Mingshan Sun

Scatter correction method for x-ray CT using primary modulation: Phantom studies

PURPOSE Scatter correction is a major challenge in x-ray imaging using large area detectors. Recently, the authors proposed a promising scatter correction method for x-ray computed tomography (CT) using primary modulation. Proof of concept was previously illustrated by Monte Carlo simulations and physical experiments on a small phantom with a simple geometry. In this work, the authors provide a quantitative evaluation of the primary modulation technique and demonstrate its performance in applications where scatter correction is more challenging. METHODS The authors first analyze the potential errors of the estimated scatter in the primary modulation method. On two tabletop CT systems, the method is investigated using three phantoms: A Catphan 600 phantom, an anthropomorphic chest phantom, and the Catphan 600 phantom with two annuli. Two different primary modulators are also designed to show the impact of the modulator parameters on the scatter correction efficiency. The first is an aluminum modulator with a weak modulation and a low modulation frequency, and the second is a copper modulator with a strong modulation and a high modulation frequency. RESULTS On the Catphan 600 phantom in the first study, the method reduces the error of the CT number in the selected regions of interest (ROIs) from 371.4 to 21.9 Hounsfield units (HU); the contrast to noise ratio also increases from 10.9 to 19.2. On the anthropomorphic chest phantom in the second study, which represents a more difficult case due to the high scatter signals and object heterogeneity, the method reduces the error of the CT number from 327 to 19 HU in the selected ROIs and from 31.4% to 5.7% on the overall average. The third study is to investigate the impact of object size on the efficiency of our method. The scatter-to-primary ratio estimation error on the Catphan 600 phantom without any annulus (20 cm in diameter) is at the level of 0.04, it rises to 0.07 and 0.1 on the phantom with an elliptical annulus (30 cm in the minor axis and 38 cm in the major axis) and with a circular annulus (38 cm in diameter). CONCLUSIONS On the three phantom studies, good scatter correction performance of the proposed method has been demonstrated using both image comparisons and quantitative analysis. The theory and experiments demonstrate that a strong primary modulation that possesses a low transmission factor and a high modulation frequency is preferred for high scatter correction accuracy.

Efficient scatter correction using asymmetric kernels

X-ray cone-beam (CB) projection data often contain high amounts of scattered radiation, which must be properly modeled in order to produce accurate computed tomography (CT) reconstructions. A well known correction technique is the scatter kernel superposition (SKS) method that involves deconvolving projection data with kernels derived from pencil beam-generated scatter point-spread functions. The method has the advantages of being practical and computationally efficient but can suffer from inaccuracies. We show that the accuracy of the SKS algorithm can be significantly improved by replacing the symmetric kernels that traditionally have been used with nonstationary asymmetric kernels. We also show these kernels can be well approximated by combinations of stationary kernels thus allowing for efficient implementation of convolution via FFT. To test the new algorithm, Monte Carlo simulations and phantom experiments were performed using a table-top system with geometry and components matching those of the Varian On-Board Imager (OBI). The results show that asymmetric kernels produced substantially improved scatter estimates. For large objects with scatter-to-primary ratios up to 2.0, scatter profiles were estimated to within 10% of measured values. With all corrections applied, including beam hardening and lag, the resulting accuracies of the CBCT reconstructions were within ±25 Hounsfield Units (±2.5%).

Rapid Monte Carlo simulation of detector DQE(f)

