CN103455989A - Method for improving imaging quality of limited-angle CT through combination of ultrasonic image - Google Patents

Abstract

The invention discloses a method for improving the imaging quality of limited-angle CT (Computed Tomography). The method comprises the following steps that an X-ray emitter and an X-ray receiver are arranged above target tissue and below the target tissue respectively; the emitter is made to move around the tissue and emit X rays within a limited angle, and meanwhile, projection data are collected; primary reconstruction is conducted by the utilization of a synchronous algebraic reconstruction technology; a three-dimensional ultrasonic image of the tissue is obtained; registering is conducted on the ultrasonic image through the combination of a primary CT reconstruction result; effective gradient information of the ultrasonic image is extracted, an improved synchronous algebraic reconstruction technology is used for conducting reconstruction again and an limited-angle CT imaging result with quality improved is obtained. According to the method for improving the imaging quality of the limited-angle CT, the characteristic that ultrasonic imaging is clear at the edge of the z direction is utilized, the problems that the limited-angle CT reconstruction result is serious in image artifact and fuzzy in boundary in the z direction, and innovation is achieved.

Description

A kind of method that improves limited angle CT image quality in conjunction with ultrasonoscopy

Technical field

The present invention relates to biomedicine and image processing field, particularly a kind of method that improves limited angle CT (Computed Tomography, computed tomography) image quality in conjunction with ultrasonoscopy.

Background technology

The three-dimensional structure that traditional CT imaging technique is rebuild interior of articles according to the different piece of destination organization to different these characteristics of the attenuation coefficient of X ray.The method is mainly to utilize cone-beam type X ray transmitter loop around 360 ° of destination organizations, and every certain angle emission primary X-ray, and with the X ray receiver, the another side at destination organization receives when each emission, thereby obtains some groups of data for projection.The method of rebuilding the 3-D view that obtains destination organization from these data for projection has a lot, wherein a kind of of comparative maturity is synchronous algebraic reconstruction technique (Simultaneous Algebraic Reconstruction Technique, below be abbreviated as SART).The inside that this technology can effectively reconstruct destination organization forms, and have advantages of that resolution is high, edge clear, noise be little.Yet when reality is used, restriction due to factors such as reconstruction time, x-ray doses, from 360 °, target object being scanned to this condition comprehensively often is difficult to be met, the substitute is emission and the reception of carrying out primary X-ray in a certain limited angle scope every certain angle, namely limited angle CT imaging technique.This technology makes detector plane and destination organization keep static, in a limited angle directly over destination organization, makes X ray transmitter surrounding target organize arc to move, thereby collects some data for projection.Yet compare with full angle CT imaging, a large amount of reductions due to data volume, limited angle CT imaging results (hereinafter referred to as the z direction) artifact on the direction perpendicular to receiver is serious, resolution is low, very large with the difference of the rebuilt tissue of reality, greatly affected the effective utilization of this technology in each field, needed badly to optimize and improve.

Three-dimensional ultrasound pattern is rebuild a kind of supplementary and auxiliary as what radioscopic image was rebuild, its result is high in the resolution of z direction, edge clear, and artifact than CT imaging results a little less than, exactly can make up the deficiency that limited angle CT is imaged on the z direction, effectively improve the quality of limited angle CT imaging results in the z direction.By this method after the marginal information of substitution ultrasonoscopy, can become under the prerequisite of image advantage retaining X ray, greatly improve its picture quality, be significant for the practice of limited angle CT imaging technique.

Summary of the invention

Goal of the invention: technical matters to be solved by this invention is unintelligible at z direction coboundary for limited angle CT imaging results, and the problem that artifact is serious provides a kind of ten thousand methods that improve limited angle CT image quality in conjunction with ultrasonoscopy.

In order to solve the problems of the technologies described above, the invention discloses a kind of method that improves limited angle CT image quality in conjunction with ultrasonoscopy, comprise the following steps:

Step 1, at upper and lower X ray transmitter and the receiver placed respectively of destination organization;

Step 2 makes the transmitter winding knit to move and launches X ray, recording projection data simultaneously in limited angle;

Step 3, utilize the SART algorithm to rebuild for the first time;

Step 4, obtain the three-dimensional ultrasound pattern of tissue;

Step 5, in conjunction with first CT reconstructed results, to ultrasound image registration;

Step 6, the effective gradient information of extraction ultrasonoscopy;

Step 7, used improved SART algorithm again to rebuild, and obtains the limited angle CT imaging results that quality improves.

