Digitally Reconstructed Radiographs Research Articles

Setup margins and geometric uncertainties in intensity-modulated radiation therapy in treating pituitary adenomas: the experience of Lyon Sud Hospital

The authors retrospectively evaluated the setup uncertainties in Intensity-Modulated Radiation Therapy (IMRT) for pituitary adenomas and verified the margins used in daily practice (3 mm). Craniocaudal (CC), anteroposterior (AP) and laterolateral (LL) displacements were measured during the first 3 days of treatment and then weekly by comparing two orthogonal images obtained by an electronic system of portal imaging with Digitally Reconstructed Radiographs (DRRs). Setup Margins (SM) were defined according to the International Commission on Radiation Units (ICRU)-62 formula, the Stroom equation and the van Herk equation. The systematic (Σ) and random (σ) errors of the population were calculated as standard deviation (SD) of the population mean and the mean of SDs for every patient, respectively. Twenty patients were treated by IMRT for pituitary adenomas, and a total of 231 measurements were obtained. Σ and σ were 0.6 and 1.3 mm, 0.8 and 1 mm, 1.2 and 1.5 mm in the AP, LL and CC direction, respectively. Larger setup margin was 2.4, 2.7 and 4 mm in the AP, LL and CC direction, respectively (van Herk formula). IMRT is a highly sophisticated treatment technique that requires precise definition and optimisation of local setup errors and, finally, of the irradiated volumes. The role of image-guided RT in these kinds of treatments should be prospectively evaluated.

Evaluation of automated image registration algorithm for image-guided radiotherapy (IGRT)

The performance of an image registration (IR) software was evaluated for automatically detecting known errors simulated through the movement of ExactCouch using an onboard imager. Twenty-seven set-up errors (11 translations, 10 rotations, 6 translation and rotation) were simulated by introducing offset up to ± 15 mm in three principal axes and 0° to ± 1° in yaw. For every simulated error, orthogonal kV radiograph and cone beam CT were acquired in half-fan (CBCT_HF) and full-fan (CBCT_FF) mode. The orthogonal radiographs and CBCTs were automatically co-registered to reference digitally reconstructed radiographs (DRRs) and planning CT using 2D-2D and 3D-3D matching software based on mutual information transformation. A total of 79 image sets (ten pairs of kV X-rays and 69 session of CBCT) were analyzed to determine the (a) reproducibility of IR outcome and (b) residual error, defined as the deviation between the known and IR software detected displacement in translation and rotation. The reproducibility of automatic IR of planning CT and repeat CBCTs taken with and without kilovoltage detector and kilovoltage X-ray source arm movement was excellent with mean SD of 0.1 mm in the translation and 0.0° in rotation. The average residual errors in translation and rotation were within ± 0.5 mm and ± 0.2°, ± 0.9 mm and ± 0.3°, and ± 0.4 mm and ± 0.2° for setup simulated only in translation, rotation, and both translation and rotation. The mean (SD) 3D vector was largest when only translational error was simulated and was 1.7 (1.1) mm for 2D-2D match of reference DRR with radiograph, 1.4 (0.6) and 1.3 (0.5) mm for 3D-3D match of reference CT and CBCT with full fan and half fan, respectively. In conclusion, the image-guided radiation therapy (IGRT) system is accurate within 1.8 mm and 0.4° and reproducible under control condition. Inherent error from any IGRT process should be taken into account while setting clinical IGRT protocol.

A study on the reproducibility of the volume in 3-D volume-rendering images for radiation treatment planning

This study examined the effects of a change in the reconstruction cross-section, which depends on the slice thickness and the phantom shape, on the volume changes in a three-dimensional volume-rendering image reconstructed in a computed tomography (CT) simulation. For this, precise phantoms with different shapes and sizes were manufactured, and the three-dimensional volume-rendering images obtained from computed treatment planning for a phantom were measured. The CT simulation, which had been standardized using a phantom, was implemented with slice thicknesses of 1.25, 2.5, 5, and 10 mm. The measurements were performed on the X-axis, the Y-axis and the Z-axis for the accessory phantom to identify the following changes: length, area and volume, measurement deviation between analyzers, and maximum and minimum values. With increasing slice thickness the difference in the visual error range between analyzers increased, the cross section for the scan increased, and the thickness or density in the scan direction became nonuniform. These results are expected to help analyze the cause of the error in digitally-reconstructed radiographs and establish a more precise CT simulation.

