Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2015 Jul 1.
Published in final edited form as: Phys Med Biol. 2014 Feb 20;59(5):1305–1326. doi: 10.1088/0031-9155/59/5/1305

Experimental realization of fluence field modulated CT using digital beam attenuation

TP Szczykutowicz 1, CA Mistretta 1,2,3
PMCID: PMC4487678  NIHMSID: NIHMS574464  PMID: 24556823

Abstract

Purpose

Tailoring CT scan acquisition parameters to individual patients is a topic of much research in the CT imaging community. It is now common place to find automatically adjusted tube current options for modern CT scanners. In addition, the use of beam shaping filters, commonly called bowtie filters, is available on most CT systems and allows for different body regions to receive different incident x-ray fluence distributions. However, no method currently exists which allows for the form of the incident x-ray fluence distribution to change as a function of view angle. This study represents the first experimental realization of fluence field modulated CT (FFMCT) for a c-arm geometry CT scan. Methods: X-ray fluence modulation is accomplished using a digital beam attenuator (DBA). The device is composed of 10 iron wedge pairs that modulate the thickness of iron x-rays must traverse before reaching a patient. Using this device, experimental data was taken using a Siemens Zeego c-arm scanner. Scans were performed on a cylindrical polyethylene phantom and on two different sections of an anthropomorphic phantom. The DBA was used to equalize the x-ray fluence striking the detector for each scan. Non DBA, or “flat field” scans were also acquired of the same phantom objects for comparison. In addition, a scan was performed in which the DBA was used to enable volume of interest (VOI) imaging. In VOI, only a small sub-volume within a patient receives full dose and the rest of the patient receives a much lower dose. Data corrections unique to using a piece-wise constant modulator were also developed.

Results

The feasibility of FFMCT implemented using a DBA device has been demonstrated. Initial results suggest dose reductions of up to 3.6 times relative to “flat field” CT. In addition to dose reduction, the DBA enables a large improvement in image noise uniformity and the ability to provide regionally enhanced signal to noise using VOI imaging techniques.

Conclusions

The results presented in this paper take the field of FFMCT from the theoretical stage to that of possible clinical implementation. FFMCT, as shown in this paper, can reduce patient dose while maintaining or improving image quality. In addition, the DBA has been experimentally shown to be well suited to implement entirely new imaging methods like photon counting and VOI imaging.

Keywords: fluence field modulated CT, volume of interest imaging, dynamic collimator

1. Introduction: a short history of FFM in CT imaging

Recently, there has been a lot interest within the CT imaging community concerning fluence field modulation (Kolditz et al. 2010, Heuscher & Noo 2011, Heuscher & Noo 2012, Bartolac et al. 2010, Bartolac et al. 2011, Bartolac & Jaffray 2012, Chen et al. 2011, Stiller et al. 2012, Sperl et al. 2010, Burion et al. 2011, Hsieh & Pelc 2013a, Hsieh & Pelc 2013b). The ability to modulate the incident x-ray fluence in CT imaging has been proposed to implement volume of interest imaging (VOI) (Bartolac et al. 2010, Bartolac et al. 2011, Bartolac & Jaffray 2012, Kolditz et al. 2010, Heuscher & Noo 2011, Heuscher & Noo 2012, Chen et al. 2011), improve standard CT imaging (Stiller et al. 2012, Sperl et al. 2010, Burion et al. 2011, Hsieh & Pelc 2013a, Hsieh & Pelc 2013b), and may be a necessary technology if photon counting CT is to overcome detector saturation and charge trapping problems (McCollough et al. 2012, Shikhaliev 2009).

CT has steadily been moving towards a more patient specific dose delivery from its inception. Tube current modulation (Kalender et al. 1999) and the use of bowtie filters (Graham et al. 2007, Toth et al. 2007) are two common methods used to reduce patient dose, detector dynamic range requirements, and x-ray scatter. The positive results of modulating the incident exposure as a function of the imaging object are well known in 2-d projection imaging as well and work was carried out in the 1980’s that lead to the development of a clinical product capable of dose modulation for different parts of a single radiograph (Plewes 1983, Plewes & Vogelstein 1983, Veldkamp et al. 2005, Molloi et al. 2001, Liu et al. 2011).

The work presented in this study explains the first experimental realization of a device capable of modulating the x-ray dose within the fan beam for non-inverse geometry CT (Szczykutowicz & Mistretta 2013a, Szczykutowicz & Mistretta 2013b, Szczykutowicz & Mistretta 2012, Szczykutowicz & Mistretta 2013c, Szczykutowicz & Mistretta 2013d). The device used to obtain such a modulation is referred to as digital beam attenuator (DBA). The DBA changes the form of the x-ray fluence incident to a patient as a function of fan angle by changing the thickness of a piecewise-constant attenuator as a function of fan angle. The change in attenuation is obtained by sliding pairs of wedge shaped filters over each other. As the amount of overlap between the wedges increases, the composite thickness of the wedges as seen by x-rays traveling from the focal spot to the patient also increases (Szczykutowicz & Mistretta 2013a). The device is shown in Figure 1.

Figure 1.

Figure 1

Open in a new tab

(a) Current experimental prototype set-up in a c-arm interventional suite. (b) Image of DBA wedges, linear actuators, and control circuits.

2. Methods

The method sub-sections will describe the DBA-FFMCT prototype system and two imaging applications the DBA can be used for. In the first application, the DBA will be used to provide a uniform detector signal, i.e. the DBA will mimic the function of a bowtie filter. In the second application, the DBA will be used to implement VOI imaging.

2.1. Data acquisition

A c-arm (Zeego, Siemens AG, Forchheim Germany) was used to acquire x-ray projections of a cylindrical 14.5 cm polyethylene phantom and an anthropomorphic phantom (Atom dosimetry verification phantom 706, CIRS Inc., Norfolk, Virginia). The cylindrical phantom was constructed at the University of Wisconsin Madison by the authors. The phantoms were connected to a rotary stage (Model M-495CC, Newport Corp., Evry France) controlled by a motor controller (Model ESP 300, Newport Corp., Irvine CA USA) which was fed serial commands via Matlab (Mathworks, Natick MA USA). C-arm CT acquisitions were simulated by keeping the c-arm fixed and rotating the phantom.

