Dual-energy CT has attracted much attention in recent years. Most recently, a fast-kVp switching
(FKS) dual-energy method has been presented with clinical and phantom results to demonstrate
its efficacy. The purpose of our study was to quantitatively compare the CTDIW of FKS and
routine CT exams under the body and head conditions. For a fair comparison, the low contrast
detectability (LCD) was matched before measuring dose. In FKS protocols, an x-ray generator
switch rapidly between 140kVp and 80kVp in adjacent views, and the effective tube current is
around 600mA. In addition to the tube voltage and current, the flux ratio between high and low
kVp is optimized by asymmetric sampling of 35%-65%. The head and body protocols further
differ by the gantry speed (0.9sec/1.0sec) and type of bowtie filter (head/body). For baseline
single-energy, we followed the IEC standard head and body protocols (120kV, 1sec, 5mm) but
iteratively adjusted the tube current (mA) in order to match the LCD. CTDIW was measured
using either a 16 cm (for head scanning) or a 32 cm (for body scanning) PMMA phantom of at
least 14 cm in length. The LCD was measured using the water section of Catphan 600. To make
the study repeatable, the automated statistical LCD measurement tool available on GE Discovery
CT750 scanner was used in this work. The mean CTDIW for the head and body single-energy
acquisitions were 57.5mGy and 29.2mGy, respectively. The LCD was measured at 0.45% and
0.42% (slice thickness=5mm, object size=3mm, central 4 images), respectively. The average
CTDIW for FKS head and body scans was 70.4mGy and 33.4mGy, respectively, at the optimal
monochromatic energy of 65 keV. The corresponding LCD was measured at 0.45% and 0.43%,
respectively. This demonstrates that, with matching LCD, CTDIW of FKS is comparable to that
of routine CT exams under head and body conditions.
With the advancement of Computed Tomography technology, improving image quality while reducing patient dose has
been a big technical challenge. The recent CT750 HD system from GE Healthcare provides significantly improved
spatial resolution and the capability to reduce dose during routine clinical imaging. This paper evaluates the image
quality of this system. Spatial resolution, dose reduction, noise, and low contrast detectability have been quantitatively
characterized. Results show a quantifiable and visually discernable higher spatial resolution for both body and cardiac
scanning modes without compromise of image noise. Further, equivalent image quality performance with up to 50%
lower dose has been achieved.
Future generations of CT systems would need a mean to cover an entire organ in a single rotation. A way to accomplish
this is to physically increase detector size to provide, e.g., 120~160mm z (head-foot) coverage at iso. The x-ray cone
angle of such a system is usually 3~4 times of that of a 64-slice (40mm) system, which leads to more severe cone beam
artifacts in cardiac scans. In addition, the extreme x-ray take-off angles for such a system cause severe heel effect,
which would require an increase in anode target angle to compensate for it. One shortcoming of larger target angle is
that tube output likely decreases because of shorter thermal length. This would result in an increase of image noise. Our
goal is to understand from a physics and math point of view, what is the clinical acceptable level of artifacts, resolution,
and noise impact. The image artifacts are assessed through computer simulation of a helical body phantom and visual
comparison of reconstructed images between a 140mm system and a 64-slice system. The IQ impact from target angle
increase is studied analytically and experimentally by first finding the proper range of target angles that give the
acceptable heel effect, then estimating the impact on peak power (flux) and z resolution using an empirical model of
heel effect for given target angle and analytical models of z resolution and tube current loading factor for given target
thermal length. The results show that, for a 140mm system, 24.5% of imaging volume exhibits more severe cone beam
artifacts than a 64-slice system, which also brings up a patient dose concern. In addition, this system may suffer from a
36% peak power (flux) loss, which is equivalent to about 20% image noise increase. Therefore, a wide coverage CT
system using a single x-ray source is likely to face some severe challenges in IQ and clinical accuracy.