PURPOSE Performance optimization of indirect x-ray detectors requires proper characterization of both ionizing (gamma) and optical photon transport in a heterogeneous medium. As the tool of choice for modeling detector physics, Monte Carlo methods have failed to gain traction as a design utility, due mostly to excessive simulation times and a lack of convenient simulation packages. The most important figure-of-merit in assessing detector performance is the detective quantum efficiency (DQE), for which most of the computational burden has traditionally been associated with the determination of the noise power spectrum (NPS) from an ensemble of flood images, each conventionally having 10(7) - 10(9) detected gamma photons. In this work, the authors show that the idealized conditions inherent in a numerical simulation allow for a dramatic reduction in the number of gamma and optical photons required to accurately predict the NPS. METHODS The authors derived an expression for the mean squared error (MSE) of a simulated NPS when computed using the International Electrotechnical Commission-recommended technique based on taking the 2D Fourier transform of flood images. It is shown that the MSE is inversely proportional to the number of flood images, and is independent of the input fluence provided that the input fluence is above a minimal value that avoids biasing the estimate. The authors then propose to further lower the input fluence so that each event creates a point-spread function rather than a flood field. The authors use this finding as the foundation for a novel algorithm in which the characteristic MTF(f), NPS(f), and DQE(f) curves are simultaneously generated from the results of a single run. The authors also investigate lowering the number of optical photons used in a scintillator simulation to further increase efficiency. Simulation results are compared with measurements performed on a Varian AS1000 portal imager, and with a previously published simulation performed using clinical fluence levels. RESULTS On the order of only 10-100 gamma photons per flood image were required to be detected to avoid biasing the NPS estimate. This allowed for a factor of 10(7) reduction in fluence compared to clinical levels with no loss of accuracy. An optimal signal-to-noise ratio (SNR) was achieved by increasing the number of flood images from a typical value of 100 up to 500, thereby illustrating the importance of flood image quantity over the number of gammas per flood. For the point-spread ensemble technique, an additional 2× reduction in the number of incident gammas was realized. As a result, when modeling gamma transport in a thick pixelated array, the simulation time was reduced from 2.5 × 10(6) CPU min if using clinical fluence levels to 3.1 CPU min if using optimized fluence levels while also producing a higher SNR. The AS1000 DQE(f) simulation entailing both optical and radiative transport matched experimental results to within 11%, and required 14.5 min to complete on a single CPU. CONCLUSIONS The authors demonstrate the feasibility of accurately modeling x-ray detector DQE(f) with completion times on the order of several minutes using a single CPU. Convenience of simulation can be achieved using GEANT4 which offers both gamma and optical photon transport capabilities.

Fast 4D cone-beam reconstruction using the McKinnon-Bates algorithm with truncation correction and nonlinear filtering

A challenge in using on-board cone beam computed tomography (CBCT) to image lung tumor motion prior to radiation therapy treatment is acquiring and reconstructing high quality 4D images in a sufficiently short time for practical use. For the 1 minute rotation times typical of Linacs, severe view aliasing artifacts, including streaks, are created if a conventional phase-correlated FDK reconstruction is performed. The McKinnon-Bates (MKB) algorithm provides an efficient means of reducing streaks from static tissue but can suffer from low SNR and other artifacts due to data truncation and noise. We have added truncation correction and bilateral nonlinear filtering to the MKB algorithm to reduce streaking and improve image quality. The modified MKB algorithm was implemented on a graphical processing unit (GPU) to maximize efficiency. Results show that a nearly 4x improvement in SNR is obtained compared to the conventional FDK phase-correlated reconstruction and that high quality 4D images with 0.4 second temporal resolution and 1 mm3 isotropic spatial resolution can be reconstructed in less than 20 seconds after data acquisition completes.

Rapid Scatter Estimation for CBCT using the Boltzmann Transport Equation

Scatter in cone-beam computed tomography (CBCT) is a significant problem that degrades image contrast, uniformity and CT number accuracy. One means of estimating and correcting for detected scatter is through an iterative deconvolution process known as scatter kernel superposition (SKS). While the SKS approach is efficient, clinically significant errors on the order 2-4% (20-40 HU) still remain. We have previously shown that the kernel method can be improved by perturbing the kernel parameters based on reference data provided by limited Monte Carlo simulations of a first-pass reconstruction. In this work, we replace the Monte Carlo modeling with a deterministic Boltzmann solver (AcurosCTS) to generate the reference scatter data in a dramatically reduced time. In addition, the algorithm is improved so that instead of adjusting kernel parameters, we directly perturb the SKS scatter estimates. Studies were conducted on simulated data and on a large pelvis phantom scanned on a tabletop system. The new method reduced average reconstruction errors (relative to a reference scan) from 2.5% to 1.8%, and significantly improved visualization of low contrast objects. In total, 24 projections were simulated with an AcurosCTS execution time of 22 sec/projection using an 8-core computer. We have ported AcurosCTS to the GPU, and current run-times are approximately 4 sec/projection using two GPU’s running in parallel.

