Digital x-ray tomosynthesis system
First Claim
1. A digital x-ray tomosynthesis system comprising:
- A) at least one x-ray source,B) a two-dimensional digital x-ray image sensor,C) an x-ray data collection-positioning mechanism for positioning, relative to each other, said at least one x-ray source, said image sensor and an object for collecting with said image sensor x-ray attenuation data representing attenuation of a large number of rays of x-radiation from said at least one source through said object to said image sensor, andD) a computer processor programmed with an algorithm for producing tomographic or three-dimensional images of said object or portions of said object using a least squares technique utilizing generalized non-specific functions;
wherein in said algorithm a three-dimensional object being imaged is represented by a scalar function d(xtomo,ytomo,h) in the dimension of Hounsfield units with one Hounsfield unit equal to a 0.1 % difference in the density of water and the range of Hounsfield units is from −
1000 (air) to 1000 (bone) with 0 Hounsfields as the density of water, wherein the attenuation of x-rays directed along a line (trajectory s) is given by where (xtomo,ytomo) are the coordinates where trajectories cross the tomographic plane at z=h.
1 Assignment
0 Petitions
Accused Products
Abstract
Method and device for digital x-ray tomosynthesis. Tomographic and/or three-dimensional images of an object are obtained with an x-ray source and a digital x-ray image sensor. The source, object and sensor are positioned relative to each other and attenuation data is obtained for a large number of rays of x-radiation through the object. A special algorithm is provided to convert the data into images. To calculate the images the algorithm uses iterative processes with a least squares type technique but with generalized (as opposed to specific) functions. The algorithm solves for the functions which are the images. Preferred embodiments include a system having an x-ray point source with a cone of diverging x-rays, a two-dimensional digital x-ray image sensor, two linear translation stages to independently move both the x-ray source and the digital x-ray image sensor, two rotation mechanisms to rotate the two linear translation stages, a microprocessor to control the data acquisition, and a computer programmed with a special algorithm to calculate the tomographic images. A plurality of sets of digital data (representing x-ray algorithm images of an object) are acquired by the digital x-ray image sensor, with the x-ray source and the digital x-ray image sensor located at different positions and angles relative to the object. The digital data representing the x-ray attenuation images is stored in the computer. Special mathematical algorithms then compute multiple images of the object using the acquired digital data. These images could include multiple tomographic images, a three-dimensional image, or a multiple three-dimensional images.
21 Citations
19 Claims
-
1. A digital x-ray tomosynthesis system comprising:
-
A) at least one x-ray source, B) a two-dimensional digital x-ray image sensor, C) an x-ray data collection-positioning mechanism for positioning, relative to each other, said at least one x-ray source, said image sensor and an object for collecting with said image sensor x-ray attenuation data representing attenuation of a large number of rays of x-radiation from said at least one source through said object to said image sensor, and D) a computer processor programmed with an algorithm for producing tomographic or three-dimensional images of said object or portions of said object using a least squares technique utilizing generalized non-specific functions; wherein in said algorithm a three-dimensional object being imaged is represented by a scalar function d(xtomo,ytomo,h) in the dimension of Hounsfield units with one Hounsfield unit equal to a 0.1 % difference in the density of water and the range of Hounsfield units is from −
1000 (air) to 1000 (bone) with 0 Hounsfields as the density of water, wherein the attenuation of x-rays directed along a line (trajectory s) is given bywhere (xtomo,ytomo) are the coordinates where trajectories cross the tomographic plane at z=h. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19)
where
m0i(x,y)=w(x,y)Ai(s){circle around (×- )}psƒ
(x,y)
are noise-free measurements for each N source positions (Xi,Yi);
i=1,N, ni(x,y) is noise, wi(x,y) is the white field image acquired during calibration, and psƒ
(x,y) is a point spread function.
-
-
4. The system as in claim 3 wherein m0i(x,y) is also represented by:
-
m0i(x,y)=w(x,y)A
-
-
5. The system as in claim 4 wherein an equation for m0i(x,y) is inverted to express a tomographic slice image data d(x,y,h) in terms of the acquired noise free data m0(x,y) by minimizing the following generalized chi-squared function:
-
versus a function d(xtomo,ytomo,h), where e is a regularization parameter chosen so that the first term averages 1 per measurement and m0 is implicitly a function of d(xtomo,ytomo,h).
-
-
6. The system as in claim 5 wherein the equation set forth in claim 4 is simplified to produce an equation:
-
and equation set forth in claim 2 is expressed as;
-
-
7. The system as in claim 6 wherein a noise correlation is a noise spectrum function (NPS) using the following equation:
-
(ni(x0,y0)ni(x1,y1))avg=∫
NPS(x0−
x,y0−
y,x1−
x,y1−
y,x,y)dxdNPS is approximated by;
∫
NPS(x0−
x,y0−
y,x1−
x,y1−
y,x,y)dxdy=nps(x0−
x1,y0−
y1)√
{square root over (noise1(x0,y0))}√
{square root over (noisei(x1,y1))}where NPS is the noise power spectrum correlation function normalized to 1 at 0 lp/mm.