All scans were acquired at 125 kVp, using the full 40 cm of detector coverage in the fan beam direction and 2.8 cm (at 1 m from the x-ray source) FOV in the cone-angle direction. The 40×30 cm at panel on the Siemens Zeego scanner was read out at a detector pitch of 0.154 mm. No detector pixel averaging was performed in the “in-slice” direction, but some detector pixel averaging was performed in the z direction to mimic different dose level acquisition as described in Section 3.1.2. The DBA scan did not employ any Cu filtration, however, a 0.9 mm Cu filter was used for the non-DBA scans. The tube current and time were 20 mA and 30 ms for each non-DBA projection and 20 mA and 200 ms for the DBA projections. Acquisition times were roughly 2 hours. For each scan, 200 projections were acquired over 360 degrees. This relatively long time was due to the time it took to: (1) read in the raw detector data; (2) process the raw detector data using the software provided by Siemens; and (3) position the wedges and index the phantom (the time to calculate how to move the wedges was negligible compared to these other tasks). On a clinical realization of the DBA, all of these steps would be fully integrated into the CT system and the limiting factor in data acquisition time would most likely be the time to move the wedges. Raw data access was provided by Siemens. “Raw” refers to data with detector dead pixel, gain, and offset corrections but without any additional image post processing or conditioning (i.e. edge enhancement or beam hardening correction).

2.2. DBA prototype

The DBA prototype is pictured in Figure 1. Figure 2 identifies some of the DBA dimension parameters listed in Table 1. tt is the toe thickness of each wedge, th is the heel thickness of each wedge that moves (i.e. only one wedge in each wedge pair moves, the other is stationary and can be made smaller), tmin is the minimum wedge thickness (i.e. corresponding to the wedge position of minimum overlap), tmax is the maximum wedge thickness (i.e. corresponding to the wedge position of maximum overlap), lmo is the length of wedge irradiated at the plane of the DBA, l is the length of the moving DBA wedge, w is the wedge width, and dmax is the distance moved by the wedge in order to change from minimum to maximum thickness. A wedgelet is the volume defined by all x-rays traversing a given wedge pair. The DBA was positioned just outside the x-ray tube housing at a distance of 31.75 cm from the x-ray source. As shown in Szczykutowicz & Mistretta (2013a), DBA performance in terms of dynamic range and “relative” tube loading increases as the DBA is moved closer to the x-ray tube. Without removing any collimator components, 31.75 cm was as close as the DBA could get to the x-ray source. The DBA wedges were made from Fe as this material showed promise based on the results presented in Szczykutowicz & Mistretta (2013a). The DBA wedges rest on an acrylic baseplate and are coated in graphite lubricant to reduce friction between wedges. The current prototype design relies on gravity to keep the wedges in place, so additional engineering would have to be employed to make the design capable of being mounted on a rotating gantry.

Figure 2.

Figure 2

Open in a new tab

Depiction of how the composite wedge thickness changes as a function of wedge overlap. (Left) The wedge position of minimum overlap. (Right) The position of maximum overlap. The gray trapezoid shown over the wedges represents the area irradiated by the x-ray cone beam. Note: the images ae not drawn to the scale as listed in Table 1.

Table 1.

Dimensions for the prototype

Parameter value
th 1.5 cm
tt 0.13 mm
l 17 cm
lmo 3.9 cm
tmin 3.7 mma
tmax th + th
dmax 10 cm
w 1.75 cm
Open in a new tab
a

Assuming only half of the detector in the z-direction is being utilized. If the entire 29.6 cm of the detector is used, tmin = 7.1 mm. These thicknesses represent highly unoptimized configurations (Szczykutowicz & Mistretta 2013a). In a clinical implementation, the DBA would be positioned much closer to the x-ray source, making this thickness approximately 1 millimeter depending on how thin the wedge toe can be constructed (Szczykutowicz & Mistretta 2013a).

The DBA device and drive system was constructed in-house. For each wedge pair, one wedge was held fixed and one was free to move (Szczykutowicz & Mistretta 2013a). Ten 10 cm stroke length linear actuators (Model L-12P with 50:1 gearing and 100 mm stroke length, Firgelli Technologies Inc., Victoria BC Canada) moved the wedges. The linear actuators were coupled with control boards (Model LAC, Firgelli Technologies Inc., Victoria BC Canada) and controlled by a 12 channel USB motion controller (12 channel Mini Maestro, Pololu Robotics and Electronics, Las Vegas NV USA) which was sent serial commands via Matlab (Mathworks, Natick MA USA).

2.3. DBA positioning LUT

The detector signal as a function of wedge and tissue thickness in the beam must be determined before the DBA can be used such that the wedges can be intelligently positioned in order to obtain a given detector signal. Since the presence of varying amounts of DBA wedge and tissue will harden the beam differently (e.g. different amounts of wedge will be required to modulate the detector signal the same amount for different tissue thicknesses), a two dimensional look-up-table (LUT) was created to determine how to position a wedge to obtain a given detector signal for a given patient thickness.

A wedge positioning LUT was created by imaging different PMMA thicknesses. A total of 8 different PMMA thicknesses were imaged ranging from 0 to 17 cm. The 10 DBA wedges were positioned from the amount of least overlap to the most overlap and kept fixed for each PMMA thickness (i.e. the first wedge pair was set to the minimum thickness, the next wedge pair was set slightly thicker and so on until the 10th wedge pair was set to the maximum wedge thickness). The minimum and maximum wedge thicknesses corresponded to 3.7 and 16.3 mm respectively. The x-ray collimators were set to the smallest field of irradiation to minimize the effects of x-ray scatter, which corresponded to 2.8 cm at 1 m from the x-ray source on our system. Regions of interest (ROI) were placed on the PMMA thickness images corresponding to the different wedge thicknesses. The mean values from these ROIs were calculated. This resulted in a LUT table having 8 by 10 values (number of PMMA thicknesses by number of different wedge thicknessess). A two dimensional exponential fit was applied to this data and used to create the LUT. Figure 3 depicts the fitted LUT and the original data.

Figure 3.

Figure 3

Open in a new tab

LUT used to position wedges. The fitted LUT is rendered below the original data by 100 detector signal units for comparison.

The LUT was used in data acquisition to determine the wedge position required to achieve the desired detector signal as follows. For a given wedgelet, the current wedge thickness and detector value for that wedgelet were used to interpolate the equivalent thickness of PMMA for that wedgelet using the LUT. Once the equivalent PMMA thickness was determined, it was used to interpolate the required wedge thickness that will produce the desired detector value (i.e. if the desired detector value was already met, the wedge thickness will not change, if the detector value was too small/large the wedge thickness will decrease/increase). This was done for every wedgelet for every view angle. The error caused by assuming all phantom material is PMMA like and any error caused by the use of a single exponential to fit data with beam hardening present will result not result in much error in the ability of this wedge positioning algorithm to function due to the iterative nature of the wedge positioning. In other words, if the estimate for the wedge position is too low or too high becasue of an assumption made, the next view angle will be closer to the ideal wedge position value irregardless of slight inaccuracies in the LUT.

