Limits to dose reduction from iterative reconstruction and the effect of through-slice blurring
S. S. Hsieh, N. J. Pelc
SPIE Medical Imaging 2016: Physics of Medical Imaging, vol. 9783, pp. 97831C, 2016
Abstract
Iterative reconstruction methods have become very popular and show the potential to reduce dose. We present a limit to the maximum dose reduction possible with new reconstruction algorithms obtained by analyzing the information content of the raw data, assuming the reconstruction algorithm does not have a priori knowledge about the object or correlations between pixels. This limit applies to the task of estimating the density of a lesion embedded in a known background object, where the shape of the lesion is known but its density is not. Under these conditions, the density of the lesion can be estimated directly from the raw data in an optimal manner. This optimal estimate will meet or outperform the performance of any reconstruction method operating on the raw data, under the condition that the reconstruction method does not introduce a priori information. The raw data bound can be compared to the lesion density estimate from FBP in order to produce a limit on the dose reduction possible from new reconstruction algorithms. The possible dose reduction from iterative reconstruction varies with the object, but for a lesion embedded in the center of a water cylinder, it is less than 40%. Additionally, comparisons between iterative reconstruction and filtered backprojection are sometimes confounded by the effect of through-slice blurring in the iterative reconstruction. We analyzed the magnitude of the variance reduction brought about by through-slice blurring on scanners from two different vendors and found it to range between 11% and 48%.
Striped ratio grids for scatter estimation
S. S. Hsieh, A. S., Wang, J. Star-Lack
SPIE Medical Imaging 2016: Physics of Medical Imaging, vol. 9783, pp. 97830J, 2016
Abstract
Striped ratio grids are a new concept for scatter management in cone-beam CT. These grids are a modification of conventional anti-scatter grids and consist of stripes which alternate between high grid ratio and low grid ratio. Such a grid is related to existing hardware concepts for scatter estimation such as blocker-based methods or primary modulation, but rather than modulating the primary, the striped ratio grid modulates the scatter. The transitions between adjacent stripes can be used to estimate and subtract the remaining scatter. However, these transitions could be contaminated by variation in the primary radiation. We describe a simple nonlinear image processing algorithm to estimate scatter, and proceed to validate the striped ratio grid on experimental data of a pelvic phantom. The striped ratio grid is emulated by combining data from two scans with different grids. Preliminary results are encouraging and show a significant reduction of scatter artifact.
“Conventional” CT images from spectral measurements
P. L. Rajbhandary, N. J. Pelc
SPIE Medical Imaging 2016: Physics of Medical Imaging, vol. 9783, pp. 97831Q, 2016
Abstract
Spectral imaging systems need to be able to produce "conventional" images, and it's been shown that systems with energy discriminating detectors can achieve higher CNR than conventional systems by optimal weighting. Combining measured data in energy bins (EBs) and also combining basis material images have previously been proposed, but there are no studies systematically comparing the two methods. In this paper, we analytically evaluate the two methods for systems with ideal photon counting detectors using CNR and beam hardening (BH) artifact as metrics. For a 120-kVp polychromatic simulations of a water phantom with low contrast inserts, the difference of the optimal CNR between the two methods for the studied phantom is within 2%. For a polychromatic spectrum, beam-hardening artifacts are noticeable in EB weighted images (BH artifact of 3.8% for 8 EB and 6.9% for 2 EB), while weighted basis material images are free of such artifacts.
