D Pokhrel

Optimization of an adaptive neural network to predict breathing.

PURPOSE To determine the optimal configuration and performance of an adaptive feed forward neural network filter to predict breathing in respiratory motion compensation systems for external beam radiation therapy. METHOD AND MATERIALS A two-layer feed forward neural network was trained to predict future breathing amplitudes for 27 recorded breathing histories. The prediction intervals ranged from 100 to 500 ms. The optimal sampling frequency, number of input samples, training rate, and number of training epochs were determined for each breathing history and prediction interval. The overall optimal filter configuration was determined from this parameter survey, and its accuracy for each breathing example was compared to the individually optimal filter setups. Prediction accuracy was also compared to breathing stability as measured by the autocorrelation of the breathing signal. RESULTS The survey of filter configurations converged on a standard setup for all examples of breathing. For 24 of the 27 breathing histories the accuracy of the standard filter for a 300 ms prediction interval was within a few percent of the individually optimized filter setups; for the remaining three histories the standard filter was 5%-15% less accurate. CONCLUSIONS A standard adaptive neural network filter setup can provide approximately optimal breathing prediction for a wide range of breathing patterns. The filter accuracy has a clear correlation with the stability of breathing.

Bow-tie wobble artifact: effect of source assembly motion on cone-beam CT.

PURPOSE To investigate the cause of a bow-tie wobble artifact (BWA) discovered on Varian OBI CBCT images and to develop practical correction strategies. METHOD AND MATERIALS The dependence of the BWA on phantom geometry, phantom position, specific system, and reconstruction algorithm was investigated. Simulations were conducted to study the dependence of the BWA on scatter and beam hardening corrections. Geometric calibration was performed to rule out other gantry-angle dependent mechanical non-idealities as BWA causes. Air scans were acquired with ball-bearing markers to study the motions of the x-ray head assembly as functions of gantry angle. Based on measurements, we developed hypothesis regarding the BWA cause. Simulations were performed to validate our hypothesis. Two correction strategies were implemented: a measurement-based method, which acquires gantry-dependent normalization projections (NPs); and a model-based method that involves numerically shifting the single-angle NP to compensate for the previously-measured bow-tie-filter (BTF) motion. RESULTS The BWA has a diameter of approximately 15 cm, is centered at the isocenter, and is reproducible independent of phantom, position, system, reconstruction, and standard corrections, but only when the BTF is used. Measurements identified a 2D sinusoidal gantry-angle-dependent motion of the x-ray head assembly, and it was the BTF motion (>3 mm amplitude projected onto the detector) resulting an intensity mismatch between the all-angle CBCT projections and a single-angle NP that caused the BWA. Both correction strategies were demonstrated effective. CONCLUSIONS A geometric mismatch between the BTF modulation patterns on CBCT projections and on the NP causes the BWA. The BTF wobble requires additional degrees of freedom in CBCT geometric calibration to characterize.

Reconstruction of brachytherapy seed positions and orientations from cone-beam CT x-ray projections via a novel iterative forward projection matching method

PURPOSE To generalize and experimentally validate a novel algorithm for reconstructing the 3D pose (position and orientation) of implanted brachytherapy seeds from a set of a few measured 2D cone-beam CT (CBCT) x-ray projections. METHODS The iterative forward projection matching (IFPM) algorithm was generalized to reconstruct the 3D pose, as well as the centroid, of brachytherapy seeds from three to ten measured 2D projections. The gIFPM algorithm finds the set of seed poses that minimizes the sum-of-squared-difference of the pixel-by-pixel intensities between computed and measured autosegmented radiographic projections of the implant. Numerical simulations of clinically realistic brachytherapy seed configurations were performed to demonstrate the proof of principle. An in-house machined brachytherapy phantom, which supports precise specification of seed position and orientation at known values for simulated implant geometries, was used to experimentally validate this algorithm. The phantom was scanned on an ACUITY CBCT digital simulator over a full 660 sinogram projections. Three to ten x-ray images were selected from the full set of CBCT sinogram projections and postprocessed to create binary seed-only images. RESULTS In the numerical simulations, seed reconstruction position and orientation errors were approximately 0.6 mm and 5 degrees, respectively. The physical phantom measurements demonstrated an absolute positional accuracy of (0.78 +/- 0.57) mm or less. The theta and phi angle errors were found to be (5.7 +/- 4.9) degrees and (6.0 +/- 4.1) degrees, respectively, or less when using three projections; with six projections, results were slightly better. The mean registration error was better than 1 mm/6 degrees compared to the measured seed projections. Each test trial converged in 10-20 iterations with computation time of 12-18 min/iteration on a 1 GHz processor. CONCLUSIONS This work describes a novel, accurate, and completely automatic method for reconstructing seed orientations, as well as centroids, from a small number of radiographic projections, in support of intraoperative planning and adaptive replanning. Unlike standard back-projection methods, gIFPM avoids the need to match corresponding seed images on the projections. This algorithm also successfully reconstructs overlapping clustered and highly migrated seeds in the implant. The accuracy of better than 1 mm and 6 degrees demonstrates that gIFPM has the potential to support 2D Task Group 43 calculations in clinical practice.