2.4. Image reconstruction

Image reconstruction was performed using a FBP algorithm with a ramp filter (Kak & Slaney 1988). In order to obtain the projection values required for image reconstruction, the detector signal values for each wedge configuration without any phantom object in place were determined (i.e. Io (the number of photons incident to an attenuator) must be determined for each I (the number of photons exiting the attenuator) measurement in order to calculate p (the line integral of linear attenuation coefficients through the attenuator) using the well know Beer’s law relationship given by I = Ioe−p). These types of projection images are commonly referred to as “air scan” images. For the DBA scans, this meant acquiring a number of images equal to the number of view angles since each view angle had a unique wedge configuration. For the non-DBA scans, 10 “air-scan” images were acquired and averaged to reduce noise. The raw projection data was processed to mitigate artifacts before being input into the FBP image reconstruction framework as will be explained in Sections 2.7, and 2.8.

2.5. Simulated imaging task

To allow for a qualitative comparison of the benefits of the DBA, low contrast objects were added to the experimental images. These low contrast objects can be seen in Figures 11, 12, and 15. The low contrast object’s sinogram data was simulated using the same geometry as the experimental set-up and added to the experimental sinogram data. The low contrast objects were all simulated to be circular, each having a radius of 3 mm and a contrast level equal to 40% that of ICRU-44 soft tissue(ICRU 1989) at 50 keV.

Figure 11.

Figure 11

Open in a new tab

(a) non-DBA image and a (b) DBA image acquired at the same dose. (c) non-DBA image acquired at 2.0 times the dose of (a) or (b). The images are displayed at [−0.013 0.039] mm−1.

Figure 12.

Figure 12

Open in a new tab

(a) non-DBA image and a (b) DBA image acquired at the same dose. (c) non-DBA image acquired at 3.6 times the dose of (a) or (b). The images are displayed at [−0.013 0.039] mm−1.

Figure 15.

Figure 15

Open in a new tab

(a) non-VOI and (b) DBA-VOI images with low contrast dots overlayed. Note, the VOI is outlined in Figure 8.

2.6. Dose metric for experimental data

To compare dose levels between DBA and non-DBA scans, all of the rays actually hitting the phantom were determined for an entire CT scan. Then, a separate “air scan” was performed in which the detector signal corresponding to these rays was determined. The sum of the detector signals corresponding to these rays was then taken. To compare dose levels between two different scans, these sums were then divided and it was assumed this ratio was equal to the ratio of total absorbed dose. Figure 4 depicts an image based tutorial on how the dose metric value is computed. Figure 4a is the projection sinogram image. This image is used to create a mask image, Figure 4b, by performing a simple thresholding. The thresholding segments out all areas of the sinogram that do not correspond to phantom material (i.e. these areas correspond to air). Since we are not interested in the dose to air, these areas are excluded from the dose comparison. The mask image is then multiplied with a raw detector signal sinogram, Figure 4c, that corresponds to what the detector would have measured if no object were present (i.e. the raw detector signal from an “air scan”). This yields an image equal to the raw detector signal only for rays incident onto the phantom (Figure 4d). The image shown in Figure 4c must be acquired for each simulated and experimental scan as this image is used in the CT normalization process, so the use of this metric does not require any extra data to be acquired.

Figure 4.

Figure 4

Open in a new tab

Overview of the relative dose calculation. The projection data obtained with the phantom in place (a) is thresholded to create a mask image (b) which is multiplied by the incident fluence (c) to produce an image of the incident fluence that only strikes the imaging object (d).

It should be noted that this dose comparison is essentially an extension of the dose constraint used by Gies et al. (1999) for view-to-view dose modulation to dose modulation within the fan beam. In Gies work, dose was calculated as the total number of photons delivered to an object during a CT acquisition. Gies only counted photons being delivered along the iso-ray, as this was the projection which determined the dose modulation in Gies work. To extend such a metric to modulation within the fan, it is then natural to count the number of photons within each region of dose modulation, which is what the method presented in this section does. This method has limitations, it assumes that a given amount of fluence delivered to the edge of an object deposits the same dose as if it were delivered to the center. This assumption is violated as scattered radiation would be more likely to escape the object for rays near the edge and be absorbed for rays near the center. This method also assumes the beam quality for the “flat-field” and DBA scans are the same. The DBA scans used anywhere from 3.7 to 16.3 mm of Fe filtration for a given wedgelet while the “flat-field” scans used a constant 0.9 mm of Cu. Due to dose deposition differences between different beam qualities and the detector’s energy response, this metric will be effected by a change in beam quality.

We are currently experimenting with using TLD chips arranged in a three dimensional anthropomorphic phantom to compare dose levels between DBA and non-DBA imaging techniques and will present these more robust and accurate dose comparison results in a separate paper.

2.7. Conjugate ray and shift correction

The authors have noticed very severe line artifacts in the projection data after the CT normalization step which we assumed were due to the lack of beam hardening (Szczykutowicz & Mistretta 2013b) and or scatter corrections (Szczykutowicz & Mistretta 2012). These artifacts can be seen in Figure 5a. Early on, however, it was noticed that these artifacts have a pattern in most cases, bright on one side of the wedge boundaries and dark on the other side. These types of artifacts were then postulated to be caused by a shift in the DBA between the actual data scan and the “air scan”. To correct for them, we manually shift the “air scan” data set until these artifacts are minimized (compare Figures 5a and 5b).

Figure 5.

Figure 5

Open in a new tab

(a) No corrections. (b) Shift correction. (c) Shift and conjugate view correction. The sinogram used to reconstruct each image is located directly above the image.

The process of a simple shift producing such a reduction in the line artifact we observed in the experimental data is expected to some degree. The platform that the DBA rests on top of is constructed from wood, and due to the geometry of the c-arm, the wooden platform cannot be made very solid since it must straddle the c-arm (see Figure 1a). Therefore, as our scan times are on the order of hours, it can be expected that this platform may shift slightly during this time. Typical shift corrections are less than 1 mm in magnitude.

