S736
International Journal of Radiation Oncology Biology Physics
3377
Materials/Methods: Daily images from five prostate patients with three implanted intraprostatic fiducial markers, acquired before setup for patients were off-line analyzed. Shifts in the anterior-posterior (AP), superior-inferior (SI), and left-right (LR) dimensions were calculated by image registration based on kV fiducial imaging and CBCT volumetric images. CBCT shifts were based on implanted fiducial markers and on soft tissues (ignoring fiducials). Based on these shift values, four strategies, designed to account for potential patient positioning errors when IGRT is not daily performed, were compared: “Shrinking Action Level” (9 mm, 4 images), “No Action Level” (4 images), “extended No Action Level” (4 images), “Newcastle Approach“ (5mm), as well as no correction as a benchmark. The PTV margins, defined by the van Herk’s formula, for every one of the above methods were used to compare the efficiency of these strategies. A total of 170 image registrations for 5 patients were analyzed. Results: Intra-observer reproducibility (standard deviation) of 2D kV imaging was found statistically smaller (p<0.01) than the ones of both types of CBCT-based registrations, at all 3 directions. Mean intra-observer difference between kV portals and fiducial -based cone beam CT was: 0.7 1.4 mm (LR) , 0.4 2.3 mm (AP) and 1.0 2.8 mm (SI). Mean difference between fiducial -based CBCT and soft tissue -based CBCT shifts was: 1.5 2.4mm (LR), 3.8 3.5 mm (AP) and 4.1 3.0 mm (SI). The inter-observer reproducibility was in mm: 0.8 (RL), 0.9 (AP), 0.8 (SI) for kV imaging, 2.3, 3.0, 2.8 for fiducial -based CBCT and 3.3, 4.1, 4.2 respectively for soft tissue -based CBCT. Among the correction strategies, the “extended No Action Level” strategy gave the smaller CTV-to PTV margin (6.4 mm), significantly smaller than that of the “Newcastle Approach” (8.2mm), the “Shrinking Action Level” (8.3mm) and the “No Action Level” (9.2 mm) strategies. Without any correction the margin needed for the group of 5 patients was 11.6 mm. Conclusions: Two-dimensional orthogonal kV imaging found to be the most reproducible method of IGRT for prostate cancer, both in inter- and intra-observer evaluation, followed by the fiducial -based CBCT. These two methods were found to agree well. Soft tissue based CBCT was found to be the less reproducible and in higher disagreement with the two methods, especially in AP and SI directions. Author Disclosure: D. Lazos: None. W.F. Mourad: None. D. Hauerstock: None. L. Harrison: None. E. Furhang: None. F. Trichter: None. R. Ennis: None.
A Novel Open Architecture Purpose Built Phased Coil Array for Head and Neck MR-SIM: Characterization, Protocol Optimization, and Imaging Performance Using Subjects Immobilized in the Treatment Position G. Perkins, R. Hammoud, S. Pienaar, S. Paloor, and N. Al Hammadi; Hamad Medical Corporation, DOHA, Qatar Purpose/Objective(s): To assess the geometric accuracy and imaging performance of a newly developed phased array coil system optimized for MR imaging of patients immobilized with the type S frame for head and neck radiation therapy. Materials/Methods: Protocol optimization for the novel coil was undertaken using a GE Optima450-GEM 70cm wide bore unit adapted for radiation therapy MR-SIM localization. Pulse sequence parameters for T1FSE, T2 FRFSE and T1 and T2 CUBE (including matrix, slice thickness, bandwidth) were optimized to maximize SNR while minimizing the effect of chemical shift. Phantom measurements were then performed to quantify system related residual geometric distortions after application of a 3D commercial gradient distortion correction algorithm for the effective FOV and the pulse sequences utilized. The optimized protocols were used to image seven volunteers positioned using an MR compatible type - S system with neutral head supports. MRI images (from the base of brain to below the clavicles) were acquired using the novel phased array coil combination placed over the head, neck and thoracic region. The volunteers were re-imaged without immobilization in a standard diagnostic head coil, which served as the benchmark for image quality. Signal to noise, image homogeneity and image artifacts were compared between the two