Computerized estimation of patient setup errors in portal images based on localized pelvic templates for prostate cancer radiotherapy

We have developed a computerized method for estimating patient setup errors in portal images based on localized pelvic templates for prostate cancer radiotherapy. The patient setup errors were estimated based on a template-matching technique that compared the portal image and a localized pelvic template image with a clinical target volume produced from a digitally reconstructed radiography (DRR) image of each patient. We evaluated the proposed method by calculating the residual error between the patient setup error obtained by the proposed method and the gold standard setup error determined by consensus between two radiation oncologists. Eleven training cases with prostate cancer were used for development of the proposed method, and then we applied the method to 10 test cases as a validation test. As a result, the residual errors in the anterior–posterior, superior–inferior and left–right directions were smaller than 2 mm for the validation test. The mean residual error was 2.65 ± 1.21 mm in the Euclidean distance for training cases, and 3.10 ± 1.49 mm for the validation test. There was no statistically significant difference in the residual error between the test for training cases and the validation test (P = 0.438). The proposed method appears to be robust for detecting patient setup error in the treatment of prostate cancer radiotherapy.

Fast reconstructed radiographs from octree-compressed volumetric data

Simulated 2D X-ray images called digitally reconstructed radiographs (DRRs) have important applications within medical image registration frameworks where they are compared with reference X-rays or used in implementations of digital tomosynthesis (DTS). However, rendering DRRs from a CT volume is computationally demanding and relatively slow using the conventional ray-casting algorithm. Image-guided radiation therapy systems using DTS to verify target location require a large number DRRs to be precomputed since there is insufficient time within the automatic image registration procedure to generate DRRs and search for an optimal pose. DRRs were rendered from octree-compressed CT data. Previous work showed that octree-compressed volumes rendered by conventional ray casting deliver a registration with acceptable clinical accuracy, but efficiently rendering the irregular grid of an octree data structure is a challenge for conventional ray casting. We address this by using vertex and fragment shaders of modern graphics processing units (GPUs) to directly project internal spaces of the octree, represented by textured particle sprites, onto the view plane. The texture is procedurally generated and depends on the CT pose. The performance of this new algorithm was found to be 4 times faster than that of a ray-casting algorithm implemented using NVIDIA™Compute Unified Device Architecture (CUDA™) on an equivalent GPU (~95% octree compression). Rendering artifacts are apparent (consistent with other splatting algorithm), but image quality tends to improve with compression and fewer particles are needed. A peak signal-to-noise ratio analysis confirmed that the images rendered from compressed volumes were of marginally better quality to those rendered using Gaussian footprints. Using octree-encoded DRRs within a 2D/3D registration framework indicated the approach may be useful in accelerating automatic image registration.

Fast DRR generation for 2D to 3D registration on GPUs

The generation of digitally reconstructed radiographs (DRRs) is the most time consuming step on the CPU in intensity based two-dimensional x-ray to three-dimensional (CT or 3D rotational x-ray) medical image registration, which has application in several image guided interventions. This work presents optimized DRR rendering on graphical processor units (GPUs) and compares performance achievable on four commercially available devices. A ray-cast based DRR rendering was implemented for a 512 × 512 × 72 CT volume. The block size parameter was optimized for four different GPUs for a region of interest (ROI) of 400 × 225 pixels with different sampling ratios (1.1%-9.1% and 100%). Performance was statistically evaluated and compared for the four GPUs. The method and the block size dependence were validated on the latest GPU for several parameter settings with a public gold standard dataset (512 × 512 × 825 CT) for registration purposes. Depending on the GPU, the full ROI is rendered in 2.7-5.2 ms. If sampling ratio of 1.1%-9.1% is applied, execution time is in the range of 0.3-7.3 ms. On all GPUs, the mean of the execution time increased linearly with respect to the number of pixels if sampling was used. The presented results outperform other results from the literature. This indicates that automatic 2D to 3D registration, which typically requires a couple of hundred DRR renderings to converge, can be performed quasi on-line, in less than a second or depending on the application and hardware in less than a couple of seconds. Accordingly, a whole new field of applications is opened for image guided interventions, where the registration is continuously performed to match the real-time x-ray.