In the present invention, preferably, X ray transmitter in described step 1 emission cone-beam type X ray, the size that destination organization is pressed close to the face of X ray receiver should be less than the receiver size, makes X ray by the thing piece substantially all be received device received.The X ray transmitter be positioned at receiver directly over, and reach certain vertical range with receiver, make X ray coverage goal tissue fully.

In the present invention, preferably, the subpoint of X ray transmitter on the receiver plane of take in described step 2 is rotation center, the initial position of transmitter of take is the centre of motion, the straight line of crossing this rotation center and being parallel to a certain limit of receiver is axle, and the arc that the X ray transmitter is done to be limited in scope around this axle in the plane perpendicular to receiver moves.The central angle that this scope is right by arc means, i.e. the determined central angle of left and right frontier point of rotation center and transmitter moving range, and its middle left and right frontier point is about centre of motion symmetry.In gatherer process, the X ray transmitter is from the centre of motion, first to a certain frontier point, move, the certain angle of every movement is launched X ray to receiver, and receiver receives backsight for collecting one group of data for projection, after arriving the frontier point of this direction, transmitter is retracted to the centre of motion, to another frontier point, move again, through same emission, receiving course, after transmitter arrives another frontier point, gatherer process finishes, and collects some groups of data for projection.

In the present invention, preferably, use the SART reconstruction algorithm in described step 3, reconstruction regions is carried out to projection repeatedly-back projection's iteration, the value of measured actual projection value correction reconstruction regions voxel in the projection value that constantly utilization is estimated and step 2, finally make result meet optimal condition, obtain first limited angle CT imaging results.

In the present invention, preferably, in described step 4 sonac respectively the destination organization upper surface and lower surface moves and to the destination organization internal emission, receive ultrasound wave, thereby obtain upper surface and two groups of ultrasound imaging datas of lower surface.The moving range of sonac will cover whole target tissue region, so that the registration of limited angle CT imaging results and ultrasonic imaging result.

In the present invention, preferably, in described step 5, because the characteristics of 3-D supersonic imaging are to launch along ultrasound wave, receive direction (being the z direction) carries out imaging, the image that namely become series of parallel in the plane of xOz or yOz (without loss of generality, this sentences three-dimensional ultrasound pattern and is reconstituted in the plane that is parallel to xOz as example), and the spatial resolution along the y direction is very low, therefore when carrying out registration, at first according to the first result of rebuilding of limited angle CT, adopt the method for linear interpolation or rejecting inactive area, adjust the size and the position that are parallel to the plane endosphere of xOz in the three dimentional reconstruction result, make in these planes the lump size in the CT reconstructed results in the lump in ultrasonic reconstructed results and respective planes, position consistency.Then the method by linear interpolation expands ultrasonic reconstructed results on the y direction, increase the data volume of three-D ultrasonic result in the y direction, thereby make the three-D ultrasonic result in the plane that is parallel to yOz also reach consistent with lump position, the size of CT imaging results, finally suitably adjust again the imaging region of CT imaging, make two kinds of reconstructed results reach registration.

In the present invention, preferably, described step 6 is chosen the gradient information of ultrasonic imaging result within being parallel to the plane of xOz, this gradient has x and z both direction, to each voxel, the value that its Grad in the x-direction is expressed as this voxel deducts the difference of its right-hand adjacent voxels, and the value that Grad in the z-direction is expressed as this voxel deducts the difference of its top adjacent voxels.The image of low quality for some ultrasonic imaging, according to above method, to every image that is parallel to the xOz plane, ask the gradient profile noise after gradient large, the lump edge fog, boundary line is thick, can under the prerequisite that does not affect general image information, to the ultrasonic imaging result, use the methods such as medium filtering, full variation optimization, also can carry out to the gradient image calculated medium filtering or simple morphology processing, thereby obtain little, the sharp-edged gradient image of noise.

