Automatic system calibration method of Xray CT

0Associated
Cases 
0Associated
Defendants 
0Accused
Products 
0Forward
Citations 
0
Petitions 
1
Assignment
First Claim
1. A method of reconstructing a computed tomography (CT) image, the method comprising:
 i) obtaining an initial CT image;
ii) performing a reconstruction algorithm on the initial CT image to obtain a reconstructed image of the initial CT image;
iii) using the reconstructed image to adjust one or more parameters associated with the image comprising using a Locally Linear Embedding (LLE) method on each of the one or more parameters;
iv) using the adjusted one or more parameters to perform the reconstruction algorithm on the reconstructed image to obtain an updated reconstructed image; and
v) repeating steps iii)iv), using the mostrecently updated reconstructed image in each repeated step iii), and updating the same one or more parameters in each repeated step iii) until a threshold value of a predetermined characteristic is met;
wherein the LLE method comprises;
with the K nearest vectors, representing the original data vector linearly with its neighboring vectors;
1 Assignment
0 Petitions
Accused Products
Abstract
Systems and methods for geometric calibration and image reconstruction in computed tomography (CT) scanning using iterative reconstruction algorithms are provided. An iterative reconstruction algorithm can be used to reconstruct an improved image, and then the improved image can be used to adjust inaccurate parameters by using a Locally Linear Embedding (LLE) method. Adjusted parameters can then be used to reconstruct new images, which can then be used to further adjust the parameters. The steps of this iterative process can be repeated until a quality threshold is met.
17 Citations
No References
Ultra low radiation dose Xray CT scanner  
Patent #
US 8,005,184 B2
Filed 04/24/2007

Current Assignee
Wisconsin Alumni Research Foundation

Sponsoring Entity
Wisconsin Alumni Research Foundation

FACE RECOGNITION METHOD  
Patent #
US 20100135541A1
Filed 01/06/2009

Current Assignee
National Tsing Hua University

Sponsoring Entity
National Tsing Hua University

Ultra Low Radiation Dose XRay CT Scanner  
Patent #
US 20080267484A1
Filed 04/24/2007

Current Assignee
Wisconsin Alumni Research Foundation

Sponsoring Entity
Wisconsin Alumni Research Foundation

Artifact correction  
Patent #
US 20070276215A1
Filed 03/08/2005

Current Assignee
Koninklijke Philips N.V.

Sponsoring Entity
Koninklijke Philips N.V.

Image clustering with metric, local linear structure, and affine symmetry  
Patent #
US 20050141769A1
Filed 11/15/2004

Current Assignee
Honda Motor Company

Sponsoring Entity
Honda Motor Company

Method for embedding and extracting digital data in images and video  
Patent #
US 6,332,030 B1
Filed 01/14/1999

Current Assignee
Regents of the University of California

Sponsoring Entity
Regents of the University of California

Matching of Regions of Interest Across Multiple Views  
Patent #
US 20120088981A1
Filed 10/06/2011

Current Assignee
Siemens Healthcare GMBH

Sponsoring Entity
Siemens Medical Solutions USA Incorporated

System and method for reconstruction of Xray images  
Patent #
US 8,189,735 B2
Filed 07/22/2010

Current Assignee
General Electric Company

Sponsoring Entity
General Electric Company

FEATURE CONVERSION DEVICE, SIMILAR INFORMATION SEARCH APPARATUS PROVIDED THEREWITH, CODING PARAMETER GENERATION METHOD, AND COMPUTER PROGRAM  
Patent #
US 20120219212A1
Filed 02/27/2012

Current Assignee
Denso IT Laboratory Inc.

Sponsoring Entity
Denso IT Laboratory Inc.

COMPUTERIZED IMAGE RECONSTRUCTION METHOD AND APPARATUS  
Patent #
US 20150030227A1
Filed 02/01/2013

Current Assignee
The Research Foundation for The State University of New York

Sponsoring Entity
The Research Foundation for The State University of New York

STORED LUMINESCENCE COMPUTED TOMOGRAPHY  
Patent #
US 20150157286A1
Filed 12/08/2014

Current Assignee
Rensselaer Polytechnic Institute

Sponsoring Entity
Rensselaer Polytechnic Institute

DYNAMIC BOWTIE FILTER FOR CONEBEAM/MULTISLICE CT  
Patent #
US 20160113602A1
Filed 10/27/2015

Current Assignee
Rensselaer Polytechnic Institute

Sponsoring Entity
Rensselaer Polytechnic Institute

XRAY PHASECONTRAST IMAGING  
Patent #
US 20160135769A1
Filed 11/17/2015

Current Assignee
Rensselaer Polytechnic Institute

Sponsoring Entity
Rensselaer Polytechnic Institute

XOPTOGENETICS / UOPTOGENETICS  
Patent #
US 20160166852A1
Filed 12/16/2015

Current Assignee
Rensselaer Polytechnic Institute

Sponsoring Entity
Rensselaer Polytechnic Institute

Attenuation map reconstruction from TOF PET data  
Patent #
US 9,600,910 B2
Filed 01/08/2015

Current Assignee
Rensselaer Polytechnic Institute

Sponsoring Entity
Rensselaer Polytechnic Institute

Computed tomography based on linear scanning  
Patent #
US 9,730,657 B2
Filed 12/17/2014