Volume-of-change cone-beam CT for image-guided surgery

C-arm cone-beam CT (CBCT) can provide intraoperative 3D imaging capability for surgical guidance, but workflow and radiation dose are the significant barriers to broad utilization. One main reason is that each 3D image acquisition requires a complete scan with a full radiation dose to present a completely new 3D image every time. In this paper, we propose to utilize patient-specific CT or CBCT as prior knowledge to accurately reconstruct the aspects of the region that have changed by the surgical procedure from only a sparse set of x-rays. The proposed methods consist of a 3D–2D registration between the prior volume and a sparse set of intraoperative x-rays, creating digitally reconstructed radiographs (DRRs) from the registered prior volume, computing difference images by subtracting DRRs from the intraoperative x-rays, a penalized likelihood reconstruction of the volume of change (VOC) from the difference images, and finally a fusion of VOC reconstruction with the prior volume to visualize the entire surgical field. When the surgical changes are local and relatively small, the VOC reconstruction involves only a small volume size and a small number of projections, allowing less computation and lower radiation dose than is needed to reconstruct the entire surgical field. We applied this approach to sacroplasty phantom data obtained from a CBCT test bench and vertebroplasty data with a fresh cadaver acquired from a C-arm CBCT system with a flat-panel detector. The VOCs were reconstructed from a varying number of images (10–66 images) and compared to the CBCT ground truth using four different metrics (mean squared error, correlation coefficient, structural similarity index and perceptual difference model). The results show promising reconstruction quality with structural similarity to the ground truth close to 1 even when only 15–20 images were used, allowing dose reduction by the factor of 10–20.

GPU Accelerated Generation of Digitally Reconstructed Radiographs for 2-D/3-D Image Registration

Recent advances in programming languages for graphics processing units (GPUs) provide developers with a convenient way of implementing applications which can be executed on the CPU and GPU interchangeably. GPUs are becoming relatively cheap, powerful, and widely available hardware components, which can be used to perform intensive calculations. The last decade of hardware performance developments shows that GPU-based computation is progressing significantly faster than CPU-based computation, particularly if one considers the execution of highly parallelisable algorithms. Future predictions illustrate that this trend is likely to continue. In this paper, we introduce a way of accelerating 2-D/3-D image registration by developing a hybrid system which executes on the CPU and utilizes the GPU for parallelizing the generation of digitally reconstructed radiographs (DRRs). Based on the advancements of the GPU over the CPU, it is timely to exploit the benefits of many-core GPU technology by developing algorithms for DRR generation. Although some previous work has investigated the rendering of DRRs using the GPU, this paper investigates approximations which reduce the computational overhead while still maintaining a quality consistent with that needed for 2-D/3-D registration with sufficient accuracy to be clinically acceptable in certain applications of radiation oncology. Furthermore, by comparing implementations of 2-D/3-D registration on the CPU and GPU, we investigate current performance and propose an optimal framework for PC implementations addressing the rigid registration problem. Using this framework, we are able to render DRR images from a 256×256×133 CT volume in ~24 ms using an NVidia GeForce 8800 GTX and in ~2 ms using NVidia GeForce GTX 580. In addition to applications requiring fast automatic patient setup, these levels of performance suggest image-guided radiation therapy at video frame rates is technically feasible using relatively low cost PC architecture.

Poster - Thur Eve - 75: Towards MR only simulation: MR based digitally reconstructed radiograph of head and neck.