TH-D-351-06: Comparison Between 2D Monte Carlo Modeled and Experimental Cone-Beam CT X-Ray Projections

Purpose: Fast and accurate modeling of cone‐beam CT(CBCT) x‐ray projection data can improve cone‐beam CT(CBCT)image quality either by conditioning projection data prior to image reconstruction or by supporting rigorous comparative simulation studies of competing image reconstruction and processing algorithms. In this study, we compare Monte Carlo‐ computed x‐ray projections with projections experimentally acquired from our Varian Trilogy CBCTimaging system for phantoms of known design. Method and Materials: Our recently developed Monte Carlo photon‐transport code, PTRAN, was used to compute primary and scatter projections for cylindrical phantoms of known diameter (CatPhan and NA model 76‐410) with and without bow‐tie filter and antiscatter grid for both full‐ and half‐fan geometries. The simulations were based upon measured 120 kVp spectra, beam profiles, and flat‐panel detector (4030CB) point‐spread functions. The beam‐stop array method was used to acquire scatter and SPR distributions from the OBI images. The biasing of scatter measurements due to the long detector PSF tails was corrected either by a lead mask or by deconvolution. Computed projections were compared to flat‐ and dark‐field corrected 4030CB images.Results: The simulated primary profiles agree with experiment within 3%, while the simulated scatter profiles agree within 8–10%. Both PSF measurements and mask measurements indicate that scatter radiation values can be biased by as much as 7% detector PSF tails. Conclusion: In agreement with the literature, the difference between simulated and measured projection data is of the order of 6–8%. Higher accuracy can be achieved mainly by improving the beam modeling and correcting the non linearities induced by the detector PSF. This project was supported in part by grants from Varian Medical Systems and NCI (R01 CA 75371).

Experimental validation of a Monte Carlo-based kV x-ray projection model for the Varian linac-mounted cone-beam CT imaging system

Fast and accurate modeling of cone-beam CT (CBCT) x-ray projection data can improve CBCT image quality either by linearizing projection data for each patient prior to image reconstruction (thereby mitigating detector blur/lag, spectral hardening, and scatter artifacts) or indirectly by supporting rigorous comparative simulation studies of competing image reconstruction and processing algorithms. In this study, we compare Monte Carlo-computed x-ray projections with projections experimentally acquired from our Varian Trilogy CBCT imaging system for phantoms of known design. Our recently developed Monte Carlo photon-transport code, PTRAN, was used to compute primary and scatter projections for cylindrical phantom of known diameter (NA model 76-410) with and without bow-tie filter and antiscatter grid for both full- and half-fan geometries. These simulations were based upon measured 120 kVp spectra, beam profiles, and flat-panel detector (4030CB) point-spread function. Compound Poisson- process noise was simulated based upon measured beam output. Computed projections were compared to flat- and dark-field corrected 4030CB images where scatter profiles were estimated by subtracting narrow axial-from full axial width 4030CB profiles. In agreement with the literature, the difference between simulated and measured projection data is of the order of 6-8%. The measurement of the scatter profiles is affected by the long tails of the detector PSF. Higher accuracy can be achieved mainly by improving the beam modeling and correcting the non linearities induced by the detector PSF.

TH‐D‐BRC‐06: The Investigation and Correction of a Bowtie‐Related Cone‐Beam CT Circular Band Artifact