-
-
8. The system as in claim 7 an “
- integrated Hounsfield”
measurement M and its noise free version M0 are is defined according to the equations set forth below;where Moi is a transform of raw measurements m0.
- integrated Hounsfield”
-
9. The system as in claim 8 wherein a first order Taylor expansion is used to bring a slowly varying white field term wi(x,y) outside of the convolution, where the difference of M and M0, i.e.
-
( x , y ) - M0 i ( x , y ) = ( m i ( x , y ) - m0 i ( x , y ) ) ⊗ ( psf - 1 ( x , y ) ) sec θ i ( x , y ) w i ( x , y ) A ′ ( M i ( x , y ) sec θ i ( x , y ) ) is a transformed noise term and where
-
-
10. The system as in claim 9 wherein wi(x,y)A′
- (Mi(x,y)secθ
i(x,y)) is the derivative of a signal versus thickness defining “
dsignal”
which can be calculated or measured and utilized to calculate;
mi(x,y)−
m0i(x,y)=[(M0i(x,y)−
Mi(x,y))dsignali(x,y)secθ
i(x,y)]{circle around (×
)}psƒ
(x,y)so that the optimization function set for the in claim 2 is expressed as;
- (Mi(x,y)secθ
-
11. The system as in claim 10 wherein the convolution in the middle of the last equation of claim 10 defines a detector quantum efficiency
dqe=psƒ- (nps−
1){circle around (×
)}psƒthat defines a ratio which represents a “
thickness noise”
squared and Δ
t2 is treated as a function only of thickness and can be calculated or measured.
- (nps−
-
12. The system as in claim 11 wherein a noise function
-
i ( x , y ) = noise i ( x , y ) dsignal i ( x , y ) sec θ i ( x , y ) - 1 = sec θ i ( x , y ) - 1 Δ t2 ( M i ( x , y ) sec θ i ( x , y ) ) and the last equation set forth in claim 10 is expressed as which is solved by calculating the derivative of χ
2 versus D(x0,y0,z0) and calculating where this derivative equals 0 to produce;
-
-
13. The system as in claim 12 where D is defined from a generator that is discrete, i.e.:
-
so that the equation set forth in claim 11 becomes;
-
-
14. The system as in claim 13 wherein the i-summation are identically assumed to be equal to zero, so that:
-
15. The system as in claim 14 wherein the equation in claim 14 is simplified by defining a function C as
-
( x , y ) = ∫ 0 Z max δ [ x + ( α i - α j ) z , y + ( β i - β j ) z ] ⅆ z wherein zmax in claim 14 equation is determined by the height of an object that is imaged, and the integral excludes absorption below the detector or above zmax in the solution so that the last equation of claim 14 is expressed as; which is solved for G, then D is solved from G and d is solved from D.
-
-
16. The system as in claim 15 wherein s in the equation set forth in claim 15 is assumed to be constant, and the equation is inverted using Fourier transforms wherein the convolutions become products
ℑ- (dqe)ℑ
(Mi)=ℑ
(dqe)ℑ
(Cij)ℑ
(Gj)+ε
σ
i2ℑ
(Gi)where ℑ
denotes a Fourier transform so that the equation is then expressed as
ℑ
(G1)=(ℑ
(dqe)ℑ
(Cij)+ε
σ
i2δ
ij)−
1ℑ
(dqe)ℑ
(Mi)and the inverses are performed individually over each spatial frequency.
- (dqe)ℑ
-
17. The system as in claim 16 wherein inverse matrices are computed once and stored as a look-up table for improved computational speed and stored data is required for each noise level and each value of zmax.
-
18. The system as in claim 15 wherein a non-constant s is assumed and multiple iterations are utilized with a constant σ
-
trial, solving equation the last equation in claim 12 using a Fourier transform method, calculating an error term, then iterating by reconstructing for the error term but using successively different values for the trial σ
value σ
trial, wherein an the error term is calculated as followsthe error term is inverted using a constant σ
dqe{circle around (×
)}(Cij{circle around (×
)}Δ
Gi)+ε
σ
trial2Δ
Gi=error,then a reconstructed generator coefficients are updated
Gi→
Gi+Δ
Githen the process is repeated with a different value of σ
trial so that eventually, the “
error”
becomes sufficiently small that solution is obtained.
-
trial, solving equation the last equation in claim 12 using a Fourier transform method, calculating an error term, then iterating by reconstructing for the error term but using successively different values for the trial σ
-
19. The system as in claim 18 wherein boundary conditions are handled by reflection conditions with tapering and a minimization function is:
-
so the first term equal (# of tube positions)×
(# of pixels) when ε
is set correctly.
-
Specification