In addition to a simple shift correction, a data replacement procedure was also performed. This correction step uses the fact that for a complete 360 degrees worth of CT projection data, each data point can be represented by another data point (Hsieh 2003). For a given datum (in the case of CT projection data we can refer to a specific datum as a ray), the other datum representing the same information is commonly called a conjugate ray. Therefore, the correction scheme employed by us to correct for artifacts occurring at the wedge boundaries was to replace the rays corresponding to the wedge boundaries with conjugate ray values. Taking advantage of the data redundancy to correct for rays at the wedge boundaries should work so long as the wedges are not symmetric with respect to the iso-ray. If the wedges are symmetric with the iso-ray, the conjugate views used to replace wedge boundary values will also be wedge boundary values. This is easy to see considering the transform for conjugate views involves the negative of the fan angle as shown in Hsieh (2003). Up to 10 detector pixels were replaced for a given wedge boundary depending on the width of the artifact region.

Figure 5c depicts the result of applying both the shift and the conjugate view angle correction. It can be appreciated how the artifact level steadily decreases with the application of each correction. There are, however, still some remaining ring artifacts. These artifacts are most likely due to the lack of a beam hardening and scatter correction. Both beam hardening and scatter have been shown to induce ring artifacts in DBA-FFMCT images (Szczykutowicz & Mistretta 2012, Szczykutowicz & Mistretta 2013b).

2.8. Experimental Beam Hardening Correction

No beam hardening corrections were applied to obtain the experimental images shown in Section 4.1; however, the DBA implemented VOI acquisitions (results shown in Section 4.2) quickly made apparent the need for a beam hardening correction for DBA-VOI imaging. The reason the DBA-VOI scans exhibited more beam hardening artifacts was that instead of equalizing the detector fluence, which acted to create a more uniformly hardened beam, the DBA-VOI scans used a binary modulation scheme which created large differences in wedge thickness. These large differences, from 0.37 cm to 1.63 cm (the difference between the maximum and minimum wedge thicknesses) will create large differences in the mean beam energy as a function of detector position. Figure 6 depicts the sinogram and line profile for a DBA acquisition pre and post correction. To correct for this artifact, a single pass soft tissue beam hardening correction (Szczykutowicz & Mistretta 2013a) was adapted for use with DBA experimental data. The projection line integral can be given by

p=−ln∫dED(E)·N(E)·e−μ(E)water·lwater−μ(E)leaf·lleaf·k∫dED(E)·N(E)·e−μ(E)leaf·lleaf·k. (1)

Figure 6.

Figure 6

Open in a new tab

Line integral sinograms without (a) and with (b) beam hardening correction applied. (c) and (d) line profiles through (a) and (b) respectively. The remaining artifacts left over in (b) and (d) were corrected using the conjugate ray replacement scheme.

The goal of the beam hardening correction is to solve for lwater. The terms present in Equation 1 are the incident x-ray spectra N(E), the detector energy response D(E), the water and wedge thicknesses (lwater and lwedge), and the energy dependent linear attenuation coefficients (μ(E)water and μ(E)wedge). lwedge can be determined from the wedge position file that is written out during each DBA acquisition. μ(E)water and μ(E)wedge were assumed to match the values given in ICRU-44 (ICRU 1989) and taken from the NIST database (Hubbell & Seltzer 1995). The difficulty with beam hardening corrections in general, and in this application, is having good estimates for D(E) and N(E). In lieu of actually measuring these quantities or obtaining them from the manufacturer, a crude guess was taken for each factor. It was assumed that the detector energy response was 100% efficient for all energies making D(E) = E. The polychromatic x-ray generator developed by (Siewerdsen et al. 2004) was used to generate a polychromatic spectrum N(E) using the same kVp as the experimental data and assuming 2.5 mm of Al filtered the beam. The value of k was empirically tuned to provide the lowest degree of artifact in CT images reconstructed using the correction. The relationship between the index positions in the wedge positioning file and the actual thickness of the wedge was folded into the correction factor k. This relationship should be linear as the wedges are triangular prisms and their composite thickness is directly related to the overlap between two such shapes. It should be stressed that this beam hardening correction should not be considered robust. It was developed to mitigate the beam hardening artifacts observed in a data set in which the phantom was only composed of a single material. As was the case in previous DBA research (Szczykutowicz & Mistretta 2013a), a more complicated beam hardening correction may be required for an imaging object containing combinations of materials. Equation 1 was minimized as a function of lwater using an unconstrained nonlinear function optimizer from Matlab (Mathworks, Natick MA USA).

3. Experimental study design

3.1. Equalized detector fluence FFMCT

In the first experimental validation of DBA-FFMCT, the DBA was used to provide the detector with a uniform detected signal. Both the cylindrical polyethylene and the anthropomorphic phantom were used to experimentally verify the ability of the DBA to implement uniform detector signal FFMCT. This was accomplished using the LUT presented in Section 2.3 and the means over each wedgelet were calculated according to the algorithm described in Szczykutowicz & Mistretta (2013a) and in Szczykutowicz & Mistretta (2013b). The q value was set to 0.05, which corresponds to setting the wedge thickness based on the detector signal that is 5% and lower than the maximum detector signal within a wedgelet. The Nd parameter was set to the same arbitrary detector signal value for all views. Nd was set to a so called “arbitrary value” since the detector signal values were not scaled to any physical quantity, however, while the absolute value of Nd was arbitrary, adjusting Nd can be thought of as having the same effect as changing the mA for a non-DBA CT acquisition (i.e. for a fixed q, the dose will scale linearly with Nd).

The first view angle for the DBA scans was acquired with the wedges fully retracted (i.e. the wedges were still present in the beam even when fully retracted; the minimum thickness corresponds to the wedges being fully retracted). Subsequent view angles were positioned using the previous view angle’s detector signal. This type of wedge positioning method does not require any a priori patient attenuation information. It does, however, assume that the patient attenuation is a slowly varying function of view angle such that the information from a previously acquired view angle accurately models the current view angle’s attenuation profile.

Figure 7 outlines the work flow for an experimental DBA scan. This figure represents the use of the DBA for a bowtie mimicking, or uniform detector signal scan. For a VOI scan, the wedge positions would be pre-determined using a priori information.

Figure 7.

Figure 7

Open in a new tab

This flowchart elucidates the general overview of the steps involved in acquiring and processing the experimental data. The variable Nq is the desired detector signal for each wedgelet. The variable q is used to calculate the mean signal for each wedgelet’s projection onto the detector (the mean signal is used in wedge positioning(Szczykutowicz & Mistretta 2013a, Szczykutowicz & Mistretta 2013b)). In stead of calculating the mean over the entire wedgelet’s projection on the detector, only the smallest q percent of the max signal for each wedgelet is used.