Current Assignee
Rensselaer Polytechnic Institute

Sponsoring Entity
Rensselaer Polytechnic Institute

NANOPARTICLEENABLED XRAY MAGNETIC RESONANCE IMAGING (NXMRI)  
Patent #
US 20170043041A1
Filed 04/21/2015

Current Assignee
Rensselaer Polytechnic Institute

Sponsoring Entity
Rensselaer Polytechnic Institute

30 Claims
 1. A method of reconstructing a computed tomography (CT) image, the method comprising:
i) obtaining an initial CT image; ii) performing a reconstruction algorithm on the initial CT image to obtain a reconstructed image of the initial CT image; iii) using the reconstructed image to adjust one or more parameters associated with the image comprising using a Locally Linear Embedding (LLE) method on each of the one or more parameters; iv) using the adjusted one or more parameters to perform the reconstruction algorithm on the reconstructed image to obtain an updated reconstructed image; and v) repeating steps iii)iv), using the mostrecently updated reconstructed image in each repeated step iii), and updating the same one or more parameters in each repeated step iii) until a threshold value of a predetermined characteristic is met; wherein the LLE method comprises; with the K nearest vectors, representing the original data vector linearly with its neighboring vectors;  View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18)
 19. A method of reconstructing a computed tomography (CT) image, the method comprising:
i) obtaining an initial CT image; ii) performing a reconstruction algorithm on the initial CT image to obtain a reconstructed image of the initial CT image; iii) using the reconstructed image to adjust one or more parameters associated with the image comprising using a Locally Linear Embedding (LLE) method on each of the one or more parameters; iv) using the adjusted one or more parameters to perform the reconstruction algorithm on the reconstructed image to obtain an updated reconstructed image; and v) repeating steps iii)iv), using the mostrecently updated reconstructed image in each repeated step iii), and updating the same one or more parameters in each repeated step iii) until a threshold value of a predetermined characteristic is met; wherein the LLE method comprises; calculating, with the weight coefficients W=(w_{ik}), the global internal coordinate Y=(y_{i}) by solving the equation;  View Dependent Claims (20, 21, 22, 23, 24)
 25. A method of reconstructing a computed tomography (CT) image, the method comprising:
i) obtaining an initial CT image; ii) performing a reconstruction algorithm on the initial CT image to obtain a reconstructed image of the initial CT image; iii) using the reconstructed image to adjust one or more parameters associated with the image comprising using a Locally Linear Embedding (LLE) method on each of the one or more parameters; iv) using the adjusted one or more parameters to perform the reconstruction algorithm on the reconstructed image to obtain an updated reconstructed image; and v) repeating steps iii)iv), using the mostrecently updated reconstructed image in each repeated step iii), and updating the same one or more parameters in each repeated step iii) until a threshold value of a predetermined characteristic is met; wherein the LLE method comprises solving the following linear system of equations;
Au=b_{i},where u=(u_{1}, u_{2}, . . . , u_{J}) is an image represented as a J dimensional vector, J is the number of pixels, b_{i}=(b_{1}, b_{2}, . . . , b_{L}) is data, L is the number of vector elements, and A=(a_{jk}) is a projection matrix related to the one or more (geometric) parameters, wherein performing the reconstruction algorithm comprises calculating a projection matrix, which is affected by the one or more (geometric) parameters;
A=A(P),where P is an estimated parameter vector, which includes the one or more parameters, wherein the projections of the projection matrix are calculated using a distancedriven method, wherein using the reconstructed image to adjust one or more parameters comprises minimizing the mean squared error between the projection data and reprojected projection data, which can be formulated as;
P=arg min∥
b_{i}−
{tilde over (b)}_{ij}∥
_{2}^{2}s.t. A(P)u=b_{i},where b_{i }is the projection vector obtained from the measurement along different projection views, {tilde over (b)}_{ij }is the corresponding reprojected projection vector from a reconstructed image with sampled parameters, and P is the updated vector of parameters, and wherein the reprojected projection vector can be calculated within a densely sampled parametric range;
{tilde over (P)}_{j}=(p_{j1},p_{j2}, . . . ,p_{jn}). View Dependent Claims (26, 27, 28, 29, 30)
1 Specification
The present application is a national stage application of International Application No. PCT/US2016/023460, filed Mar. 21, 2016, which claims the benefit of U.S. Provisional Application Ser. No. 62/135,861, filed Mar. 20, 2015, both of which are incorporated herein by reference in their entirety, including any figures, tables, and drawings.
This invention was made with government support under grant No. R01 EB016977 and U01 EB017140 awarded by National Institutes of Health. The government has certain rights in the invention.
In the medical imaging field, Xray computed tomography (CT) provides critical diagnostic information. Recently, CT techniques have been developed to provide quality images at low radiation dose. Patient motion or inaccurate machinery can lead to inaccurate projection angles or distance between the Xray source and the center of the object being scanned, thereby resulting in inaccurate results. Another challenging problem in this field is geometric calibration, such as for Carm CT and ultrahigh resolution CT. This problem is also related to rigid patient motion compensation, because motion is relative between imaging components and a patient body.
To perform geometric calibration and motion correction, a number of methods have been proposed recently. Analytic methods with a calibration phantom and iterative methods with or without a calibration phantom have been proposed. Analytic methods are widely used in industrial CT and can be based on the identification of elliptical parameters in conebeam geometry. Some calibration methods are iterative, such as optimizationbased calibration for conebeam CT, selfcalibration for conebeam CT, and selfcalibration for conebeam microCT. There is an overlap in the literature on geometric calibration and motion reduction. While some motion reduction methods utilize fast scanning, even with multisourcedetector systems or while avoiding motionaffected data, other methods estimate patient motion and compensate for its effect. Each of these methods has limitations, though.