Cone-beam filtered backprojection (CB-FBP) is one of the major reconstruction algorithms for digital tomosynthesis. In conventional FBP, the photon fluxes in projections are evenly distributed along the X-ray beam. Due to the limited view angles and finite detector dimensions, this uniform weighting causes non-uniformity in the recon images and leads to cone-beam artifact. In this paper, we propose a 3-D view weighting technique in combination with FBP to combat this artifact. An anthropomorphic chest phantom was placed at supine position to enable the imaging of chest PA view. During a linear sweep of X-ray source, 41 X-ray images at different projection angles were acquired with the following protocol: 120kVp, 160mA, and 0.64mAs/exposure. To create the worst scenario for testing, we chose 60 degrees as the sweep angle in this exam. The data set was reconstructed with conventional CB-FBP and proposed algorithm under the same parameters: FOV = 40x40 cm^2, and slice thickness = 4mm. 3 recon slices were randomly selected for review with slice height = 10.5/14.5/17.5cm. Results were assessed qualitatively by human observers and quantitatively through ROI measurement. In each slice, three pre-defined ROIs (50x50 pixels)--ROI A and B are in artifact more pronounced area, and ROI C is in relatively artifact-free area--are extracted and measured. The non-uniformity error was defined as the ratio of MEAN(AVG(C-A), AVG(C-B)) / AVG(C). The average non-uniformity error over the three test images was 0.428 for without view weighting and only 0.041 for with view weighting.
In digital tomosynthesis, one of the limitations is the presence of out-of-plane blur due to the limited angle acquisition. The point spread function (PSF) characterizes blur in the imaging volume, and is shift-variant in tomosynthesis. The purpose of this research is to classify the tomosynthesis imaging volume into four different categories based on PSF-driven focus criteria. We considered linear tomosynthesis geometry and simple back projection algorithm for reconstruction. The three-dimensional PSF at every pixel in the imaging volume was determined. Intensity profiles were computed for every pixel by integrating the PSF-weighted intensities contained within the line segment defined by the PSF, at each slice. Classification rules based on these intensity profiles were used to categorize image regions. At background and low-frequency pixels, the derived intensity profiles were flat curves with relatively low and high maximum intensities respectively. At in-focus pixels, the maximum intensity of the profiles coincided with the PSF-weighted intensity of the pixel. At out-of-focus pixels, the PSF-weighted intensity of the pixel was always less than the maximum intensity of the profile. We validated our method using human observer classified regions as gold standard. Based on the computed and manual classifications, the mean sensitivity and specificity of the algorithm were 77+/-8.44% and 91+/-4.13% respectively <i>(t=-0.64, p=0.56, DF=4)</i>. Such a classification algorithm may assist in mitigating out-of-focus blur from tomosynthesis image slices.
In this study, we investigate the relationship between quantum noise and spatial resolution for volumetric CT. Both theoretical analysis and experiments were performed to investigate their relationship. In theory, quantum noise can be derived from its relationship to dose, in-plane spatial resolution, recon kernel, and signal-to-noise ratio (SNR). In the experiments, by scanning a Teflon sphere phantom, the 3-D MTF was measured from the edge profile along the spherical surface. Cases of different resolutions (and noise levels) were generated by adjusting recon kernel. To reduce bias, the total photon fluxes were matched: 120kVp, 260mA, and 1sec per gantry rotation. In the end, all data sets were reconstructed using modified FDK algorithm under the same condition: FOV=10cm and slice thickness=0.625mm. Finally, we investigated the efficiency of an image-space adaptive smoothing filter as a noise reduction tool and its impact on spatial resolution. The theoretical analysis indicated that the variance of noise is proportional to at least 4th power of the spatial resolution. Our experimental results supported this conclusion by showing the relationship is 4.6th (helical) or 5th (axial) power. Results also showed that, with properly designed image-space smoothing filters, it is feasible to reduce quantum noise (and the power relationship to a lower order) with smaller loss of spatial resolution.