Raw data normalization for a multi source inverse geometry CT system
J. Baek, B. De Man, D. Harrison, N. J. Pelc
Optics Express, vol. 23, no. 6, pp. 7514-7526
Abstract
A multi-source inverse-geometry CT (MS-IGCT) system consists of a small 2D detector array and multiple x-ray sources. During data acquisition, each source is activated sequentially, and may have random source intensity fluctuations relative to their respective nominal intensity. While a conventional 3rd generation CT system uses a reference channel to monitor the source intensity fluctuation, the MS-IGCT system source illuminates a small portion of the entire field-of-view (FOV). Therefore, it is difficult for all sources to illuminate the reference channel and the projection data computed by standard normalization using flat field data of each source contains error and can cause significant artifacts. In this work, we present a raw data normalization algorithm to reduce the image artifacts caused by source intensity fluctuation. The proposed method was tested using computer simulations with a uniform water phantom and a Shepp-Logan phantom, and experimental data of an ice-filled PMMA phantom and a rabbit. The effect on image resolution and robustness of the noise were tested using MTF and standard deviation of the reconstructed noise image. With the intensity fluctuation and no correction, reconstructed images from simulation and experimental data show high frequency artifacts and ring artifacts which are removed effectively using the proposed method. It is also observed that the proposed method does not degrade the image resolution and is very robust to the presence of noise.
Fluid-filled dynamic bowtie filter: a feasibility study
P. Shunhavanich, S. S. Hsieh, N. J. Pelc
SPIE Medical Imaging 2015: Physics of Medical Imaging, vol. 9412, pp. 94121L, 2015
Abstract
By varying its thickness to compensate for the different path length through the patient as a function of fan angle, a pre-patient bowtie filter modulates flux distribution to reduce patient dose, scatter, and detector dynamic range, and to improve image quality. A dynamic bowtie filter is superior to its traditional, static counterpart in its ability to adjust its thickness along different fan and view angles to suit a specific patient and task. Among the proposed dynamic bowtie designs, the piecewise-linear and the digital beam attenuators offer more flexibility than conventional filters, but rely on analog positioning of a limited number of wedges. In this work, we introduce a new approach with digital control, called the fluid-filled dynamic bowtie filter. It is a two-dimensional array of small binary elements (channels filled or unfilled with attenuating liquid) in which the cumulative thickness along the x-ray path contributes to the bowtie’s total attenuation. Using simulated data from a pelvic scan, the performance is compared with the piecewise-linear attenuator. The fluid-filled design better matches the desired target attenuation profile and delivers a 4.2x reduction in dynamic range. The variance of the reconstruction (or noise map) can also be more homogeneous. In minimizing peak variance, the fluid-filled attenuator shows a 3% improvement. From the initial simulation results, the proposed design has more control over the flux distribution as a function of both fan and view angles.
Statistical bias in material decomposition in low photon statistics region
P. L. Rajbhandary, N. J. Pelc.
SPIE Medical Imaging 2015: Physics of Medical Imaging, vol. 9412, pp. 94124W, 2015
Abstract
We show that in material decomposition, statistical bias exists in the low photon regime due to non-linearity including but not limited to the log operation and polychromatic measurements. As new scan methods divide the total number of photons into an increasing number of measurements (e.g., energy bins, projection paths) and as developers seek to reduce radiation dose, the number of photons per measurement will decrease and estimators should be robust against bias at low photon counts. We study bias as a function of total flux and spectral spread, which provides insight when parameters like material thicknesses, number of energy bins, and number of projection views change. We find that the bias increases with lower photon counts, wide spectrum, with more number of energy bins and more projection views. Our simulation, with ideal photon counting detectors, show biases up to 2.4 % in basis material images. We propose a bias correction method in projection space that uses a multi dimensional look up table. With the correction, the relative bias in CT images is within 0.5 ± 0.17%.