3.1.1. Noise metrics for equalized detector fluence FFMCT

To assess the degree of variability in noise over non-DBA and DBA-FFMCT images, the noise standard deviation was measured in two different ROIs on the cylindrical phantom and the ratio was calculated. The two regions were chosen to represent areas of maximum and minimum (referred to as regions A and B in Table 2; these regions are depicted in Figure 10) noise within the polyethelene portions of the cylindrical phantom. Ideally, many realizations of the same phantom would be acquired so a noise standard deviation map could be created from an ensemble of pixel values, but due to the relatively long CT image acquisition times (≈2 hours per CT acquisition) this was not feasible.

Table 2.

Comparison of non-DBA to DBA scans (images shown in Figure 10). Noise values are standard deviation (mm−1).

Scan Type Detector Dynamic Range Noise in A Noise in B Noise ratio Relative Dose Equal Noise Relative Dose
non-DBA 13.8 0.0076 0.0055 1.38 2.13 1.89
DBA 7.9 0.0081 0.0079 1.01 1 1
Open in a new tab
Figure 10.

Figure 10

Open in a new tab

(a) Non-DBA CT image. Region A is the 4 sided black polygon and region B is the 6 sided green polygon. (b) DBA CT image. Depicted in (c) and (d) are the noise standard deviation maps corresponding to figures (a) and (b) respectively. The window and level of the standard deviation maps have been set to display ± 10% of the mean pixel standard deviation from region A as listed in Table 2. (c) and (d) represent the standard deviation of the linear attenuation coefficient (mm−1).

In addition to measuring the noise standard deviation inside ROIs, a noise standard deviation map was also calculated for each image. 16 by 16 voxel regions were defined for the non-DBA and DBA-FFMCT images. The standard deviation within each ROI was then determined and a map of these value displayed. The resulting standard deviation values should only be interpreted as representing noise values in the uniform regions of the cylindrical phantom as the metric will give edges a high standard deviation value.

3.1.2. Dose comparison for equalized detector fluence FFMCT

The relative dose level was adjusted between the non-DBA and DBA-FFMCT scans by averaging adjacent detector rows. In this study, we averaged at most 25 detector rows which corresponds to a slice thickness of 3.85 mm. This was done in stead of simply modifying the tube output (i.e. mAs time product) due to the scan times lasting on the order of a few hours. This method of adjusting the dose (and correspondingly the noise) was done with careful attention to ensure the detector was operating in a flat region of the detector quantum efficiency (DQE) as a function of detector entrance exposure. Plots of the zero frequency DQE as a function of detector entrance exposure taken from Jaffray & Siewerdsen (2000) and modern Varian amorphous silicon CsI at panels were compared to the detector entrance exposure used in our studies (the lowest of which was approximately 0.01 mR) to confirm we were operating in a uniform region of DQE. While we did not measure the DQE as a function of dose for our detector, the similarity of our detector to current and published work made us confident that we were operating in a relatively at part of DQE as a function of dose. Averaging detector data will also introduce some partial volume artifacts which can be seen in some of our experimental results. There are also some image details visible in some images that are not visible in others, again, these differences are due to the averaging performed in the z direction on the detector data to adjust the dose per image. For this preliminary study, these differences should not distract from the relevant image quality characteristics, that of image noise and noise uniformity.

3.2. VOI imaging experimental study design

The VOI imaging was implemented by first identifying a small (2.5 cm) VOI within the phantom. Then, this VOI was projected onto the DBA wedges for each view angle. Those wedges overlapping the VOI more than 50% were set to the minimum thickness while all other wedges were set to the maximum thickness. Technically, this is a rudimentary way to implement VOI (Heuscher & Noo 2011, Heuscher & Noo 2012) relative to the work of Bartolac et al. (Bartolac et al. 2011, Bartolac et al. 2010, Bartolac & Jaffray 2012), but for the initial proof of concept implementation, it serves to validate the use of the DBA for VOI. Future work will include implementing more complex VOI wedge positioning algorithms. The selected VOI is depicted on Figure 8. The VOI was purposely chose to not lie along iso-center, therefore the wedges were forced to “follow” the VOI as the phantom is rotated as is shown in in Figure 9.

Figure 8.

Figure 8

Open in a new tab

Location of the VOI.

Figure 9.

Figure 9

Open in a new tab

Wedge position “sinogram”. In this binary sinogram, white and black represent wedges at minimum and maximum thicknesses respectively. This “fluence sinogram” allows an appreciation of how the wedges “follow” the VOI as the projection data is acquired.

For comparison, a regular non-VOI CT image was also acquired using the same technique parameters as the DBA-VOI scan, but with all of the DBA wedges set to the minimum thickness for all view angles. This acquisition allowed the noise levels inside and outside of the VOI to be compared for the DBA-VOI and the non-VOI scans. In addition to comparing noise standard deviation values between the VOI and the rest of the phantom, noise standard deviation maps (as described in Section 3.1.1) were also generated for DBA-VOI and non-VOI scans. These noise standard deviation maps allow for the morphological differences in the noise standard deviation distribution with and without VOI to be qualitatively appreciated.

4. Results and Discussion

4.1. Equalized detector fluence FFMCT experimental results

4.1.1. Cylindrical phantom

Figures 10a and 10b compare the non-DBA and DBA acquired CT images. As it is difficult to tell the differences in noise uniformity between these images visually, noise maps have been created which show the pixel standard deviation inside 16 by 16 voxel regions in the images as shown in Figures 10c and 10d. Important to notice is the difference in uniformity of the noise between the DBA and non-DBA scans. This uniform noise represents a scan in which only enough dose was delivered to obtain a certain level of noise over the entire image. In other words, the smallest dose required for a clinician to make a diagnosis can be realized with the DBA since no dose is wasted in making lower than clinically required noise levels within an image. The actual noise level can be adjusted by changing the tube current and will have to be determined empirically in the clinical setting for a given imaging task. With this said, care must also be taken in futures studies to understand how the changing thickness of DBA filtration will effect image contrast. A change in beam quality due to a modulator thickness change will be present for all attenuation-based fluence field modulators, not just the prototype studied in this work. Table 2 compares the noise ratio between the DBA and non-DBA scans. The non-DBA scan suffers from almost a 40% difference in noise between the two ROIs. The DBA scan shows almost no change in noise between the two regions implying just enough dose was delivered to provide for a uniform noise distribution over the entire image.