image datasets in a blinded study. Visibility of ten anatomical structures was evaluated. For statistical analysis, Cohen’s kappa, Wilcoxon matched pairs and t-testing were utilized. Results: A geometrically accurate, high resolution MR-SIM imaging protocol (T1/T2CUBE. 2D T1/T2 FSE) was developed with 3mm slice thickness/no gap that provides excellent image quality from the base of brain to below the clavicles in the treatment position. From phantom studies, residual distortions were found to be 1.0mm within a 10 cm radius and < 1.5 mm within a 15 cm radius of the scan centre. Using a RBW of 50-62.5 minimized chemical shift (<1mm within a 10cm radius) at 1.5T while providing acceptable SNR. The coil provided higher SNR values in all anatomical structures, but less uniformity than the standard diagnostic head coil (full comparative SNR/CNR/geometric data will be presented). No significant difference for image quality and artifacts were demonstrated between the coils in the blinded study. Conclusions: The novel open architecture purpose built phased coil array provides unparalleled coverage and good image quality of the subject immobilized for head and neck radiation therapy and is a promising solution for MR-SIM in the head and neck region. Author Disclosure: G. Perkins: None. R. Hammoud: None. S. Pienaar: None. S. Paloor: None. N. Al Hammadi: None.
3378 Assessment of Fiducial-based 2D kV Orthogonal Imaging, Fiducialbased CBCT, and Soft-tissue-based CBCT for Prostate Cancer Patients With Implanted Fiducial Markers D. Lazos,1 W.F. Mourad,1,2 D. Hauerstock,1 L. Harrison,1 E. Furhang,1 F. Trichter,3 and R. Ennis3; 1Beth Israel Medical Center, Continuum Cancer Centers of New York, New York, NY, 2Department of Radiation Oncology, Albert Einstein College of Medicine of Yeshiva University, Bronx, NY, 3St Luke’s - Roosevelt Hospital, Continuum Cancer Centers of New York, New York, NY Purpose/Objective(s): To assess the intra- observer, inter- observer and inter-modality reproducibility of fiducial-based 2D orthogonal kV imaging, fiducial-based-cone beam CT (CBCT) and soft tissue-based CBCT image registration for daily localization of prostate in radiation therapy, as well as to compare four strategies for correction of potential patient positioning errors.
3379 Time Series Analysis of Interfraction Patient Setup in Image Guided Radiation Therapy A. Hu and S. Qi; UCLA, Los Angeles, CA Purpose/Objective(s): It is attempting to predict daily setup shifts or anatomic changes for adaptive radiation therapy. However, it is not clear whether the shifts are potentially predictable or completely random. Time series analysis, a widely used statistical method in other discipline such as stock prediction, has rarely been used in analysis of the shifts or internal organ motion. The purpose of this study was to explore the temporal characteristics of daily shifts through time series analysis in prostate and head and neck (H&N) cancer radiation treatment. Materials/Methods: Daily shifts of media-lateral (ML), cranial-caudal (CC) and posterior-anterior (AP) and roll from 31 H&N and 42 prostate cases were analyzed retrospectively. All patients were treated with IMRT using megavoltage cone beam or fan beam CT as imaging guidance for daily setup. We selected patients who were treated with at least 15 fractions. Time series analysis was performed for daily shifts in three translational dimensions and/or rotational roll on each patient respectively using time series analysis module. Autocorrelation, trend and pattern were examined via analysis of correlograms of autocorrelation function and partial autocorrelation function. Autoregressive-moving average (ARMA) modeling was used to construct model for shift prediction. Results: A total of 124 series from 31 H&N patients and 168 series from prostate patients were analyzed. Most of the series did not demonstrate trend or autocorrelation, i.e., daily shifts were random and did not appear