First results from a prototype dynamic attenuator system
S. S. Hsieh, M. V. Peng, C. A. May, P. Shunhavanich, N. J. Pelc
SPIE Medical Imaging 2015: Physics of Medical Imaging, vol. 9412, pp. 94121N, 2015
Abstract
The dynamic, piecewise-linear attenuator has been proposed as a concept which can shape the radiation flux incident on the patient. By reducing the signal to photon-rich measurements and increasing the signal to photon-starved measurements, the piecewise-linear attenuator has been shown to improve dynamic range, scatter, and variance and dose metrics in simulation. The piecewise-linear nature of the proposed attenuator has been hypothesized to mitigate artifacts at transitions by eliminating jump discontinuities in attenuator thickness at these points. We report the results of a prototype implementation of this concept. The attenuator was constructed using rapid prototyping technologies and was affixed to a tabletop x-ray system. Images of several sections of an anthropormophic pediatric phantom were produced and compared to those of the same system with uniform illumination. The thickness of the illuminated slab was limited by beam collimation and an analytic water beam hardening correction was used for both systems. Initial results are encouraging and show improved image quality, reduced dose and low artifact levels.
Multivariate Gaussian model based Cramér-Rao lower bound evaluation of the in-depth PCXD
Y. Yao, N. J. Pelc
SPIE Medical Imaging 2015: Physics of Medical Imaging, vol. 9412, pp. 941213, 2015
Abstract
Purpose: In-Depth photon counting detectors (PCXD) use an edge-on configuration and have multi-layer segmentations. The benefit of this configuration for additional spectral information depends on the energy response. Also, inter-layer cross-talk introduces correlation to the signal collected from each layer, which makes the independent Poisson model no longer valid for estimating the Cramér-Rao lower bound (CRLB) of the material decomposition variance. We proposed to use a multivariate Gaussian model as the substitute address the data correlation. Methods: A 120 kVp incident spectrum was simulated and transmitted through 25cm of water and 1cm of calcium. 5- layer In-Depth and 1-layer Edge-On PCXDs with full energy resolution were simulated using Monte Carlo methods. We selected Si, GaAs and CdTe as detector materials. The detectors were defined to have 1mm wide pixels and thickness of 70mm (Si), 10.5mm (GaAs) and 3mm (CdTe). Geant4 was used and energy response functions (ERF) capturing secondary events were obtained, together with the Gaussian parameter estimates. We evaluated the CRLBs of the In-Depth and Edge-On detectors for each material and the systematic variance bounds were compared. Results: For uncorrelated data, the CRLB can assume Poisson statistics. As the data becomes more correlated, the Poisson CRLB fails to capture the cross-talk effect, but a Gaussian model can, and is accurate if the number of photons is not small. The CRLB analysis shows that the effects of the ERF and the noise correlation are significant. If cross-talk can be corrected, the depth information proves to be beneficial and can reduce the variance lower bound by 3% to 10% depending on the detector material. Conclusions: The multivariate Gaussian model was validated to be a good substitute to the Poisson model for PCXD CRLB estimation. It can avoid the errors that would otherwise be caused by correlated measurements.
A Dynamic Attenuator Improves Spectral Imaging With Energy-Discriminating, Photon Counting Detectors
S. S. Hsieh, N. J. Pelc
IEEE Transactions on Medical Imaging, vol. 34, no. 3, pp. 729-739, 2015
Abstract
Energy-discriminating, photon counting (EDPC) detectors have high potential in spectral imaging applications but exhibit degraded performance when the incident count rate approaches or exceeds the characteristic count rate of the detector. In order to reduce the requirements on the detector, we explore the strategy of modulating the X-ray flux field using a recently proposed dynamic, piecewise-linear attenuator. A previous paper studied this modulation for photon counting detectors but did not explore the impact on spectral applications. In this work, we modeled detection with a bipolar triangular pulse shape (Taguchi et al., 2011) and estimated the Cramer-Rao lower bound (CRLB) of the variance of material selective and equivalent monoenergetic images, assuming deterministic errors at high flux could be corrected. We compared different materials for the dynamic attenuator and found that rare earth elements, such as erbium, outperformed previously proposed materials such as iron in spectral imaging. The redistribution of flux reduces the variance or dose, consistent with previous studies on benefits with conventional detectors. Numerical simulations based on DICOM datasets were used to assess the impact of the dynamic attenuator for detectors with several different characteristic count rates. The dynamic attenuator reduced the peak incident count rate by a factor of 4 in the thorax and 44 in the pelvis, and a 10 Mcps/mm (2) EDPC detector with dynamic attenuator provided generally superior image quality to a 100 Mcps/mm (2) detector with reference bowtie filter for the same dose. The improvement is more pronounced in the material images.