As the level of image noise between the DBA-FFMCT and non-DBA scans was not equal in region A, the relative dose shown in Table 2 was adjusted using the relation Dose≈1noise variance. The resulting dose for the case when the DBA and non-DBA scans exhibited equal noise (in region A) is also shown in Table 2. For this simple phantom, the dose of the non-DBA scan was 1.9 times that of the DBA scan. Larger reductions in dose can be expected when the phantom provides more changes in attenuation along the fan angle and as is the case with mA modulation schemes as a function of view angle as was shown by Gies et al. (1999).

The dynamic range did decrease with the addition of the DBA as can be seen from Table 2. The reduction shown in this table, however, is not very impressive compared to some earlier work performed using the DBA where the dynamic range went from 84.2 to 3.7 after the addition of the DBA (Szczykutowicz & Mistretta 2013b). The reason this dynamic range reduction is so small is the small size of the phantom used in this experiment. In general, as the phantom size, and therefore the attenuation change with fan angle decreases, reductions in detector dynamic range and dose due to the DBA decrease. This is an important point since the anthropomorphic phantom used in the following sections only represents a typical medium sized adult. Therefore, the dose reductions cited in the next section would be even larger for a larger patient with the same tissue morphology.

4.1.2. Anthropomorphic phantom

Figures 11 and 12 compare non-DBA scans at different dose levels to DBA-FFMCT scans. The high dose non-DBA scans were dosed (as described in Section 3.1) such that the noise level along the long axis of the shoulder phantom and in the mediastinum region of the thorax phantom matched the respective DBA scans. Figures 11 and 12 clearly show how the use of the DBA can enable better image quality at the same dose (compare the first two columns in Figures 11 and 12) or comparable image quality at a lower dose (compare the last two columns in Figures 11 and 12).

Some morphological details may not match between the DBA and non-DBA scans shown in Figures 11 and 12. This is due to the fact that the scans were acquired at different times and therefore the phantom was set-up in a slightly different position for each scan. In addition, the dose levels for each scan were adjusted by detector averaging as described in Section 3.1.

4.2. VOI imaging experimental results

Figures 13a and 13b compare DBA-VOI and non-VOI CT images while Figures 14a and 14b show the voxel standard deviations for 16 by 16 voxel regions.

Figure 13.

Figure 13

Open in a new tab

(a) non-VOI CT image. (b) DBA-VOI CT image. Note, the VOI is outlined in Figure 8.

Figure 14.

Figure 14

Open in a new tab

(a) and (b) depict the noise maps for figures (a) and (b) shown in Figure 13 respectively. Note, the VOI is outlined in Figure 8. The images represent the standard deviation of the linear attenuation coefficient (mm−1).

The ratio of the VOI noise to the peak noise for the DBA-VOI and non-VOI cases was 1.53 and 1.07 respectively as shown in Table 3. The noise for the non-VOI case is larger in the center of the phantom than at the periphery, this effect is due to the higher uncertainty of measurements corresponding to the center of the phantom (Kak 1979) (no bowtie filter was used in the present study). The DBA-VOI case exhibits a much larger change in noise as expected. To illustrate why VOI imaging is beneficial, Figure 15 compares the non-VOI image to the DBA-VOI image (equal noise within the VOI region only). The noise inside the VOI was adjusted using detector averaging as described in Section 3.1 until equal between the non-VOI and DBA-VOI cases. It can be appreciated from Figure 15 how, if in this case the imaging task is to locate the small low contrast dot within the VOI region, the DBA-VOI enables such a diagnosis. The DBA-VOI acquisition does so without overdosing the patient and providing better than needed image quality over the rest of the phantom as does the non-VOI acquisition (i.e. the non-VOI image provides high SNR outside of the VOI which wastes dose, as can be noticed by the ease at which the low contrast dot can be visualized outside of the VOI region).

Table 3.

Comparison of non-DBA to DBA scans (images shown in Figure 13)

Scan Type Noise in VOI (mm−1) Peak Noise (mm−1) Noise ratio
non-DBA-VOI 0.0067 0.0072 1.07
DBA-VOI 0.0070 0.0107 1.53
Open in a new tab

The motivation for the adjustment of the noise within the VOI between the non-VOI and DBA-VOI scans via detector averaging is not intuitive. One may assume that since the VOI is receiving the same incident dose in both scans that the image noise within this region would also be constant. This assumption, however, does not take into account the possibility of noisy projection data corresponding to areas outside of the VOI influencing the noise within the VOI. This is possible because the filtered back projection algorithm used to reconstruct the images involves a non-local filtering step (Hsieh 2003) that may transfer some of the high noise properties from outside the VOI into the VOI. In addition, in this first application of DBA-VOI imaging, the VOI wedges were positioned such that the VOI only received full dose when a given wedgelet overlapped the VOI by 50% or more. This means that for some view angles part of the VOI did not receive full dose (the dose given in the non-VOI scan when all wedges were at the minimum position).

Figure 16 compares the detector signal for a projection from the VOI-DBA and non-VOI imaging acquisitions. The non-VOI detector signal is 13.02 times that of the VOI signal, on average, for all detector positions except those positions corresponding to the VOI. For these detector positions, the non-VOI signal is only slightly larger than the VOI data. In the absence of scatter, it would be expected that the two scanning modes would have equal signal levels over areas corresponding to the VOI. However, the increased scatter present in the non-VOI scan will likely raise the non-VOI signal. More scatter is expected to be present in the non-VOI scan since the dose to regions outside the VOI is higher in the non-VOI projections relative to the VOI-DBA projections.

Figure 16.

Figure 16

Open in a new tab

Comparison of detector signals between the non-VOI and DBA-VOI imaging.

5. Conclusions

FFMCT has successfully been implemented with the DBA, which represents, to the authors knowledge, the first experimental realization of this scanning technique for CT using a non inverse CT geometry. Such a modulation represents another step along the way of patient specific imaging in the long line of patient specific dose delivery optimization using kVp selection, bowtie filter selection, and angular/z-axis mA modulation. Artifact free DBA images have been presented and the data corrections unique to using a DBA device have been developed. These corrections are nothing novel, they simply represent modifying existing CT data corrections methods for use on data acquired using a DBA. The use of existing data corrections algorithms should decrease the amount of work required to implement the DBA in the clinic.