Scatter correction with kernel perturbation

X-ray scatter degrades image contrast, uniformity and CT number accuracy in cone-beam computed tomography (CBCT). Correction methods based on the scatter kernel superposition (SKS) technique are efficient and suitable for many clinical applications but still produce residual errors due to limitations in the scatter kernel models. To reduce these errors, we propose to generate a first-pass reconstruction using a set of default SKS parameters followed by limited Monte Carlo simulations that are then used to perturb and refine key kernel parameters in order to obtain an improved second-pass correction. To test the approach, we used the fast adaptive scatter kernel model (fASKS) employing asymmetric kernels for the first-pass scatter correction and then used GEANT4 to simulate scatter-to-primary ratios in selected projections allowing for refined scatter estimates. The results show that a minimal number of projections require simulation in order to adequately perturb scatter kernel parameters for all projections. Compared to the default asymmetric kernels, the refined kernels reduced CT number errors from 24 HU to 15 HU in a large pelvis phantom resulting in a more uniform and accurate image.

User-friendly, ultra-fast simulation of detector DQE(f)

Development of the indirect scintillating detector is hindered not only by the cost and lead-time of manufacturing but also the computational resources required for numerical modeling. The simulation is bogged down by the number of x-ray photons (gammas) required to duplicate the experimental flood image ensemble necessary to characterize the noise power spectrum (NPS), a key input into the detective quantum efficiency (DQE). The simulation approach presented in this work exploits our previously reported procedure named Fujita-Lubberts- Swank (FLS)6 . This novel technique computes the Lubberts NPS from an ensemble of single gamma point spread functions (PSF) and, as a result, allows for a significant reduction in the number of simulated particles, enabling full DQE(f) simulations with optical transport in less than one CPU-hour. For a given detector and spectrum, the FLS execution time is determined primarily by the number of gamma and optical photons initiated. The optimal number of each varies with the detector specifics. In this work, we present a different simulation paradigm in which Geant4 was customized to allow for the user to specify the quantities of detected gammas, and detected opticals per gamma. These quantities were empirically shown to be constant over a small selection of different detector types. While work still needs to be done to explore the range of detectors for which this technique will work, we demonstrate a concept which brings added convenience and efficiency to FLS detector simulations.

WE‐C‐217BCD‐11: Coupled Radiative and Optical Geant4 Simulation of MV EPIDs Based on Thick Pixelated Scintillating Crystals

PURPOSE One way to greatly reduce the incidence of metal artifacts produced in kilovoltage (kV) CT images is by using megavoltage (MV) photons that penetrate high-Z objects, thus providing a measurable signal. For do se-efficient imaging, a high detective quantum efficiency (DQE) MV detector is desired. This study validates the coupled radiation and optical Geant4 simulation results against experimental data from various prototype pixelated scintillator MV detectors and determines the essential optical parameters which control the detector performance. METHODS Experimental data obtained with a 6MV radiation source from 8 different detectors was considered. The detectors used CsI, CdW and BGO as scintillating crystals and polystyrene septal wall material. Accurate Geant4 models of the detectors were implemented and coupled radiation and optical simulations were performed. The unknown optical properties of the models were determined by minimizing the difference between the modulation transfer functions (MTF) of the simulated data obtained with the slanted slit technique and the experimental MTFs. With the set of optical properties fixed, further simulation validation was performed against the experimental normalized noise power spectrum (NNPS(f)) and the experimental DQE(f) curves for each detector. All the simulations were performed on a computer cluster deployed on the Amazon EC2 platform. RESULTS The optimal values for the free optical parameters are 10%, 95% and 90% for the top surface reflectivity, the crystal-sept a surface reflectivity, and the Lambertian component contribution to the reflected beam from the crystal-septa interface respectively. The absolute difference between experimental and simulated data was below 10% for all the data sets. CONCLUSIONS To our knowledge this study is the first to present a full optical and radiative DQE(f) model using Geant4 that shows an excellent match with experimental data. The model indicates that improved performance can be obtained using more specular septa which are optically opaque. Support: NIH-T32-CA09695, NIH-1R01CA138426 NIH T32-CA09695, NIH R01- CA138426, Several authors work for Varian Medical Systems.