To develop a practical method to obtain bony structures from Magnetic Resonance (MR) images and to create MR- based digitally reconstructed radiographs (DRR) for MR only simulation. Using T1-weighted MR images, air regions including the sinuses and the airway in the head and neck were manually contoured. The bone and soft tissue masks were automatically generated based on the statistical data calculated from the air contour and MR intensities. "CT like" MR images were generated by mapping the MR intensities of the voxels within these masks into the CT number ranges of these tissues. The MR-based DRRs created from "CT like" MR images were quantitatively evaluated using the co-registered MR and CT images of 10 stereotactic radiosurgery CNS patients. Ten anatomical control points, set on the contours of the skull segmented using a threshold of 300 HU were used to determine the differences in distance between MR-based DRRs and CT-based DRRs, and to evaluate the geometrical accuracy of MR-based DRRs. The bony structures were visible in the MR-based DRRs. The mean geometric difference and standard deviation between the ten anatomical control points on MR- based and CT-based DRRs were -0.03±1.11 mm (including uncertainty of image fusion). The maximum distance difference was 1.67mm. The study provides a practical method to generate MR- based DRRs from MR-only simulations of the head and neck regions. The image quality and anatomical accuracy of MR-based DRRs is comparable to that of CT-based DRRs.

4D digitally reconstructed radiography for verifying a lung tumor position during volumetric modulated arc therapy

We have proposed four dimensional (4D) digitally reconstructed radiography (DRR) for verifying a lung tumor position during volumetric modulated arc therapy (VMAT). An internal target volume (ITV) was defined based on two clinical target volumes (CTVs) delineated on maximum exhalation and maximum inhalation images acquired by 4D planning computed tomography (CT). A planning target volume (PTV) was defined by adding a margin of 5 mm to the ITV on the maximum exhalation 3D CT images. A single-arc VMAT plan was created on the same CT data using Pinnacle SmartArc with a maximum multi-leaf collimator leaf speed of 1 mm/degree, thereby resulting in quasi-conformal field shapes while optimizing each beam intensity for each gantry angle. During VMAT delivery, cone-beam CT (CBCT) projection data were acquired by an on-board kilovoltage X-ray unit and a flat panel 2D detector. Four CBCT image sets with different respiratory phases were reconstructed using in-house software, where respiratory phases were extracted from the projection data. Subsequently a CTV was delineated on each of the 4D CBCT images by an oncologist. Using the resulting 4D CBCT data including the CTV contours, 4D DRRs during the VMAT delivery were calculated as a function of gantry angle. It was confirmed that the contoured CTV was within the radiation field during the four-fraction lung VMAT delivery. The proposed 4D DRR may facilitate the verification of the position of a respiratory moving lung tumor during VMAT delivery on each treatment day.

Volume-of-interest cone-beam CT using a 2.35 MV beam generated with a carbon target

This is a proof-of-concept study addressing volume of interest (VOI) cone beam CT (CBCT) imaging using an x-ray beam produced by 2.35 MeV electrons incident on a carbon linear accelerator target. Methodology is presented relevant to VOI CBCT image acquisition and reconstruction. Sample image data are given to demonstrate and compare two approaches to minimizing artifacts arising from reconstruction with truncated projections. Dosimetric measurements quantify the potential dose reduction of VOI acquisition relative to full-field CBCT. The dependence of contrast-to-noise ratio (CNR) on VOI dimension is investigated. A paradigm is presented linking the treatment planning process with the imaging technique, allowing definition of an imaging VOI to be tailored to the geometry of the patient. Missing data in truncated projection images are completed using a priori information in the form of digitally reconstructed radiographs (DRRs) generated from the planning CT set. This method is compared to a simpler technique of extrapolating truncated projection data prior to reconstruction. The utility of these approaches is shown through imaging of a geometric phantom and the head-and-neck section of a lamb. The total scatter factor of the 2.35 MV∕carbon beam on field size is measured and compared to a standard therapeutic beam to estimate the comparative dose reduction inside the VOI. Thermoluminescent dosimeters and Gafchromic film measurements are used to compare the imaging dose distributions for the 2.35 MV∕carbon beam between VOI and full-field techniques. The dependence of CNR on VOI dimension is measured for VOIs ranging from 4 to 15 cm diameter. Without compensating for missing data outside of truncated projections prior to reconstruction, pronounced boundary artifacts are present, in three dimensions, within 2-3 cm of the edges of the VOI. These artifacts, as well as cupping inside the VOI, can be reduced substantially using either the DRR filling or extrapolation techniques presented. Compared to 6 MV, the 2.35 MV∕carbon beam shows a substantially greater dependence of total scatter factor on field size, indicating a comparative advantage of the VOI approach when combined with the low-Z target beam. Dosimetric measurements in the anthropomorphic head phantom demonstrate a dose reduction by up to 15% and 75% inside and outside of the VOI, respectively, compared to full-field imaging. For the 2.35 MV∕carbon beam, CNR was shown to be approximately invariant with VOI dimension for bone and lung objects. The low-Z target, VOI CBCT technique appears to be feasible and combines the desirable characteristics of the low-Z target beam with regard to CNR, with the capacity to localize the imaging dose to the anatomy relevant to the image guidance task.