The subject invention provides novel and advantageous systems and methods for geometric calibration and image reconstruction in computed tomography (CT) scanning (e.g., Xray CT scanning) using one or more iterative reconstruction algorithms. An iterative reconstruction algorithm (e.g., based on Total Variation) can be used to reconstruct an improved image, and then the improved image can be used to adjust inaccurate parameters, for example by using a Locally Linear Embedding (LLE) method. Adjusted parameters can then be used to reconstruct new images, which can then be used to further adjust the parameters. The steps of this iterative process can be repeated until a quality threshold is met. This can lead to automatic system calibration for Xray CT, as opposed to related art parameter adjustment methods and systems based on accurate but expensive machinery.
In an embodiment, a method of reconstructing a CT image can include: obtaining an initial CT image; performing a reconstruction algorithm on the initial CT image to obtain a reconstructed image of the initial CT image; using the reconstructed image to adjust one or more parameters associated with the image (e.g., using an LLE method); using the adjusted one or more parameters to perform the reconstruction algorithm on the reconstructed image to obtain an updated reconstructed image; and iteratively repeating the parameter adjustment and updating of the image reconstructions (using the mostrecently updated reconstructed image each time and updating the same one or more parameters each time) until a threshold value of a predetermined characteristic is met.
In a further embodiment, a CT system can include: a radiation source; a detector for detecting radiation from the radiation source; and a computer including a computer readable medium having computerexecutable instructions stored thereon for performing a method of the subject invention.
The subject invention provides novel and advantageous systems and methods for geometric calibration and image reconstruction in computed tomography (CT) scanning (e.g., Xray CT scanning) using one or more iterative reconstruction algorithms. An iterative reconstruction algorithm (e.g., based on Total Variation) can be used to reconstruct an improved image, and then the improved image can be used to adjust inaccurate parameters, for example by using a Locally Linear Embedding (LLE) method. Adjusted parameters can then be used to reconstruct new images, which can then be used to further adjust the parameters. The steps of this iterative process can be repeated until a quality threshold is met. This can lead to automatic system calibration for Xray CT, as opposed to related art parameter adjustment methods and systems based on accurate but expensive machinery. Thus, methods and systems of the subject invention can significantly decrease costs in Xray CT. Also, system geometric parameters can be extracted directly from projection data under practical conditions.
Embodiments of the subject invention can adjust inaccurate parameters in CT scanning using a mathematical method, thereby helpfully reconstructing CT images with better quality. Because this method is an iterative method, in an embodiment, parallel calculation (e.g., a GPU method) can be used and can decrease calculation time.
In many embodiments of the subject invention, geometric calibration (e.g., iteratively updating parameters) can be performed on CT scanning (e.g., Xray CT scanning) by coupling LLE and reprojection. An initial parameterization can be used to reconstruct a CT image, and reprojected projection vectors that sample the parametric range densely can be calculated. With the reprojected projection vectors and the original projection vectors, weight coefficients and neighbors for LLE can be calculated, and the parameter estimation can be updated. An image can be iteratively reconstructed until a satisfactory quality is achieved (for example, until one or more threshold values in one or more characteristics of interest is/are met). In certain embodiments, LLE and reprojection can be used on fanbeam imaging geometry CT scanning.
LLE was proposed in 2000 as an unsupervised manifold learning algorithm for dimensionality reduction (Roweis et al., “Nonlinear dimensionality reduction by locally linear embedding”, Science 290.5500 (2000): 23232326; and Roweis et al., “An introduction to locally linear embedding”, http://www.cs.toronto.edu/˜roweis/lle/publications.html (2000); both of which are hereby incorporated herein by reference in their entireties). Related to classic dimensionality reduction methods such as principal component analysis (PCA) and multidimensional scaling (MDS), LLE computes eigenvectors but does so locally to preserve intrinsic geometry embedded in a high dimensional space. LLE is easy to implement and yet gives excellent performance when data are sufficiently dense. LLE can include three steps: Step 1 can include finding the K nearest neighbors for each highdimensional point in terms of the Euclidean distance; step 2 can include representing each point linearly with its neighbors and calculating weight coefficients for its neighbors; and step 3 can include mapping high dimensional data with the weight coefficients to a low dimensional representation on an intrinsic manifold.
In many embodiment of the subject invention, with the data vector b_{i}, the first step of LLE can be to find its K nearest neighbors in its dataset vectors {tilde over (b)}_{ij }according to the Euclidean distance:
d_{ij}=∥b_{i}−{tilde over (b)}_{ij}∥_{2}^{2}. (1)
With the K nearest vectors, the second step can be to represent the original data vector linearly with its neighboring vectors:
where {tilde over (b)}_{ik }are the K nearest neighbors and w_{ik }are respectively weight coefficients, and
The weight matrix can be solved by minimizing the following error:
Using the constraint
to solve Equation (3) is equivalent to solving the linear system:
where C=(c_{jk}) is the local covariance matrix calculated as:
