend the addition of 5 mm margins in all directions for a total of 10 mm to take into account the change in position of the tumor relative to the spine from the time of simulation to treatment. Author Disclosure: J.A. James: A. Employee; clinical staff at the University of Louisville. B. Lynch: A. Employee; clinical staff for the Oncology Services of North Alabama. C. Swanson: A. Employee; clinical staff for Baptist Health. B. Wang: A. Employee; Faculty at the University of Louisville. N.E. Dunlap: A. Employee; Faculty at the University of Louisville.

3611 Feasibility of Gating Using a Magnetic-Resonance Image Guided Radiation Therapy (MR-IGRT) System R. Kashani,1 K. Tanderup,1 J.R. Victoria,2 L. Santanam,1 H. Wooten,1 O.L. Green,1 J.F. Dempsey,2 and S. Mutic1; 1Washington University School of Medicine, St. Louis, MO, 2ViewRay, Cleveland, OH