T1/T2*-weighted MRI provides clinically relevant pseudo-CT density data for the pelvic bones in MRI-only based radiotherapy treatment planning

Background and purpose. In radiotherapy (RT), target soft tissues are best defined on MR images. In several cases, CT imaging is needed only for dose calculation and generation of digitally reconstructed radiographs (DRRs). Image co-registration errors between MRI and CT can be avoided by using MRI-only based treatment planning, especially in the pelvis. Since electron density information can not be directly derived from the MRI, a method is needed to convert MRI data into CT like data. We investigated whether there is a relationship between MRI intensity and Hounsfield unit (HU) values for the pelvic bones. The aim was to generate a method to convert bone MRI intensity into HU data surrogate for RT treatment planning. Material and methods. The HU conversion model was generated for 10 randomly chosen prostate cancer patients and independent validation was performed in another 10 patients. Data consisted of 800 image voxels chosen within the pelvic bones in both T1/T2*-weighted gradient echo and CT images. Relation between MRI intensity and electron density was derived from calibrated HU-values. The proposed method was tested by constructing five “pseudo”-CT series. Results. We found that the MRI intensity is related to the HU value within a HU range from 0 to 1400 within the pelvic bones. The mean prediction error of the conversion model was 135 HU. Dose calculation based on the pseudo-CT images was accurate and the generated DRRs were of good quality. Conclusions. The proposed method enables generation of clinically relevant pseudo-CT data for the pelvic bones from one MRI series. It is simpler than previously reported approaches which require either acquisition of several MRI series or T2* maps with special imaging sequences. The method can be applied with commercial clinical image processing software. The application requires segmentation of the bones in the MR images.

TU‐A‐BRA‐09: Phase‐Matched Digital Tomosynthesis (DTS) Imaging for Simultaneous Target Verification during Volumetric Modulated Arc Therapy (VMAT) Treatment

Purpose: To develop a phase‐matched DTS technique to allow tomographic verification of moving targets simultaneous to VMAT delivery. Methods: This study proposes to verify moving targets by comparing phase ‐matched reference DTS (Ref ‐DTS) generated from simulation 4DCT to corresponding on ‐board DTS (OBI ‐DTS) acquired during VMAT delivery. The 4D Digital Extended Cardiac Torso (XCAT) Phantom was used for demonstration. A normal patient breathing pattern was input into the XCAT phantom to generate on ‐board CBCT projections, simulation 4DCT, average intensity projection (AIP) of the 4DCT and free ‐breathing CT (FBCT).These simulation CT sets (FBCT, AIP and 4DCT) were used to project digitally reconstructed radiographs (DRRs): DRR ‐FBCT and DRR ‐AIP (both with no respiratory phase information), and DRR ‐4DCT matched to phases of onboard projections. These DRR sets were further used to reconstruct three Ref ‐DTS sets: Ref ‐DTS ‐FBCT, Ref ‐DTS ‐AIP, and phase ‐matched Ref ‐DTS ‐4DCT. Each Ref ‐DTS set was compared with OBI ‐DTS (generated from on ‐board projections) to identify relative differences between both. To quantitatively assess the feasibility of comparing each Ref ‐DTS set with OBI ‐DTS for target verification, the simulation CT sets were shifted by a given amount to simulate target motion/setup variation, and the resulting Ref ‐DTS sets were registered to OBI ‐DTS automatically. The accuracy of target verification was evaluated by the discrepancies between simulated and registered shifts. Results: For a 1mm discrepancy criterion between simulated and registered shifts, the success rates between OBI‐DTS /Ref‐DTS‐FBCT, OBI‐DTS /Ref‐DTS‐AIP, and OBI‐DTS /Ref‐DTS‐4DCT registrations are 14.3%, 28.6%, and 100% for 2mm shifts, 0%, 14.3%, 100% for 5mm shifts, and 42.9%, 28.6%, 100% for 10mm shifts, respectively. Conclusions: The simulation study indicated that the phase‐matched Ref‐DTS‐4DCT technique considerably improved the target verification accuracy than that provided by non‐phase‐information Ref‐DTS‐FBCT or Ref‐DTS‐AIP. Therefore, it has the potential to validate VMAT treatment of moving targets. (Research partially supported by grant from Varian Medical Systems) Research partially supported by grant from Varian Medical Systems.