C=(b_{i}−{tilde over (b)}_{ik})^{T}(b_{i}{tilde over (b)}_{ik}). (5)
With the weight coefficients W=(w_{ik}), the third step can include calculating the global internal coordinate Y=(y_{i}) by solving the equation:
By the RayleitzRitz theorem, the solution of Equation (6) is given by the bottom d+1 eigenvectors of the generalized eigenvalue problem:
MY=λY, (7)
where λ is the eigenvalue, and M=(I−W)^{T}(I−W).
Because the Xray source is on a circular trajectory, it can be convenient to represent fanbeam geometry in a polar coordinate system. Various practical factors can lead to inaccurate geometric parameters in a viewdependent fashion. Specifically, inaccurate projection angles and other parameters can include:
{tilde over (θ)}_{i}=θ_{i}+δ_{i}, i=1,2, . . . ,N, (8)
{tilde over (p)}_{j}=p_{j}ε_{j}, j=1,2, . . . ,M, (9)
where θ_{i }are the accurate projection angles, p_{j }are the other accurate geometric parameters including SOD, ODD, detector offset, and detector tilt angle, which do not depend on θ_{i}. N is the number of views, M is the number of other parameters, and δ_{i }and ε_{j }are respectively the angular error and the parametric error. In general, more accurate geometric parameters will lead to the ability to reconstruct a better CT image.
A rigid patient motion problem can be present in the geometric calibration during fanbeam geometry CT scanning.
In many embodiments of the subject invention, calibration with LLE can be performed. An iterative CT reconstruction approach can be modeled to solve the following linear system of equations:
Au=b_{i}, (10)
where u=(u_{1}, u_{2}, . . . , u_{J}) is an image represented as a J dimensional vector, J is the number of pixels, b_{i}=(b_{1}, b_{2}, . . . , b_{L}) is data, L is the number of vector elements, and A=(a_{jk}) is a projection matrix related to the geometric parameters. If both the system matrix and projection data are known, an iterative algorithm can be used to reconstruct a CT image with a fixed projection matrix by reducing the difference between original and reprojected data. In a reconstruction algorithm, an important step is to calculate the projection matrix, which is affected by the geometric parameters, such as projection angle and SOD; that is,
A=A(P), (11)
where P is an estimated parameter vector, which includes SOD, ODD and detector offset distance, detector tilt angle, and/or projection angle, possibly among others.
In an embodiment, projections can be calculated using a distancedriven method, such as the distancedriven method disclosed Siddon (Fast calculation of the exact radiological path for a threedimensional CT array, Medical Physics 12(2) (1985), 252255), which is hereby incorporated herein by reference in its entirety. Based on the reprojection approach in the iterative reconstruction, a new iterative method can be formulated to estimate the geometric parameters by minimizing the mean squared error between the projection data and reprojected projection data, which can be formulated as:
P=arg min∥b_{i}−{tilde over (b)}_{ij}∥_{2}^{2}s.t. A(P)u=b_{i}, (12)
where b_{i }is the projection vector obtained from the measurement along different projection views, {tilde over (b)}_{ij }is the corresponding reprojected projection vector from a reconstructed image with sampled parameters, and P is the updated vector of parameters. The reprojected projection vector can be calculated by Equations (10) and (11) within a densely sampled parametric range:
{tilde over (P)}_{j}=(p_{j1},p_{j2}, . . . ,p_{jn}). (13)
If the parametric sampling interval is sufficiently small, a true parameter vector is close to neighboring sampled parameter vectors, and the measured projection vector can be linearly expressed by the K nearest reprojected projection vectors associated with the sampled parameter vectors. That is,
where {tilde over (b)}_{ik }are the K nearest reprojection vectors associated with the corresponding K vectors of parameters {tilde over (P)}_{k}, and w_{ik }are weight coefficients. The key relationship is that the weight coefficients for Equations (14) and (15) are the same. Therefore, the real parameter estimation can be refined by searching for the K nearest reprojected projection vectors and updating the parameter vector with the weight coefficients and the corresponding sampled parameters. Consequently, the geometric calibration problem can be solved by dimensionality reduction via LLE.
With a densely sampled parametric domain and correspondingly reprojected projection vectors, the K nearest reprojected projection vectors of an original projection vector can be found and the weight coefficients can be calculated with LLE by using Equations (1), (4), and (5). With the sampled parameters and corresponding weight coefficients, a parametric update can be performed according to Equation (15).
In certain embodiments, universal quality index (UQI) can be used to evaluate reconstructed images and/or as a threshold characteristic to determine when an iterative reconstruction process should be stopped. UQI is described in detail in Wang et al. (A universal image quality index, Signal Processing Letters, IEEE 9(3) (2002), 8184), which is hereby incorporated herein by reference in its entirety. In certain embodiments, geometric calibration results can be quantified using the average angular error (
The UQI evaluates an image by integrating three factors including correlation distortion, brightness distortion, and contrast distortion. The range of UQI is between −1 and 1. The closer to 1 the UQI, the better a reconstructed image will be. Given a reconstructed image u*_{ij }and the ground truth image u_{ij }of S×T, the UQI is defined as:
The average angular error is:
where is the projection angle after calibration, and θ_{i }is the original projection angle. The parametric error is:
P_{Error}=P−P_{On}, (23)
where P is the parameter vector after calibration, and P_{On }is the real parameter vector. The average object x and ycoordinate offsets are:
where x_{i }and y_{i }are the original object x and ycoordinate offsets, while {tilde over (x)}_{i }and {tilde over (y)}_{i }are the calibrated offsets, respectively.