Tomosynthesis is widely used for three-dimensional reconstruction of objects acquired from limited angle X-ray projection imaging with stationary digital detector. Traditionally, the point-spread function (PSF) in digital tomosynthesis is assumed to be symmetrical with respect to the central axis and shift invariant. The purpose of this research is to characterize the true nature of the PSF by intensity and shape considerations. We assumed that tomosynthesis PSF depended on the imaging geometry and the reconstruction algorithms. In this paper, we describe PSF characterization with respect to the linear geometry and back projection reconstruction. We considered the following parameters: source to image distance (SID) (mm), total number of slices reconstructed after reconstruction, distance (in z-direction) from the first and the last slice to the detector (mm), resolution in X, Y & Z (pix/mm), and total number of projections. Using these parameters, we determined the PSF at every location of the reconstructed volume. The PSF was contained in the plane formed by the linear source trajectory and the point under consideration that extended through all the slices. The results show that the PSF is shift variant and unique at every location and gradually changing over the entire reconstructed volume. The shift from the central axis and central reconstructed slice caused the PSF to exhibit shear corresponding to the X-shift, tilt with the Y-shift and asymmetry with the Z-shift. In summary, we have characterized tomosynthesis PSF to be globally shift variant exhibiting shear, tilt and asymmetry.
A novel method to measure in-plane resolution (modulation transfer function, or MTF) and slice thickness (slice sensitivity profile, or SSP) of a digital radiographic tomosynthesis system is presented. With this method, one can measure these two important system IQ characteristics simultaneously without suffering from incontinuous sampling, aliasing, and partial volume effect as do the existing methods. The method is based on imaging a shallow-angled slice ramp phantom. The MTF is measured as the HWHM of the Fourier transformation of the first derivative of edge profiles. The HWHM corresponding to the sharpest of edge profile represents the in-plane resolution of the system, and the slice thickness of the system is determined from the HWHM vs. z-distance curve. The in-plane resolution result has been confirmed by the measurement from an animal skull specimen. The experiment results have shown that, for a typical 40-degree sweep, 61 projections, and using a Specialized Filtered Backprojection (SFBP) algorithm, the in-plane resolution of the measured system is close to 1 lp/mm (as measured by the HWHM of MTF), and effective slice thickness is 1.7 mm and 4.0 mm at HWHM and HW3TM, respectively. It is also observed that, while the in-plane resolution remains constant between planes at 7 cm and 30 cm above the detector plane, SSP has increased (i.e., slice thickness increased) 20% on average with the increase of the plane height. We demonstrate one of the applications of the method to optimize the sweep angle of a tomosynthesis system. The results show that, in a typical angular range from 20 to 60 degrees, the increase in sweep angle can intrinsically reduce slice thickness but less significantly impact in-plane resolution.
CT can be used to study pulmonary structure-function relationships. There is a growing clinical need to match pulmonary structures across individuals to detect abnormal structure due to disease and to compare regional pulmonary function. In this paper, we propose a novel scheme for registering and warping 3-D pulmonary CT images of different subjects in two main steps: 1) identify a set of reproducible feature points for each CT image to establish correspondences across subjects; 2) use a landmark and intensity-based consistent image registration algorithm to warp a template image volume to the rest of the lung volumes. Effectiveness of the proposed scheme is evaluated and visualized using both gray-level and segmented CT images. Results show that the proposed scheme is able to reduce landmark registration error and relative volume overlapping error from 10.5 mm and 0.70 before registration to 0.4 mm and 0.11, respectively. The proposed scheme can be used to construct a computerized human lung model (or atlas) to help detect abnormal lung structural changes.
We describe a novel method to build 3D statistical shape models for anatomic objects in tomographic images, and demonstrate the use of the model to guide image segmentation. Our method consists of two main steps. In the first step, a statistical shape model is built for a collection of training images. Boundary similarities between adjacent transverse slices are matched to guide inter-slice interpolation. Slice-by-slice correspondences are established between images in the training sets by matching mean boundary curvatures. A statistical shep model is then obtained by principal components analysis. During the second step, the model is used to guide image segmentation. Segmentation is initialized by placing the mean shape into the image under analysis. The model deforms iteratively by updating its shape and pose parameters using the principles of the active shape model. Following the active shape model, an active contour model (snake) is used to refine the object boundary. The proposed methods have been tested using ten volumetric chest HRCT images. The results show that the new method is able to automatically generate 3D object shape models without the need for manual landmark identification. The combination of the active shape model with the active contour model yields a fast, accurate object segmentation.