WE-C-217BCD-08: Rapid Monte Carlo Simulations of DQE(f) of Scintillator-Based Detectors

PURPOSE Monte Carlo simulations of DQE(f) can greatly aid in the design of scintillator-based detectors by helping optimize key parameters including scintillator material and thickness, pixel size, surface finish, and septa reflectivity. However, the additional optical transport significantly increases simulation times, necessitating a large number of parallel processors to adequately explore the parameter space. To address this limitation, we have optimized the DQE(f) algorithm, reducing simulation times per design iteration to 10 minutes on a single CPU. METHODS DQE(f) is proportional to the ratio, MTF(f)̂2 /NPS(f). The LSF-MTF simulation uses a slanted line source and is rapidly performed with relatively few gammas launched. However, the conventional NPS simulation for standard radiation exposure levels requires the acquisition of multiple flood fields (nRun), each requiring billions of input gamma photons (nGamma), many of which will scintillate, thereby producing thousands of optical photons (nOpt) per deposited MeV. The resulting execution time is proportional to the product nRun x nGamma x nOpt. In this investigation, we revisit the theoretical derivation of DQE(f), and reveal significant computation time savings through the optimization of nRun, nGamma, and nOpt. Using GEANT4, we determine optimal values for these three variables for a GOS scintillator-amorphous silicon portal imager. Both isotropic and Mie optical scattering processes were modeled. Simulation results were validated against the literature. RESULTS We found that, depending on the radiative and optical attenuation properties of the scintillator, the NPS can be accurately computed using values for nGamma below 1000, and values for nOpt below 500/MeV. nRun should remain above 200. Using these parameters, typical computation times for a complete NPS ranged from 2-10 minutes on a single CPU. CONCLUSIONS The number of launched particles and corresponding execution times for a DQE simulation can be dramatically reduced allowing for accurate computation with modest computer hardware. NIHRO1 CA138426. Several authors work for Varian Medical Systems.

TH-A-141-10: A Piecewise-Focused Pixelated Detector for MV Imaging

PURPOSE For portal imaging, high DQE detectors can be constructed from thick pixelated scintillator arrays that absorb MV x-rays. However, due to beam divergence, MTF and DQE losses can be significant for off-axis elements not focused towards the source. We present a novel focusing approach based on situating a shaped fiber optic plate (FOP) between rectilinear scintillator arrays and an amorphous silicon flat panel imager (AMFPI). METHODS The entire FOP comprises seven wedge-shaped sections that are fused together so that the center of each section points towards the source focal spot which is located 1500 mm away. The arc-shaped FOP directs light from the scintillator assembly to the AMFPI. The scintillator assembly consists of seven identical rectilinear sub-arrays, each with a 1 degree bevel, that are close-packed end-to-end. Each 15mm thick CdWO4 sub-array comprises 66×66 elements with a pixel pitch of 0.784 mm resulting in a piecewise-focused area detector having dimensions 365 mm × 52 mm. RESULTS Monte Carlo simulations of radiative and optical transport with a 6MV source predict a DQE(0) of 23%. With no beam divergence, MTF and DQE values at a spatial frequency of 0.4mm-1 are 0.5 and 15% respectively. Maximum off-axis MTF and DQE losses occur at the subarray edges (divergence angle =1 degree) and are only 6% and 12% respectively at 0.4mm-1 . If the detector were not focused, MTF and DQE losses at 0.4mm-1 would be 30% and 85% respectively. These losses would occur at edge of the detector where the beam divergence is 7 degrees. CONCLUSION A novel approach to creating a focused detector for MV portal, cone-beam or helical CT imaging is presented. Compared to previously proposed designs, all the block arrays are identical and rectilinear scintillator thus reducing costs and simplifying manufacturing processes. Assembly and experimental measurements are underway. NIH Academic-Industrial Partnership NIH RO1 CA138426; Varian Medical Systems.