In many embodiments, image reconstruction can be performed using the Ordered Subset Simultaneous Algebraic Reconstruction Technique (OSSART). OSSART is discussed in Wang et al. (Orderedsubset simultaneous algebraic reconstruction techniques, Journal of Xray Science and Technology, 12(3) (2004), 169177), which is hereby incorporated herein by reference in its entirety. In addition, Total Variation (TV) regularization can be used. TV is discussed in detail in Sidky et al. (Accurate image reconstruction from fewviews and limitedangle data in divergent beam CT, Journal of Xray Science and Technology, 14(2) (2006), 119139), which is hereby incorporated herein by reference in its entirety.
In an embodiment, a CT system can include a radiation source (e.g., an Xray source), a detector for detecting radiation (e.g., Xrays) from the radiation source, and a computer system and/or one or more computer readable media having computerexecutable instructions (stored thereon) for performing a method as disclosed herein. For example, the computerexecutable instructions can be for performing an iterative reconstruction algorithm and/or a geometric calibration for CT (e.g., Xray CT).
Embodiments of the subject invention provide calibration of a CT system (e.g., an Xray CT system, such as a twodimensional Xray CT system) of viewwise random geometric parameters. LLE can be used as an important step to estimate geometric parameters subject to inherent lowdimensional consistency, and this has been demonstrated to provide significant improvements (see the Examples). LLE can find the K nearest reprojected projection vectors with the corresponding sampled parameters and update geometric parameters based on the linear combination of sampled parameters with weighting coefficients calculated via LLE.
Unified image reconstruction and parameter estimation schemes disclosed herein can include iteratively updating the projection matrix and an underlying image by reprojection and LLE. This can be applied to CT system (e.g., Xray CT system) calibration and rigid patient motion compensation. Though fanbeam geometry and rigid patient motion have been specifically discussed, this is for exemplary purposes only and should be construed as limiting. Embodiments of the subject invention can be used with other types of geometry (e.g., conebeam geometry) and other types of movement (e.g., nonrigid patient motion).
Embodiments of the subject invention can use one or more iterative reconstruction algorithms (e.g., based on Total Variation) to reconstruct better images, and then use the reconstructed images to adjust inaccurate parameters with LLE method. Adjusted parameters can then be used to reconstruct new images, and then the new images can be used to adjust inaccurate parameters. This iterative process can be repeated (e.g., until one or more threshold values of one or more characteristics (e.g., UQI) is met) to obtain desirable parameters. This can lead to automatic system calibration for CT (e.g., Xray CT), thereby allowing for financial savings by eliminating the need for certain expensive machinery. Inaccurate parameters can be adjusted mathematically, allowing reconstruction of CT images with better quality.
For (Xray) CT, geometric calibration and patient motion compensation (e.g., rigid patient motion compensation) are interrelated issues for optimization of image reconstruction quality. Noncalibrated system geometry and patient movement during a CT scan will result in streaklike blurring and other artifacts in reconstructed images. The LLE approach disclosed herein addresses this challenge and can be performed under a rigid twodimensional object assumption, thereby addressing challenges of geometric calibration and patient motion compensation in a more general way than any related art methods. Projections can be linearly represented by upsampled neighbors via LLE, and CT system parameters can be iteratively estimated from projection data.
Methods and systems of the subject invention can be used in current CT systems, especially in lowcost CT systems having inaccurate machinery (and therefore in need of parameter adjustment). In related art CT systems, large amounts of money are spent on accurate machinery, leading to highcost CT systems that may not be affordable to small hospitals and clinics. The mathematical parameter adjustment methods and systems of the subject invention, which don'"'"'t require highcost, highaccuracy machinery, can adjust parameters only with computer equipment (e.g., a processing device such as a computer and/or one or more computer readable media). Methods and systems of the subject invention are also valuable for CT manufacturers by allowing reduction in the production cost of the CT system, which could lead to the development of a lowcost system and/or a portable CT system that could increase overall CT system sales.
The methods and processes described herein can be embodied as code and/or data. The software code and data described herein can be stored on one or more computerreadable media, which may include any device or medium that can store code and/or data for use by a computer system. When a computer system reads and executes the code and/or data stored on a computerreadable medium, the computer system performs the methods and processes embodied as data structures and code stored within the computerreadable storage medium.
It should be appreciated by those skilled in the art that computerreadable media include removable and nonremovable structures/devices that can be used for storage of information, such as computerreadable instructions, data structures, program modules, and other data used by a computing system/environment. A computerreadable medium includes, but is not limited to, volatile memory such as random access memories (RAM, DRAM, SRAM); and nonvolatile memory such as flash memory, various readonlymemories (ROM, PROM, EPROM, EEPROM), magnetic and ferromagnetic/ferroelectric memories (MRAM, FeRAM), and magnetic and optical storage devices (hard drives, magnetic tape, CDs, DVDs); network devices; or other media now known or later developed that is capable of storing computerreadable information/data. Computerreadable media should not be construed or interpreted to include any propagating signals. A computerreadable medium of the subject invention can be, for example, a compact disc (CD), digital video disc (DVD), flash memory device, volatile memory, or a hard disk drive (HDD), such as an external HDD or the HDD of a computing device, though embodiments are not limited thereto. A computing device can be, for example, a laptop computer, desktop computer, server, cell phone, or tablet, though embodiments are not limited thereto.