For noise uniformity imaging, when the DBA is used to emulate an optimized bowtie filter for each view angle, dose reductions of 3.6 times were reported. A very simple implementation of VOI imaging was also performed using the DBA, and the results demonstrate how VOI imaging can be used to provide regional SNR prescriptions, which for some applications could reduce patient dose. With this said, it should be understood that the results presented in this paper represent dose comparisons made using a highly unoptimized comparison of the DBA to non-DBA CT techniques (current state-of-the art dose modulation in CT uses bowtie filters and angular mA modulation, we did not have access to either of these technologies for our experimental data collection). In addition, the differences in beam quality between DBA and non-DBA CT acquisitions remain to be studied as to how these differences effect the dose comparisons made in this paper.

6. Future directions

Future applications of the DBA include photon counting, projection (2-dimensional), and fluoroscopic imaging. In photon counting imaging, the DBA could be used to reduce detector saturation issues (Taguchi et al. 2010). In addition, the DBA could be used to keep the detector signal relatively constant as a function of view angle, failure to maintain a relatively constant signal as a function of view angle has been shown to induce image artifacts as was shown by Shikhaliev (2009). For projection and fluoroscopic imaging, the DBA could be used to: equalize the detector signal to minimize the amount of image post processing required to produce images that can be displayed within the same display range; reduce scatter by reducing dose to lightly attenuating areas of the image; and to “follow” moving anatomy. It is likely that the orientation of the DBA device to the patient will have a large effect on image quality and dose when used in 2-dimensional and fluoroscopic imaging applications. This is because the DBA can only provide fluence modulation over one dimension, and most areas of the body of clinical interest for interventional procedures will have large variations in attenuation in 2 dimensions. The DBA may be aligned along the patient axis for chest procedures (i.e. allowing the modulation of the DBA to compensate for the mediastinum and lung fields) or aligned across the patient to compensate for diaphragm motion.

In addition to investigating applications that the DBA can be used to implement, more work remains to investigate the performance of the DBA. Due to the attenuation based dose modulation of the DBA, the imaging spectrum is changed as a function of fan angle. Therefore, the imaging spectrum is unique for each view angle for each fan angle. This marks a major departure from current CT acquisitions, in which the spectrum, while changing because of the heel effect and the use of bowtie filters, does not change as a function of view angle (assuming the kVp is constant during the scan). It may be that the use of a beam hardening algorithm effectively mitigates any negative effects of changing the spectrum as a function of view angle. Work must also be done to quantify the scatter radiation caused by the DBA device. This source of scatter radiation could be quite large in magnitude (relative to what is currently seen on conventional CT and c-arm scanners) and add to patient dose and degrade contrast. Scatter, beam quality, noise uniformity (and also noise texture), and incident dose distribution all vary as a function of view angle and over the resulting CT images with DBA-FFMCT. Therefore, traditional metrics like CNR for image quality and CTDIvol for dose may not be adequate for “equal” comparisons between DBA-FFMCT and non-FFMCT image methods. The results of this paper should therefore be interpreted as mainly an experimental demonstration of DBA-FFMCT (and FFMCT in general) while fully acknowledging future works in this field should constantly progress in their complexity with regards to image quality and dose comparisons due to the unique nature of FFMCT.

Ideally, the next DBA prototype will be made to fit inside the existing collimator housing of c-arm and conventional CT systems. Fortunately, as the DBA is made smaller, due to the DBA design, the performance in terms of soft tissue compensation and “relative” tube loading increases as shown in Szczykutowicz & Mistretta (2013a). Therefore work is underway to minimize the device.

Lastly, one piece of the current DBA image acquisition chain needs to be eliminated. Currently, as shown in Figure 7, an “air scan” is required for each DBA acquisition. The need for a new “air scan” for each acquisition is due to the unique wedge positions used for each view angle for DBA scans. It is unlikely that the clinical work flow would allow for an “air scan” after every patient. This would be especially true for interventional suites when the patient may still be lying inside the scanner awaiting further treatment depending on the results from the CT; such a patient should not have to be removed from the scanner in order for an “air scan” to be acquired. Work is currently in progress to enable “air scan” free DBA acquisitions.

Acknowledgments

This work is supported by an NIH training grant 5T32CA009206-33 and a grant from Siemens Medical Systems. The authors would also like to thank Nick Bevins, Joe Zambelli, and Gary Frank for their advice concerning prototyping and machine shop skills. The authors would also like to thank Kevin Royalty for training in raw data access on the Zeego and Eric Oberstar for designing the original prototype. The authors also thank Triple Ring Technologies for lending the anthropomorphic phantom which was purchased under NIH grant 5RC1HL100436-02.