WE-G-217BCD-09: Calibration of a DRR Algorithm.

To calibrate an algorithm for digitally reconstructed radiographs (DRRs) such that synthetic projections cast through an object in a fan-beam CT (FBCT) optimally match cone-beam CT projections of the same object. Our DRR algorithm models the transmission of primary photons through an object with a ray-tracing algorithm that samples the CT at small steps along each source-detector pixel path. Sampled CT values are converted to linear attenuation coefficients (LACs), scaled by the step-size, and added to a running sum for each detector pixel. We made attenuation measurements of materials of known CT number to calibrate a CT to LAC conversion function for our FBCT and CBCT systems. DRRs are post- processed to reproduce the tube output, source fluence distribution, and detector response of the CBCT system. In addition, they are modified to account for scatter, beam hardening, and detector veiling glare. Finally, we determined CBCT geometry parameters using a specially designed phantom. We analyzed the quality of our DRR algorithm by directly comparing synthetic DRRs cast through the FBCT of a Catphan® phantom with actual CBCT projections of the phantom. Line plots comparing the intensity profiles of the DRRs and CBCT projections show that uncorrected DRRs do not align perfectly with the projections. However, when scatter, beam hardening, and veiling glare are modeled there is much closer agreement. Simulations with different geometry phantom shapes and sizes recommend a cylindrical phantom 150 mm in diameter. Computational times under one second per DRR are achieved using a GPU for a simulated 1024×768 pixel detector with a 512×512×281 mm CT volume and 0.5 mm ray step-size. Our DRR algorithm is able to closely reproduce actual CBCT projections taken through a test object. It is optimized specifically for the FBCT and CBCT systems in our department. Funded in part by NCI grants R01CA123299 and P01CA116602. The authors report no conflicts of interest.

SU-E-J-112: A Splatting Method to Generate DRRs for Deformed CT Volume.

To develop an efficient algorithm for generating high-quality digitally reconstructed radiographs (DRRs) for regularly and irregularly sampled volumes based on a splatting method with dynamic elliptical Gaussian kernels, and to evaluate this method using ray tracing. The traditional ray tracing method, which takes every intersected voxel into account, produces high quality DRRs but is very time consuming. Additionally, it is not suitable for handling irregularly-sampled volumes since it always requires image re-sampling, which leads to inaccuracy. We present a splatting approach to compute the 'footprint function', facilitating efficient perspective projection of elliptical Gaussian kernels at very low cost. This reported framework allows dealing with both regularly and irregularly sampled volumes effectively and efficiently. An XCAT digital phantom was used to generate 3D chest volumes at different respiratory phases, and CT projections are generated using ray tracing and the splatting method, respectively. Normalized cross correlation (NCC) is applied to evaluate the DRR similarity of two methods. Respiratory Phases one and four are used as volume datasets from the 4D XCAT digital phantom as they represent the lung at end inhale and end exhale, respectively. The dataset of Phase one is the basic regularly-sampled volume while the dataset of Phase four is deformed from that of Phase one, and resampled to a regularly sampled dataset for the ray tracing method. NCC between splatting and ray tracing DRRs are 0.9980 and 0.9977 for Phase one and Phase four, respectively. The calculation speed of the splatting method is 3 times faster than that of ray tracing. Our splatting approach can generate high-quality DRRs efficiently, and is a good alternative for current DRR generation techniques for deformed volume datasets. This research is supported by CPRIT Individual Investigator Award RP110329.