The subject invention includes, but is not limited to, the following exemplified embodiments.
A method of reconstructing a computed tomography (CT) image, the method comprising:
i) obtaining an initial CT image;
ii) performing a reconstruction algorithm on the initial CT image to obtain a reconstructed image of the initial CT image;
iii) using the reconstructed image to adjust one or more parameters associated with the image (e.g., the initial CT image or the previous reconstructed image) or the CT system from which the initial CT image was obtained;
iv) using the adjusted one or more parameters to perform the reconstruction algorithm on the reconstructed image to obtain an updated reconstructed image; and
v) repeating steps iii)iv) (using the mostrecently updated reconstructed image in each repeated step iii), and updating the same parameter(s) in each repeated step iii)) until a threshold value of a predetermined characteristic is met.
The method according to embodiment 1, wherein the reconstruction algorithm is based on Total Variation (TV).
The method according to any of embodiments 12, wherein using the reconstructed image to adjust one or more parameters associated with the image comprises is done using a Locally Linear Embedding (LLE) method on each of the one or more parameters.
The method according to embodiment 3, wherein the LLE method includes:
step 1) finding the K nearest neighbors for each highdimensional point in terms of the Euclidean distance;
step 2) representing each point linearly with its neighbors and calculating weight coefficients for its neighbors; and
step 3) mapping high dimensional data with the weight coefficients to a low dimensional representation on an intrinsic manifold.
The method according to any of embodiments 34, wherein the LLE method includes:
finding the K nearest neighbors of a data vector b_{i }in its dataset vectors {tilde over (b)}_{ij }according to the Euclidean distance:
d_{ij}=∥b_{i}−{tilde over (b)}_{ij}∥_{2}^{2}.
The method according to any of embodiments 35, wherein the LLE method (further) includes:
with the K nearest vectors, representing the original data vector linearly with its neighboring vectors:
where {tilde over (b)}_{ik }are the K nearest neighbors and w_{ik }are respectively weight coefficients, and
and the weight matrix can be solved by minimizing the following error:
The method according to embodiment 6, wherein, using the constraint
minimizing the error is equivalent to solving the linear system:
where C=(c_{jk}) is the local covariance matrix calculated as:
C=(b_{i}−{tilde over (b)}_{ik})^{T}(b_{i}−{tilde over (b)}_{ik}).
The method according to any of embodiments 37, wherein the LLE method (further) includes, calculating, with the weight coefficients W=(w_{ik}), the global internal coordinate Y=(y_{i}) by solving the equation:
The method according to embodiment 8, wherein the solution of the equation that can be solved to calculate Y=(y_{i}) is given by the bottom d+1 eigenvectors of the generalized eigenvalue problem:
MY=λY,
where λ is the eigenvalue, and M=(I−W)^{T}(I−W).
The method according to any of embodiments 39, wherein the number of nearest neighbors used in the LLE method is K=2.
The method according to any of embodiments 39, wherein the number of nearest neighbors used in the LLE method is at least 2.
The method according to any of embodiments 111, wherein the predetermined characteristic, of which a threshold value must be met to stop repeating steps iii)iv) is universal quality index (UQI).
The method according to embodiment 12, wherein the threshold value is 0.6.
The method according to embodiment 12, wherein the threshold value is at least 0.6.
The method according to embodiment 12, wherein the threshold value is 0.7.
The method according to embodiment 12, wherein the threshold value is at least 0.7.
The method according to embodiment 12, wherein the threshold value is 0.8.
The method according to embodiment 12, wherein the threshold value is at least 0.8.
The method according to embodiment 12, wherein the threshold value is 0.9.
The method according to embodiment 12, wherein the threshold value is at least 0.9.
The method according to embodiment 12, wherein the threshold value is 0.95.
The method according to embodiment 12, wherein the threshold value is at least 0.95.
The method according to embodiment 12, wherein the threshold value is 0.5.
The method according to embodiment 12, wherein the threshold value is at least 0.5.
The method according to any of embodiments 124, wherein the one or more parameters includes at least one of projection angle, sourcetoobject distance (SOD), objecttodetector distance (ODD), detector offset, detector tilt angle, object xcoordinate offset, object ycoordinate offset, and projection angular error.
The method according to any of embodiments 124, wherein the one or more parameters includes at least two of projection angle, SOD, ODD, detector offset, detector tilt angle, object xcoordinate offset, object ycoordinate offset, and projection angular error.
The method according to any of embodiments 124, wherein the one or more parameters includes at least three of projection angle, SOD, ODD, detector offset, detector tilt angle, object xcoordinate offset, object ycoordinate offset, and projection angular error.
The method according to any of embodiments 124, wherein the one or more parameters includes at least four of projection angle, SOD, ODD, detector offset, detector tilt angle, object xcoordinate offset, object ycoordinate offset, and projection angular error.
The method according to any of embodiments 124, wherein the one or more parameters includes projection angle, SOD, ODD, detector offset, and detector tilt angle.
The method according to any of embodiments 124, wherein the one or more parameters includes object xcoordinate offset, object ycoordinate offset, and projection angular error.
The method according to any of embodiments 330, wherein the LLE method includes solving the following linear system of equations:
Au=b_{i},
where u=(u_{1}, u_{2}, . . . , u_{J}) is an image represented as a J dimensional vector, J is the number of pixels, b_{i}=(b_{1}, b_{2}, . . . , b_{L}) is data, L is the number of vector elements, and A=(a_{jk}) is a projection matrix related to the one or more (geometric) parameters.