In the present invention, expansion and optimization that described step 7 is carried out traditional SART reconstruction algorithm from the angle of loss function, the loss function that traditional SART algorithm is used is least square function

its iterative formula

that solve is exactly the optimum solution x that makes L (x) minimum.The loss function of choosing in the present invention adds the ultrasonic gradient information obtained in step 6, the gradient information that namely utilizes 3-D supersonic imaging, at the gradient information of x and z both direction, limited angle CT is imaged on to x and z both direction is revised, and its loss function is $L\left(x\right)=\frac{1}{2}{\left|\right|b-\mathrm{Ax}\left|\right|}_{\mathrm{<figure-callout\; id="2"\; label="w"\; filenames="HSA0000095446480000041.png"\; state="\{\{state\}\}">w</figure-callout>}}^2+\frac{{\mathrm{\&theta;}}_1}{2}\left|\right|G_{\mathrm{us}1}-{\mathrm{<figure-callout\; id="1"\; label="G\; X"\; filenames="HSA0000095446480000041.png"\; state="\{\{state\}\}">G</figure-callout>}}_X{\left|\right|}^2+\frac{{\mathrm{\&theta;}}_3}{2}\left|\right|G_{\mathrm{us}3}-G_{X3}{\left|\right|}^2,$ By solving this optimization problem, the iterative formula of deriving after being optimized.

The accompanying drawing explanation

Below in conjunction with the drawings and specific embodiments, the present invention is done further and illustrates, above-mentioned and/or otherwise advantage of the present invention will become apparent.

Fig. 1 is the process flow diagram of the inventive method.

Fig. 2 is a kind of schematic diagram of limited angle CT projection gatherer process.

Fig. 3 is a kind of schematic diagram of 3-D supersonic imaging.

Fig. 4 is the schematic diagram of asking gradient.

Fig. 5 is the label schematic diagram to the destination organization voxel.

Embodiment:

The present invention be take traditional SART method for reconstructing as basis, and by adding the effective gradient information of ultrasonic reconstructed image, having solved traditional limited angle CT, to be imaged on the z directional image of low quality, and the problem that resolution is low has improved the quality of reconstructed image.

As shown in Figure 1, the invention discloses a kind of method that improves limited angle CT image quality in conjunction with ultrasonoscopy, comprise the following steps:

Step 1, at upper and lower X ray transmitter and the receiver placed respectively of destination organization;

Step 2 makes the transmitter winding knit to move and launches X ray, recording projection data simultaneously in limited angle;

Step 3, utilize the SART algorithm to rebuild for the first time;

Step 4, obtain the three-dimensional ultrasound pattern of tissue;

Step 5, in conjunction with first CT reconstructed results, to ultrasound image registration;

Step 6, the effective gradient information of extraction ultrasonoscopy;

Step 7, used improved SART algorithm again to rebuild, and obtains the limited angle CT imaging results that quality improves.

In the present invention, step 1, destination organization is a long 22.225cm, wide 12.7cm, the rectangular parallelepiped thing piece of high 5.08cm, comprise some spherical lumps that differ in size and rhombus lump in the thing piece; The long 23.04cm of X ray receiver, wide 19.02cm, receiver is comprised of 1152 * 951 receiving sensor unit, and each unit is a square that the length of side is 0.02cm, and the one group of data for projection therefore at every turn received is the picture that a width resolution is 1152 * 951.This thing piece is positioned between X ray transmitter and X ray receiver, presses close to X ray receiver plane, the gap of 0.65cm is arranged between the bottom surface of thing piece and receiver plane.The X ray transmitter is positioned at directly over the thing piece, with the vertical range of receiver be 90cm, make the X ray of transmitter emission can substantially cover whole destination organization.The placement schematic diagram of a kind of X ray transmitter, destination organization, X ray receiver as shown in Figure 2.

In the present invention, step 2, as shown in Figure 2, X ray transmitter initial position (centre of motion) is at a P, and P ' is the projection of P in the receiver plane.Point P is positioned at a determined plane of side OADG of destination organization, using the plane of movement of this plane as transmitter, take P ' as the center of circle, the X ray transmitter is moved as arc around a P ', the right central angle of this arc is 60 °, the central angle angle that frontier point L and frontier point R are right to the centre of motion is all 30 °, and these two frontier points are about centre of motion symmetry.X ray is from the P of the centre of motion, first mobile to the L point, and every 3 °, to X ray receiver emission primary X-ray, the data that receiver receives are one group of projection value on this angle.After X ray transmitter motor point L, transmitter is moved back to centre of motion P, and start to move to another frontier point R, similarly, every through 3 ° of emission and receptions of carrying out primary X-ray, until transmitter point of arrival R, gatherer process finishes, and collects altogether 21 groups of data for projection.