SU‐E‐J‐52: Validation of 3D Structure Projection Onto 2D DRR in Commercial Treatment Planning Systems

The use of structure overlay on setup DRRs can aid the image alignment procedure for daily image-guided setup procedures. However, the accuracy of a 3D region-of-interest (ROI) projected on a 2D digitally reconstructed radiograph (DRR) has rarely been evaluated quantitatively. The goal of this study is to test the accuracy of two commercial treatment planning systems (TPS) in producing overlay structures on setup DRRs. We designed a novel method to identify landmarks which were on the boundary of the projected ROI on a DRR. The 3D ROIvolume is composed of a stack of 2D curves. We first mathematically project each 2D curve onto a beams-eye-view (BEV) plane. Next, we detectthe boundary points of the projected curves. Those boundary points serve aslandmarks. Finally, we project the binary mask of the 3D ROI volume using ray tracing method onto the BEV plane. This projected binary mask is used to exclude the false landmarks. Once those landmarks are detected, wecompute the distance between the landmarks and ROI outlines from the TPS. We applied our validation method to 13 ROIs from a lung patient and 4 simulated ROIs on 2 BEV DRRs for two different TPS (Eclipse and Pinnacle). Average distance between the landmarks and ROIoutlines was 0.5mm for both Eclipse and Pinnacle approaches, which is close to the pixel resolution of the DRR. The maximum distance andaverage maximum distance was 2mm and 1 mm, respectively, for both TPS.The maximum distance occurred at points where the ROI curve has a sharpchange between slices. The accuracy of Eclipse and Pinnacle ROI projection method seems to be acceptable to within 1mm althoughprojection error can be as large as 2mm when structure shape has a sharp variation from one slice to the next.

SU-E-J-210: Lung Tumor Target Volume Contours on EPID Cine Mode Images.

While real time imaging of treatment through an electronic portal imaging device (EPID) is a powerful tool to monitor treatment, limited field of view and lower contrast from an MV beam can make assessment difficult for physicians. This work will develops a method to register and project contour outlines for the internal target volume (ITV) and planning target volume (PTV) of lung tumor cases onto cine mode EPID images to help physicians in interpretation during treatment. A sequence of EPID images, acquired during treatment, was registered to treatment planning computed tomography (CT) by machine geometry and patient setup with cone-beam computed tomography (CBCT). The planning CT was converted from Hounsfield scale to electron density by calibration curves of our CT simulator and digitally reconstructed radiographs (DRRs) were produced to match the EPID geometry, pixel for pixel. ITV and PTV structures as defined on the planning CT were then projected onto the DRRs. The DRRs were registered to the EPID images using cross correlation of a single template defined within the treatment aperture of each image. Once registered, the contours from the DRR were transferred to the EPID. We were able to successfully register MV DRRs to EPID images and display the projected target volumes. Without introduced motion, geometric registration and CBCT guided patient setup up were sufficient to register the contours within a single pixel, as normalized cross correlations produced no additional shift. We expect the DRR/EPID registration to be an important step when looking at cases with substantial tumor movement. The visualization of target volumes provides a tool for physicians to interpret EPID images and assess treatment, especially in cases with tumor movement. The methods developed will serve as the basis for a clinical tool providing real time contours.

Development of digital reconstructed radiography software at new treatment facility for carbon-ion beam scanning of National Institute of Radiological Sciences

To increase the accuracy of carbon ion beam scanning therapy, we have developed a graphical user interface-based digitally-reconstructed radiograph (DRR) software system for use in routine clinical practice at our center. The DRR software is used in particular scenarios in the new treatment facility to achieve the same level of geometrical accuracy at the treatment as at the imaging session. DRR calculation is implemented simply as the summation of CT image voxel values along the X-ray projection ray. Since we implemented graphics processing unit-based computation, the DRR images are calculated with a speed sufficient for the particular clinical practice requirements. Since high spatial resolution flat panel detector (FPD) images should be registered to the reference DRR images in patient setup process in any scenarios, the DRR images also needs higher spatial resolution close to that of FPD images. To overcome the limitation of the CT spatial resolution imposed by the CT voxel size, we applied image processing to improve the calculated DRR spatial resolution. The DRR software introduced here enabled patient positioning with sufficient accuracy for the implementation of carbon-ion beam scanning therapy at our center.