The method according to embodiment 31, wherein performing the reconstruction algorithm comprises calculating a projection matrix, which is affected by the one or more (geometric) parameters:
A=A(P),
where P is an estimated parameter vector, which includes the one or more parameters.
The method according to embodiment 32, wherein the projections of the projection matrix are calculated using a distancedriven method.
The method according to any of embodiments 3233, wherein using the reconstructed image to adjust one or more parameters comprises minimizing the mean squared error between the projection data and reprojected projection data, which can be formulated as:
P=arg min∥b_{i}−{tilde over (b)}_{ij}∥_{2}^{2}s.t A(P)u=b_{i},
where b_{i }is the projection vector obtained from the measurement along different projection views, {tilde over (b)}_{ij }is the corresponding reprojected projection vector from a reconstructed image with sampled parameters, and P is the updated vector of parameters.
The method according to embodiment 34, wherein the reprojected projection vector can be calculated by the equations provide in embodiments 31 and 32 within a densely sampled parametric range:
{tilde over (P)}_{j}=(p_{j1},p_{j2}, . . . ,p_{jn}).
The method according to embodiment 35 wherein a true parameter vector is close to neighboring sampled parameter vectors, and the measured projection vector can be linearly expressed by the K nearest reprojected projection vectors associated with the sampled parameter vectors, such that:
where {tilde over (b)}_{ik }are the K nearest reprojection vectors associated with the corresponding K vectors of parameters {tilde over (P)}_{k}, and w_{ik }are weight coefficients.
The method according to any of embodiments 136, wherein obtaining the initial CT image comprises obtaining an initial parameter vector including initial value(s) for the one or more parameters associated with the image (e.g., the initial CT image or the previous reconstructed image) or the CT system from which the initial CT image was obtained.
The method according to embodiment 37, performing the reconstruction algorithm comprises updating the parameter vector with updated value(s) of the one or more parameters based on the (most recently) reconstructed image).
The method according to any of embodiments 138, wherein the reconstruction algorithm comprises using an Ordered Subset Simultaneous Algebraic Reconstruction Technique (OSSART).
The method according to any of embodiments 139, wherein the reconstruction algorithm comprises using TV for regularization.
The method according to any of embodiments 140, wherein step ii) is performed by a processor.
The method according to any of embodiments 141, wherein step iii) is performed by a processor.
The method according to any of embodiments 142, wherein step iv) is performed by a processor.
The method according to any of embodiments 143, wherein step v) is performed by a processor.
A computer readable medium having computerexecutable instructions (stored thereon) for performing the method according to any of embodiments 144.
A CT system, comprising:
a radiation source (e.g., an Xray source);
a detector for detecting radiation (e.g., Xrays) from the radiation source; and
a computer having the computer readable medium according to embodiment 45.
A greater understanding of the present invention and of its many advantages may be had from the following examples, given by way of illustration. The following examples are illustrative of some of the methods, applications, embodiments and variants of the present invention. They are, of course, not to be considered as limiting the invention. Numerous changes and modifications can be made with respect to the invention.
The LLEbased calibration approach was evaluated using an abdomen image phantom from a clinic, shown in
To evaluate reconstructed images, UQI was used. Also, geometric calibration results were quantified with the average angular error
The image reconstruction process was performed using OSSART with TV regularization. As the reconstruction and calibration iteration stop criteria, satisfactory UQI values were monitored for in both the image and projection domains.
The utility of the calibration approach was evaluated for reducing projection angular and other geometric errors. The number of projection angles N=360. The projection angles were randomly perturbed as follows:
θ_{i}=360×(i−1)/N+δ_{i}, i=1,2, . . . ,N. (26)
The other parameters used are shown in Table 1, including the real values, initial values, and final estimates. The length of the detector array was 80 cm. With the geometric parameters, projection data were generated with Poisson noise, assuming that the number of incoming Xray photons was 10^{5}.
To calibrate the parameters efficiently, first the geometric parameters were calibrated one by one rather than all the parameters together being calibrated together. The calibration process proceeded in the following sequence: detector offset; SOD; ODD; detector tilt; and projection angle. The angular sampling steps for each parameter were 0.04 cm, 0.1 cm, 0.1 cm, 0.01°, and 0.02° respectively. The sampling ranges were [0 cm, 0.4 cm], [46 cm, 54 cm], [46 cm, 54 cm], [0°, 2°], and [−1°, 1°] respectively. The number of nearest neighbors was 2. The UQI threshold for stopping was 0.9 for reconstruction and 0.6 for calibration.
Referring to
The calibration algorithm discussed herein was applied in a case of real projection data with geometric parameter errors. The nominal SOD and ODD values were both 38 cm, the number of detector elements was 1024, and the length of the detector was 13.0048 cm. The number of projection angles was 900 in the angular range [0°, 360°]. An object was on the rotation stage while the rotation center was not on the line between the Xray source and the detector center, as shown in
Referring to
Calibration as discussed herein was tested for a situation involving rigid patient motion. As discussed, rigid patient motion can be represented as object xcoordinate offset error, object ycoordinate offset error, and projection angle error. Rigid patient motion calibration was simulated with an abdomen phantom. The number of projection angles and other parameters were the same as in Example 1. The object xcoordinate offset error and object ycoordinate offset error were all randomly selected within the range of [−1 cm, 1 cm], and the projection angle errors were randomly chosen within the range of [−1°, 1°].