Volume 84 Number 3S Supplement 2012 to be correlated with previous days’ shifts. However, approximately 22% of H&N cases had autocorrelation in AP direction and 13% in other two directions. For prostate, approximately 29% of the AP shifts, 24% of the ML shifts, 14% of the CC shifts and 14% of the rotation were autocorrelated. Autocorrelation were normally found in one or two directions but rarely found in all directions for patients at both tumor sites. Patient characteristics such as initial body weight or weight loss did not seem to be associated with autocorrelation series. The shifts in 52% H&N and 60% prostate patients appeared to be random and demonstrated no autocorrelation in any of the shift directions. Autoregression (AR), moving average (MA) and ARMA modeling were then performed for autocorrelated series. First order model AR(1) was adequate for most autocorrelated series. Conclusions: Daily shifts for the majority of prostate and H&N patients were random in any directions. However, interfraction shifts of a subset of patients demonstrated autocorrelation and appeared to be predictable by previous days’ shifts. Although the factors to identify these subsets of patients are unclear in this initial analysis, we remain hopeful that these factors will be identified in future investigations. Author Disclosure: A. Hu: None. S. Qi: None.
3380 Analysis of Tumor-Motion Surrogate Signals Under Different Coaching Conditions Using Empirical Mode Decomposition and Hilbert-Huang Transformation S. Han-Oh,1 R. George,2 and R. Hales3; 1Johns Hopkins University, Baltimore, MD, 2University of Maryland School of Medicine, Baltimore, MD, 3Johns Hopkins University School of Medicine, Baltimore, MD Purpose/Objective(s): To investigate a novel technique, called empirical mode decomposition (EMD) and Hilbert Huang Transformation (HHT), by applying to tumor-motion surrogate signals under free (FB), audio-guided (AG), and audio-visual guided (AVG) coaching conditions to find whether the a patient-specific breathing frequency in FB is preserved under different coaching conditions. Materials/Methods: EMD decomposes a nonlinear and non-stationary signal into the intrinsic mode functions (IMFs) in the time domain which are a complete and orthogonal set to describe the signal. HHT is used to obtain instantaneous frequencies of each IMF as a function of time. The tumor-motion surrogate signals from 20 lung-cancer patients were acquired using an infrared reflector on the patients’ abdomen. Each patient participated in FB, AG, and AVG breathing sessions. These 3 types of breathing signals were decomposed into its own set of the IMFs by the EMD and then, the HHT was applied to each IMFs to obtain instantaneous frequency. The averaged instantaneous frequency over time was used to represent the corresponding IMF. Statistical significance in differences of frequencies among inter-patients and 3-coaching types were tested using ANOVA. Correlation of each IMF with the surrogate signal was used to determine the IMF representing the surrogate signal best. Results: Each surrogate signal was decomposed to 10 1 IMFs on average, regardless of the coaching types. The decomposed IMFs from each surrogate signal showed three categories of frequencies: (1) high frequencies (1 - 10 Hz) such as a noise-like signal, (2) medium frequencies (0.1 - 0.7 Hz), which is a true breathing signal, and (3) low frequencies (0.09 - 0.01 Hz) that is a baseline drift. The IMFs with high and low frequencies were independent of the patients and the coaching types (p Z 1 and 0.69 for high and low frequencies). It indicates that these high and low frequency components originated from the tumor-motion surrogate system. Each surrogate signal showed a prominent correlation with one of its IMFs: 0.747 for FB, 0.770 for AG, and 0.782 for AVG, on average. Low correlations were shown with the rest of the IMFs: 0.139 for FB, 0.125 for AG, and 0.133 for AVG, on average. The IMF with a high correlation for FB showed an average difference of 0.11 Hz and 0.09 Hz compared to that of AG and AVG, respectively. These differences were not statistically significant (p > 0.05). Conclusions: The EMD and HHT can produce a reproducible tumormotion surrogate signal after removing unnecessary signal components. The patient-specific breathing frequency is conserved regardless of
Poster Viewing Abstracts S737 respiratory coaching types. These data support a novel technique which can improve clinical procedures that rely on tumor-motion surrogates. Author Disclosure: S. Han-Oh: None. R. George: None. R. Hales: None.