References

  1. Bartolac S, Graham S, Siewerdsen J, Jaffray D. Compensator Approaches for Intensity Modulated Computed Tomography. International Conference on Image Formation in X-ray Computed Tomography. 2010;1(1):101. [Google Scholar]
  2. Bartolac S, Graham S, Siewerdsen J, Jaffray D. Fluence field optimization for noise and dose objectives in ct. Medical Physics. 2011;38:S2. doi: 10.1118/1.3574885. [DOI] [PubMed] [Google Scholar]
  3. Bartolac S, Jaffray D. Fluence Field Modulated Computed Tomography. International Conference on Image Formation in X-ray Computed Tomography. 2012;2(2):119–122. [Google Scholar]
  4. Burion S, Sandman A, Bechtel K, Solomon E, Funk T. X-ray dose reduction by adaptive source equalization and electronic region-of-interest control. Proceedings of SPIE. 2011;Vol. 7961:79612B. [Google Scholar]
  5. Chen L, Yu L, Leng S, McCollough C. Ct volume-of-interest (voi) scanning: Determination of radiation reduction outside the voi. RSNA. 2011;SSK15-07 [Google Scholar]
  6. Gies M, Kalender W, Wolf H, Suess C, Madsen M. Dose reduction in ct by anatomically adapted tube current modulation. i. simulation studies. Medical physics. 1999;26:2235. doi: 10.1118/1.598779. [DOI] [PubMed] [Google Scholar]
  7. Graham S, Moseley D, Siewerdsen J, Jaffray D. Compensators for dose and scatter management in cone-beam computed tomography. Medical physics. 2007;34:2691. doi: 10.1118/1.2740466. [DOI] [PubMed] [Google Scholar]
  8. Heuscher D, Noo F. Nuclear Science Symposium and Medical Imaging Conference (NSS/MIC), 2011 IEEE. IEEE; 2011. Ct dose reduction using dynamic collimation; pp. 3470–3473. [Google Scholar]
  9. Heuscher D, Noo F. Ct dose reduction using dynamic collimation; The Second international conference on image formation in X-ray computed tomography; 2012. pp. 115–118. [Google Scholar]
  10. Hsieh J. Computed tomography: principles, design, artifacts, and recent advances. Vol. 114. Society of Photo Optical; 2003. [Google Scholar]
  11. Hsieh SS, Pelc NJ. The feasibility of a piecewise-linear dynamic bowtie filter. Medical physics. 2013a;40:031910. doi: 10.1118/1.4789630. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Hsieh SS, Pelc NJ. SPIE Medical Imaging. International Society for Optics and Photonics; 2013b. Optimized control of a dynamic, prepatient attenuator; pp. 86681Q–86681Q. [Google Scholar]
  13. Hubbell J, Seltzer S. Tables of x-ray mass attenuation coefficients and mass energy-absorption coefficients 1 kev to 20 mev for elements z= 1 to 92 and 48 additional substances of dosimetric interest Technical report National Inst. of Standards and Technology-PL, Gaithersburg, MD (United States). Ionizing Radiation Div. 1995 [Google Scholar]
  14. ICRU. Tissue substitutes in radiation dosimetry and measurement. Report 44. 1989 [Google Scholar]
  15. Jaffray DA, Siewerdsen JH. Cone-beam computed tomography with a flat-panel imager: initial performance characterization. Med. Phys. 2000;27:1311–1320. doi: 10.1118/1.599009. [DOI] [PubMed] [Google Scholar]
  16. Kak A. Computerized tomography with x-ray, emission, and ultrasound sources. Proceedings of the IEEE. 1979;67(9):1245–1272. [Google Scholar]
  17. Kak AC, Slaney M. Principles of Computerized Tomographic Imaging. Bellingham, WA: IEEE; 1988. [Google Scholar]
  18. Kalender W, Wolf H, Suess C. Dose reduction in CT by anatomically adapted tube current modulation. II. Phantom measurements. Medical physics. 1999;26:2248. doi: 10.1118/1.598738. [DOI] [PubMed] [Google Scholar]
  19. Kolditz D, Kyriakou Y, Kalender W. Volume-of-interest (voi) imaging in c-arm flat-detector ct for high image quality at reduced dose. Medical physics. 2010;37:2719. doi: 10.1118/1.3427641. [DOI] [PubMed] [Google Scholar]
  20. Liu X, Shaw CC, Lai CJ, Wang T. Comparison of scatter rejection and low-contrast performance of scan equalization digital radiography (sedr), slot-scan digital radiography, and full-field digital radiography systems for chest phantom imaging. Medical Physics. 2011;38:23. doi: 10.1118/1.3519903. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. McCollough C, Chen G, Kalender W, Leng S, Samei E, Taguchi K, Wang G, Yu L, Pettigrew R. Achieving routine submillisievert ct scanning: Report from the summit on management of radiation dose in ct. Radiology. 2012 doi: 10.1148/radiol.12112265. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Molloi S, Van Drie A, Wang F. X-ray Beam Equalization: Feasibility and Performance of an Automated Prototype System in a Phantom and Swine1. Radiology. 2001;221(3):668. doi: 10.1148/radiol.2213010183. [DOI] [PubMed] [Google Scholar]
  23. Plewes D. A scanning system for chest radiography with regional exposure control: Theoretical considerations. Medical physics. 1983;10:646. doi: 10.1118/1.595327. [DOI] [PubMed] [Google Scholar]
  24. Plewes D, Vogelstein E. A scanning system for chest radiography with regional exposure control: Practical implementation. Medical physics. 1983;10:655. doi: 10.1118/1.595333. [DOI] [PubMed] [Google Scholar]
  25. Shikhaliev P. Projection x-ray imaging with photon energy weighting: experimental evaluation with a prototype detector. Physics in Medicine and Biology. 2009;54:4971. doi: 10.1088/0031-9155/54/16/009. [DOI] [PubMed] [Google Scholar]
  26. Siewerdsen J, Waese A, Moseley D, Richard S, Jaffray D. Specktr: A computational tool for x-ray spectral analysis and imaging system optimization. Med. Phys. 2004;31:3057–3067. doi: 10.1118/1.1758350. [DOI] [PubMed] [Google Scholar]
  27. Sperl J, Beque D, Claus B, De Man B, Senzig B, Brokate M. Computer-assisted scan protocol and reconstruction (caspar) reduction of image noise and patient dose. Medical Imaging, IEEE Transactions on. 2010;29(3):724–732. doi: 10.1109/TMI.2009.2034515. [DOI] [PubMed] [Google Scholar]
  28. Stiller W, Veloza S, Kauczor HU. An analytic modulation function for dynamic adaption of ct beam-shaping filter geometry to subject attenuation in dependence of fan and projection angle: Comparison of calculated and mc-simulated attenuation profiles for elliptic cross-sections. RSNA. 2012;SSK17-09 [Google Scholar]
  29. Szczykutowicz T, Mistretta C. Practical considerations for intensity modulated ct. Proceedings of SPIE. 2012;Vol. 8313:83134E. [Google Scholar]
  30. Szczykutowicz T, Mistretta C. Design of a digital beam attenuation system for computed tomography: Part i. system design and simulation framework. Medical Physics. 2013a;40 doi: 10.1118/1.4773879. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Szczykutowicz T, Mistretta C. Design of a digital beam attenuation system for computed tomography: Part ii. performance study and initial results. Medical Physics. 2013b;40 doi: 10.1118/1.4773880. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Szczykutowicz T, Mistretta C. Intensity modulated ct implemented with a dynamic bowtie filter. Proceedings of SPIE. 2013c;Vol. 8313:161. [Google Scholar]
  33. Szczykutowicz T, Mistretta C. Volume of interest ct implemented with a dynamic bowtie filter. Proceedings of SPIE. 2013d;Vol. 8313:97. [Google Scholar]
  34. Taguchi K, Frey E, Wang X, Iwanczyk J, Barber W. An analytical model of the effects of pulse pileup on the energy spectrum recorded by energy resolved photon counting x-ray detectors. Medical physics. 2010;37:3957. doi: 10.1118/1.3429056. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Toth T, Ge Z, Daly M. The influence of patient centering on ct dose and image noise. Medical physics. 2007;34:3093. doi: 10.1118/1.2748113. [DOI] [PubMed] [Google Scholar]
  36. Veldkamp W, Kroft L, Mertens B, Geleijns J. Digital Slot-Scan Charge-coupled Device Radiography versus AMBER and Bucky Screen-Film Radiography: Comparison of Image Quality in a Phantom Study1. Radiology. 2005;235(3):857. doi: 10.1148/radiol.2353031919. [DOI] [PubMed] [Google Scholar]

RESOURCES