The calibration was in the following sequence: object xcoordinate offset; object ycoordinate offset; and projection angle. The sampling rates for the parameters were 0.05 cm, 0.05 cm, and 0.1°, respectively. The sampling ranges were [−1 cm, 1 cm], [−1 cm, 1 cm], and [−1°, 1°], respectively. The number of nearest neighbors was K=2. The UQI stop value for reconstruction was 0.9 and for calibration was 0.8.
Referring to
Geometric parameters have different sensitivities on reconstruction quality. In a geometric calibration, detector offset significantly affects reconstruction. Detector tilt and projection angular error have stronger effects than the SOD error and the ODD errors on reconstruction quality. This can help explain the calibration results for detector offset, detector tilt, and projection angle being better than those for SOD and ODD. In patient motion calibration, object x and ycoordinate offsets were more effective on the quality of a reconstructed image than the projection angular error was.
A closelyrelated issue is the parametric sampling rate for reprojection. The greater the sampling rate, the more accurate the calibration results will be, but the higher computational cost will be involved. Sampling rate can be chosen with balance between calibration results and computational overhead in mind.
An important issue in using TV is its weight. If it is too small, TV will not be able to reduce artifacts and noise. If it is too large, TV will oversmoothen CT images. The TV parameter depends on levels of artifacts and noise. In Examples 13, the TV parameter was empirically set to 0.1 after geometric calibration. A smaller TV parameter could be used in a case of weaker noise.
The LLEbased calibration approach was evaluated using a SheppLogan phantom.
Referring to
The LLEbased calibration approach was evaluated using a chest phantom with Poisson noise of 10^{5 }photons.
Referring to
It should be understood that the examples and embodiments described herein are for illustrative purposes only and that various modifications or changes in light thereof will be suggested to persons skilled in the art and are to be included within the spirit and purview of this application.
All patents, patent applications, provisional applications, and publications referred to or cited herein (including those in the “References” section) are incorporated by reference in their entirety, including all figures and tables, to the extent they are not inconsistent with the explicit teachings of this specification.
 G. Wang, H. Yu, and B. D. Man. “An outlook on xray CT research and development.” Medical physics 35.3 (2008): 10511064.
 G. Wang, S. Zhao, and H. Yu, et al., “Design, analysis and simulation for development of the first clinical microCT scanner.” Academic Radiol. 12.4 (2005):511525.
 C. Rock, U. Linsenmaier, R. Brandl, et al. “Introduction of a new mobile Carm/CT combination equipment (ISOC3D). Initial results of 3D sectional imaging.” Der Unfallchirurg 104.9 (2001): 827833.
 G. Rajiv, M. Grasruck, C. Suess, et al. “Ultrahigh resolution flatpanel volume CT: fundamental principles, design architecture, and system characterization.” European radiology 16.6 (2006): 11911205.
 F. Noo, R. Clackdoyle, C. Mennessier, et al. “Analytic method based on identification of ellipse parameters for scanner calibration in conebeam tomography.” Phys. Med. Biol. 45 (2000) 34893508.
 D. Panetta, N. Belcari, A. D. Guerra, et al. “An optimizationbased method for geometrical calibration in conebeam CT without dedicated phantoms.” Med. Biol. 53 (2008) 38413861.
 W. Wein, A. Ladikos, A. Baumgartner. “Selfcalibration of geometric and radiometric parameters for conebeam computed tomography.” in: Fully 3D, Potsdam, (2011): 327330.
 V. Patel, R. N. Chityala, K. R. Hoffmann, et al. “Selfcalibration of a conebeam microCT system.” Med. Phys. 36 (2009) 4858.
 B. D. Man, S. Basu, D. Beque, et al. “Multisource inverse geometry CT: a new system concept for Xray computed tomography.” Medical Imaging. International Society for Optics and Photonics (2007).
 Y. Liu, H. Liu, Y. Wang, et al. “Halfscan conebeam CT fluoroscopy with multiple sources.” Med. Phys. 28.7 (2001):14661471.
 J. Hsieh. “Computed Tomography, Principles, Design, Artifacts, and Recent Advances. Bellingham.” WA: SPIE, 2003.
 C. R. Crawford and K. F. King. “Computedtomography scanning with simultaneous patient translation.” Med. Phys. 17.6 (1990): 967982.
 G. Wang and M. W. Vannier. “Preliminary study on helical CT algorithms for patient motion estimation and compensation.” IEEE Trans. Med. Imag. 14.2 (1995): 205211.
 W. G. Lu and T. R. Mackie. “Tomographic motion detection and correction directly in sinogram space.” Phys., Med., Biol. 47.8 (2002): 12671284.
 H. Yu, and G. Wang. “Data consistency based rigid motion artifact reduction in fanbeam CT.” Medical Imaging, IEEE Transactions on 26.2 (2007): 249260.
 H. Yu, Y. Wei, J. Hsieh, et al. “Data consistency based translational motion artifact reduction in fanbeam CT.” Medical Imaging, IEEE Transactions on 25.6 (2006): 792803.
 S. Leng, B. Nett, M. Speidel, et al. “Motion artifact reduction in fanbeam and conebeam computed tomography via the fanbeam data consistency condition (FDCC).” Medical Imaging. International Society for Optics and Photonics, 2007.
 S. T. Roweis, and S. K. Lawrence. “Nonlinear dimensionality reduction by locally linear embedding.” Science 290.5500 (2000): 23232326.
 S. T. Roweis, and S. K. Lawrence. “An introduction to locally linear embedding.” unpublished. Available at: http://www.cs.toronto.edu/˜roweis/lle/publications.html (2000).
 R. A. Horn and C. R. Johnson. “Matrix Analysis.” Cambridge University Press, Cambridge, 1990
 R. L. Siddon. “Fast calculation of the exact radiological path for a threedimensional CT array.” Medical physics 12. 2 (1985): 252255.
 Z. Wang, and A. C. Bovik. “A universal image quality index.” Signal Processing Letters, IEEE 9.3 (2002): 8184.
 G. Wang, and M. Jiang. “Orderedsubset simultaneous algebraic reconstruction techniques (OSSART).” Journal of Xray Science and Technology 12.3 (2004): 169177.
 Y. E. Sidky, C. M. Kao, and X. Pan. “Accurate image reconstruction from fewviews and limitedangle data in divergentbeam CT.” Journal of Xray Science and Technology 14.2 (2006): 119139.