An algorithm to estimate the object support in truncated images
S. S. Hsieh, B. E. Nett, G. Cao, N. J. Pelc
Medical Physics, vol. 41, no. 7, pp. 071908, 2014
Abstract
Purpose: Truncation artifacts in CT occur if the object to be imaged extends past the scanner field of view (SFOV). These artifacts impede diagnosis and could possibly introduce errors in dose plans for radiation therapy. Several approaches exist for correcting truncation artifacts, but existing correction algorithms do not accurately recover the skin line (or support) of the patient, which is important in some dose planning methods. The purpose of this paper was to develop an iterative algorithm that recovers the support of the object.
Methods: The authors assume that the truncated portion of the image is made up of soft tissue of uniform CT number and attempt to find a shape consistent with the measured data. Each known measurement in the sinogram is interpreted as an estimate of missing mass along a line. An initial estimate of the object support is generated by thresholding a reconstruction made using a previous truncation artifact correction algorithm (e.g., water cylinder extrapolation). This object support is iteratively deformed to reduce the inconsistency with the measured data. The missing data are estimated using this object support to complete the dataset. The method was tested on simulated and experimentally truncated CT data.
Results: The proposed algorithm produces a better defined skin line than water cylinder extrapolation. On the experimental data, the RMS error of the skin line is reduced by about 60%. For moderately truncated images, some soft tissue contrast is retained near the SFOV. As the extent of truncation increases, the soft tissue contrast outside the SFOV becomes unusable although the skin line remains clearly defined, and in reformatted images it varies smoothly from slice to slice as expected.
Conclusions: The support recovery algorithm provides a more accurate estimate of the patient outline than thresholded, basic water cylinder extrapolation, and may be preferred in some radiation therapy applications.
The piecewise-linear dynamic attenuator reduces the impact of count rate loss with photon-counting detectors
S. S. Hsieh, N. J. Pelc
Physics in Medicine & Biology, vol. 59, no. 11, pp. 2829-2847, 2014
Abstract
Photon counting x-ray detectors (PCXDs) offer several advantages compared to standard energy-integrating x-ray detectors, but also face significant challenges. One key challenge is the high count rates required in CT. At high count rates, PCXDs exhibit count rate loss and show reduced detective quantum efficiency in signal-rich (or high flux) measurements. In order to reduce count rate requirements, a dynamic beam-shaping filter can be used to redistribute flux incident on the patient. We study the piecewise-linear attenuator in conjunction with PCXDs without energy discrimination capabilities. We examined three detector models: the classic nonparalyzable and paralyzable detector models, and a 'hybrid' detector model which is a weighted average of the two which approximates an existing, real detector (Taguchi et al 2011 Med. Phys. 38 1089-102 ). We derive analytic expressions for the variance of the CT measurements for these detectors. These expressions are used with raw data estimated from DICOM image files of an abdomen and a thorax to estimate variance in reconstructed images for both the dynamic attenuator and a static beam-shaping ('bowtie') filter. By redistributing flux, the dynamic attenuator reduces dose by 40% without increasing peak variance for the ideal detector. For non-ideal PCXDs, the impact of count rate loss is also reduced. The nonparalyzable detector shows little impact from count rate loss, but with the paralyzable model, count rate loss leads to noise streaks that can be controlled with the dynamic attenuator. With the hybrid model, the characteristic count rates required before noise streaks dominate the reconstruction are reduced by a factor of 2 to 3. We conclude that the piecewise-linear attenuator can reduce the count rate requirements of the PCXD in addition to improving dose efficiency. The magnitude of this reduction depends on the detector, with paralyzable detectors showing much greater benefit than nonparalyzable detectors.