Purpose: In the course of developing projection‐space preprocessing algorithms for improving on‐board CBCTCT number accuracy and uniformity, a persistent, prominent circular band artifact (CBA) with asymmetric illumination shadows was discovered. The CBA remain unchanged even after applying beam‐hardening, scatter subtraction, and veiling glare corrections to Varian OBI full‐fan projection data but only when a bow‐tie filter is used. This study investigates the causes and correction strategies of the CBA. Method and Materials:CBCTimages were acquired, preprocessed, and reconstructed with an in‐house FDK engine for phantoms of different diameters and locations relative to isocenter, on several OBI systems, to characterize CBA behavior and to form hypotheses as to its origin. Numerical simulations were used to evaluate all hypothesized contributing factors, as assessed by the necessary experimental measurements. A custom calibration was performed to identify the dependence of kV source location and flat‐panel detector pose as a function of gantry angle. Different correction approaches were tried on both synthetic and measured datasets, including gantry‐angle‐dependent normalization, full‐ and partial kV beam geometry calibrations, and empirical cancellation. The interplay between the CBA corrections and scatter and beam hardening corrections was also studied. Results: The CBA had a diameter of about 15 cm, was centered at the isocenter, and had similar asymmetric illumination, for all phantom dimensions, locations, and machines, and was reproducible over time. It appeared only when the bowtie filter was used. Simulations and experimental studies identified that a combination of geometric wobble and the bowtie filter slope caused the artifact. Gantry‐angle dependent calibrations of normalization were sufficient for about 80% CBA mitigation, but that complete elimination required gantry‐angle dependent beam‐hardening corrections. Conclusion:CBCT geometric wobble with the presence of bowtie filter could cause a circular band artifact. Supported by NIH P01 CA116602 and a grant from Varian Medical Systems.

Localizing intracavitary brachytherapy applicators from cone-beam CT x-ray projections via a novel iterative forward projection matching algorithm.

PURPOSE To present a novel method for reconstructing the 3D pose (position and orientation) of radio-opaque applicators of known but arbitrary shape from a small set of 2D x-ray projections in support of intraoperative brachytherapy planning. METHODS The generalized iterative forward projection matching (gIFPM) algorithm finds the six degree-of-freedom pose of an arbitrary rigid object by minimizing the sum-of-squared-intensity differences (SSQD) between the computed and experimentally acquired autosegmented projection of the objects. Starting with an initial estimate of the objects pose, gIFPM iteratively refines the pose parameters (3D position and three Euler angles) until the SSQD converges. The object, here specialized to a Fletcher-Weeks intracavitary brachytherapy (ICB) applicator, is represented by a fine mesh of discrete points derived from complex combinatorial geometric models of the actual applicators. Three pairs of computed and measured projection images with known imaging geometry are used. Projection images of an intrauterine tandem and colpostats were acquired from an ACUITY cone-beam CT digital simulator. An image postprocessing step was performed to create blurred binary applicators only images. To quantify gIFPM accuracy, the reconstructed 3D pose of the applicator model was forward projected and overlaid with the measured images and empirically calculated the nearest-neighbor applicator positional difference for each image pair. RESULTS In the numerical simulations, the tandem and colpostats positions (x,y,z) and orientations (alpha, beta, gamma) were estimated with accuracies of 0.6 mm and 2 degrees, respectively. For experimentally acquired images of actual applicators, the residual 2D registration error was less than 1.8 mm for each image pair, corresponding to about 1 mm positioning accuracy at isocenter, with a total computation time of less than 1.5 min on a 1 GHz processor. CONCLUSIONS This work describes a novel, accurate, fast, and completely automatic method to localize radio-opaque applicators of arbitrary shape from measured 2D x-ray projections. The results demonstrate approximately 1 mm accuracy while compared against the measured applicator projections. No lateral film is needed. By localizing the applicator internal structure as well as radioactive sources, the effect of intra-applicator and interapplicator attenuation can be included in the resultant dose calculations. Further validation tests using clinically acquired tandem and colpostats images will be performed for the accurate and robust applicator/sources localization in ICB patients.

TU-D-BRB-06: Localizing Intracavitary Brachytherapy Applicators from Conebeam CT X-Ray Projections Via a Novel Iterative Forward Projection Matching (IFPM) Algorithm