3381 Fluoroscopic 3D Images Based on 2D Treatment Images Using a Realistic Modified XCAT Phantom J. Lewis,1 R. Li,2 S. St. James,1 Y. Yue,1 R. Berbeco,1 and P. Mishra1; 1 Brigham and Women’s Hospital, Dana-Farber Cancer Center, Harvard Medical School, Boston, MA, 2Stanford University, Palo Alto, CA Purpose/Objective(s): To simulate the process of generating fluoroscopic 3D treatment images from 4DCT and individual 2D x-ray projections using a realistic patient data-based modified XCAT phantom. Materials/Methods: The process consists of the following steps: 4DCT based on modified XCAT: We developed a modified XCAT digital phantom based on measured patient tumor trajectories. The respiratory motion in the current XCAT phantom depends upon chest wall and diaphragm motion. To generate an XCAT phantom incorporating the measured trajectories and realistic anatomic motion, we first modified the parameters of the existing phantom to adaptively calculate the chest wall and diaphragm motion based on a specific tumor motion. Patient-specific PCA lung motion model: To build a patient-specific lung motion model a set of displacement vector fields (DVFs) corresponding to different phases of 4DCT were generated. These DVFs are generated relative to a reference (typically peak-exhale) phase of the data. DVFs are created using a graphics processing unit-implemented Demon’s algorithm deformable image registration. To compactly represent these DVFs we use principal component analysis (PCA)-based eigenvector decomposition as it captures salient characteristics of the DVFs in 3 eigenvectors. In the PCA representation DVFs and PCA coefficients are dependent on time while eigenvectors are only space-dependent, thus enabling the generation of fluoroscopic 3D images from a small set of time-dependent parameters. Fluoroscopic 3D image generation: Once the patient-specific lung motion model is obtained the next step is to generate 3D images based on the projection image. This is achieved by defining a cost function in which a projection matrix connects the 3D reference image to the single projection image. The cost function iteratively updates the PCA coefficients and DVFs to approximate the given projection image in a meansquared sense. Results: We constructed a PCA lung motion model at a resolution of (2, 2, 2.5) mm in (x, y, z). For image generation we used 3 PCA coefficients. For evaluation we used 3D images directly generated from the XCAT phantom. The error metrics used were the average of the absolute distance each pixel in a reconstructed image moves compared to the ground truth image (Erecons) versus the same measure for the reference image (Eref). The mean of Erecons is 0.079 mm and of Eref is 0.1665 mm, with standard deviation 0.0060 and 0.0678 mm respectively. Conclusions: We developed a methodology to simulate the process of generating fluoroscopic 3D treatment images based on prior information in the 4DCT and single x-ray projections measured during treatment. This was accomplished by simulating each step in the process using a modified XCAT phantom with motion matched to measured patient tumor trajectories. Author Disclosure: J. Lewis: None. R. Li: None. S. St. James: None. Y. Yue: None. R. Berbeco: None. P. Mishra: None.
3382 Spatial Accuracy and Visibility Studies of a MRI Compatible Fiducial Marker P. Wersall,1 A. Hartwig,2 E. Castellanos,3 F. Labruto,4 and A. Tilikidis1; 1 Karolinska University Hospital, Stockholm, Sweden, 2Karolinska University Hospital, Clinical Neuroscience, Stockholm, Sweden, 3 Karolinska University Hospital, Department of Oncology, Stockholm, Sweden, 4Karolinska University Hospital, Department of Radiology, Stockholm, Sweden