SU-E-J-26: Automated Estimation Method of Patient Setup Errors Using Simulated Portal Images for Prostate Cancer Radiotherapy.

We developed a novel automated estimation method for patient setup errors based on simulated and real portal images for prostate cancer radiotherapy. The estimation of patient setup errors in this study was based on a template matching technique with a cross-correlation coefficient and Sobel filter between the real portal image and localized pelvic template of reference image, which were DRR (digitally reconstructed radiography) images and simulated portal images. The simulated portal image was derived by projecting a CT image according to an inverse exponential power law of x-ray attenuation for a water-equivalent path length of each voxel of the CT image on each ray from a source to each pixel on the EPID (electric portal imaging device). A localized pelvic template of each patient in AP (anterior-posterior) or lateral view was automatically extracted from the DRR or simulated portal images by cropping a rectangular region, which was determined by using the mean pelvic template and four anatomical feature points. We applied the proposed method to three prostate cancer cases, and evaluated it using the residual error between the patient setup error obtained by proposed method and the gold standard setup error determined by two radiation oncologists. The average residual errors of the patient setup error for the DRR and simulated portal images were 0.79 and 1.26 mm in the left-right (LR) direction, 3.17 and 2.05 mm in the superior-inferior (SI) direction, 1.69 and 5.82 mm in the anterior-posterior (AP) direction, 3.84 and 6.94 mm in Euclidean distance (ED), respectively. If we used the simulated portal image for LR and SI directions and the DRR image for AP direction, the Euclidean distance was 3.22 mm. The proposed method has a potential to correctly estimate patient setup errors for prostate cancer radiotherapy.

TU-C-213CD-04: Tracking Implanted Fiducials Using Kilovoltage (kV) Projection Images: A Feasibility Study

Purpose: We have developed a method of tracking irregularly shaped implanted markers using KV projection images acquired in rotational mode and assess its potential for detecting intra‐fractional target motion. This is a feasibility study directed toward long‐range goals of acquiring such images during rotational treatment and using them for motion correction. Methods: KV projection images were acquired (Varian TrueBeam) during seven cone beam scans of two gastroesophageal and two pancreas cancer patients (IRB‐approved protocol). Each had at least one irregularly shaped radiopaque marker (Visicoil) implanted in or near the tumor. Specialized digitally reconstructed radiographs (DRRs) used for template based tracking were created from a breath‐hold planning CT at end expiration, in which the ray tracing was confined to a small volume of interest surrounding each marker. Sobel filter preprocessing of KV images served to enhance marker visibility and suppress background features. DRRs were matched with processed KV images both manually (ground truth) and automatically (normalized cross‐correlation with simplex minimization). Anthropomorphic phantom studies were also done to evaluate measurement uncertainty.Results: The mean (over patient scans) and standard deviation of the differences (Auto‐manual) were −0.04 ± 0.68 mm and 0.08 ± 0.89 mm in transverse and superior‐inferior (SI) directions respectively. The percentages of matches with difference exceeding 2 mm were 1.8% transverse and 5.0% SI. Intra‐observer consistency of manual registration was checked by repeating the manual registration for all 657 projections in one patient; the standard deviation of the difference was 0.4 mm. Phantom studies showed the measurement uncertainty of automatic registration to be approximately 0.15 mm. Conclusions: The proposed method can track arbitrary marker shapes using templates generated from a breath‐hold CT or alternatively, respiration‐correlated CT scan at one phase. Preliminary results indicate accuracy and robustness are adequate for clinical application but confirmation in larger numbers of patients is required. Research grant from Varian Medical Systems