Purpose: To present a novel method for reconstructing the 3D pose (positions & orientations) of radio‐opaque objects of known but arbitrary shape from a small set of 2D x‐ray projections, in support of intraoperative brachytherapy planning. Methods and materials: IFPM finds the object pose by minimizing the sum‐of‐squared‐intensity‐differences (SSD) between the computed and experimentally‐acquired auto‐segmented object projections. Starting with an initial estimate of the applicator pose, IFPM iteratively refines the pose parameters (positions and 3 Euler angles) until the SSD converges. The applicator model is a mesh of discrete points derived from a complex combinatorial geometric model of the actual applicator. Three pairs of computed and measured projection images, with known imaging geometry, are used. Projection images of an intrauterine tandem were acquired from Acuity digital‐simulator. Image pre‐processing step was performed to create blurred binary applicator‐only images. To quantify IFPM accuracy, the reconstructed 3D pose of the applicator model was forward projected and overlaid with the measured images and the dice similarity coefficient (DSC) was computed for each image‐pair.Results: In the numerical simulations, the tandem and colpostats positions (x, y, z) and orientations (α, β, γ) were estimated with (0.28, 0.3, 0.37) mm and (0.9, 0.8, 1.0)° accuracies, respectively. For the measured tandem images, the DSC was better than 0.88 for each image‐pair.Discussion: We have developed a new, accurate, and completely automatic method to localize radio‐opaque applicators of arbitrary shape from x‐ray projections. Also, no lateral film is required. By localizing the applicator internal structure and the sources, the effect of intra/inter‐applicator attenuation can be included in the resultant dose distribution. Further development and validation tests with tandem and colpostats will be performed for the accurate and robust applicator/sources localization in ICB patients. Supported by Varian Medical Systems

TU-D-BRB-04: Reconstruction of Brachytherapy Seed Positions and Orientations from Conebeam CT X-Ray Projections: A Novel Iterative Forward Projection Matching Algorithm

Purpose: To generalize and experimentally validate a new algorithm for reconstructing the 3D pose (position and orientation) of implanted brachytherapy seeds from a few 2D conebeam‐CT x‐ray projections. Methods and materials: The iterative forward projection matching (IFPM) algorithm finds the set of seed poses that minimizes the sum‐of‐intensity‐difference‐squared (SSD) of computed and experimentally‐acquired auto‐segmented rojections of the seed array. IFPM starts with an initial approximation to the seed configuration, e.g., the pre‐planned seed arrangement and then iteratively refines the 3D seed pose and imaging viewpoint parameters until the SSD converges. We have demonstrated the IFPM method using both synthetic projection images of clinically‐realistic Model‐6711 seed arrangements and measured projections of an in‐house precision‐machined prostate implant phantom that allows the orientations and locations of up to 100 seeds to be set to known values. The phantom was scanned using an Acuity‐digital‐simulator with full 660‐projections. Three‐to‐ten x‐ray projection images were selected from the conebeam‐CT dataset and were pre‐processed to create binary seed‐only images. In addition to comparing the reconstructed to the known seed poses, 2D matching accuracy was quantified comparing the reconstructed seed projection with the measured projection using the dice‐similarity‐coefficient (DSC). The estimated 3D seed positions were also compared with clinically obtained VariSeed‐planning coordinates derived from conebeam‐CT images.Results: For the simulations, the seed reconstruction error was better than 0.4mm/2°. For the phantom experiments, IFPM absolute accuracy was (0.56±0.45)mm for position, while and (2.9±2.8)° and (3.6±4.0)° for polar and azimuthal angles, respectively. The DSC was better than 0.76 in each image‐pair.Conclusions: We have developed a novel algorithm for accurately recovering 3D pose of implanted brachytherapy seeds from as few as 3 projections. IFPM avoids the need to match corresponding seeds in each projection and accommodates incomplete data, overlapping seed clusters, and highly‐migrated seeds. Supported by Varian Medical Systems

Brachytherapy seed localization via iterative forward projection matching (IFPM) algorithm using intraoperative cone-beam-CT sinogram projections

Purpose: To experimentally validate a new algorithm for reconstructing the 3D positions of implanted brachytherapy seeds from intraoperatively acquired 2D cone-beam CT sinogram projections. Methods and materials: The iterative forward projection matching (IFPM) consists of finding the 3D seed geometry that minimizes the sum-of-squared intensity differences between computed projections of the candidate seed configuration and experimentally acquired auto-segmented projections of the implanted seeds. Projections are convolved by 2D Gaussian kernel to provide gradient, necessary for convergence. Four Pd103 post-implant patients were scanned using an Acuity digital-simulator with a full 660 projections conebeam-CT for post-implant dosimetry. 3-10 x-ray images were selected from the CBCT sinogram and were preprocessed to create the binary seed images. The IFPM accuracy was quantified both at projections and 3D. Results: For all example cases, the mean registration error was found to be about 1.2mm; the IFPM converged in 16-21 iterations and in 4-7 iterations following reduction of Gaussian blurring function width with computation time of about 1.9-2.3 minutes/ iteration on a 2 GHz processor. Discussion: The IFPM algorithm avoids the need for matching corresponding seeds images on each projection as required by standard back-projection methods and has a potential to accommodate incomplete projection data. The method has been found to work robustly with unknown number of seeds and uncertainties in the imaging viewpoints.