In the present invention, step 3, the SART reconstruction algorithm is mainly to ask the solution x of such class equation by a kind of method of iteration:

Ax=b (1)

In formula (1), A is M * N matrix, refers to the system matrix of projective transformation in process of reconstruction, and x is N * 1 column vector, the value of all voxels that will rebuild, consists of, and represents the attenuation coefficient of each voxel, the number that N is reconstructed voxel; B is M * 1 column vector, represents projection value, and M is the total number of the projected pixel that collects.In the situation that limited angle CT imaging, the value of M is generally much smaller than N, so this equation is underdetermined equation, has countless many groups to separate.The main thought of SART algorithm is projection and back projection, first to the given initial estimate of x

then this estimated value is carried out to the estimated value that orthogonal projection obtains projection column vector b

again with the projection value b measured under first angle ₁with

poor

as the modified value correction be back projection's process, it is right that all 21 groups of projection values are completed once

correction be considered as iteration one time, final

converge to and make || b-Ax||w ²minimum x, i.e. least square solution.According to the SART theory, the introducing of weighting matrix W can reduce artifact effectively.

Understand the SART algorithm from the angle that solves optimization problem, namely requirement makes the value of the x of following loss function minimum, as shown in formula (2):

$L\left(x\right)=\frac{1}{2}{\left|\right|b-\mathrm{Ax}\left|\right|}_w^2---\left(2\right)$

Utilize gradient descent method to solve and make the minimum x of L (x) obtain the iterative formula of SART algorithm:

$x_j^{(t+1)}=x_j^{\left(t\right)}+\mathrm{\&lambda;}\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\frac{A_{\mathrm{ij}}}{A_{+,j}}\&CenterDot;\frac{b_j-\left({\mathrm{Ax}}^{\left(t\right)}\right)}{A_{i,+}}---\left(3\right)$

Wherein, x _j ^(t+1)mean the value that in voxel, j is ordered after (t+1) inferior iteration, the sequence number that i is got projection value, A _{+, j}, A _{i ,+}value be respectively:

$A_{+,j}=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\left|A_{\mathrm{ij}}\right|,A_{i,+}=\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}\left|A_{\mathrm{ij}}\right|(i=\mathrm{1,2},...,M;j=\mathrm{1,2},...,N)---\left(4\right)$

B _ibe the value of i subpoint, (Ax ^(t)) _ibe the value of i the subpoint of estimating, λ is relaxation factor, has influence on speed of convergence and reconstructed results.In order to guarantee convergence, in this example, the value of λ is taken as 0.1, and total iterations is taken as 3 times.

As shown in Figure 5, by using the SART algorithm, the first reconstruction obtains limited angle CT imaging results, the three-dimensional dimension of this reconstructed results is m * n * l, in this example, corresponding size is 1008 * 524 * 430, this each voxel of rebuilding space is 0.02cm in the length of x and y direction, at z direction height, be 0.01cm, be that corresponding physical size is 20.16cm * 10.48cm * 4.3cm, its bottom surface is apart from receiver plane 1.57cm, this undersized is in the physical size of destination organization, but all lump information in can the inclusion piece in reconstruction regions.

In the present invention, step 4, the 3-D supersonic imaging schematic diagram as shown in Figure 3, arrow wherein means the sonac emission, receives hyperacoustic process, by sonac respectively in translation, the scanning process on upper and lower two surfaces of destination organization, obtain two groups of three dimentional reconstruction results of destination organization, two groups of such results are to rebuild resolution near the direction of sensor separately high, and lump is clear.Due to these two groups as a result some zone can overlap, so can be by a series of mark and fusion treatment, thereby these two groups of results are merged and are obtained complete three dimentional reconstruction result.

The result that three dimentional reconstruction obtains is high at the plane intrinsic resolution that is parallel to xOz, edge clear, but low along the y directional resolution.In this example, the ultrasonic imaging result is of a size of 192 * 489 * 498, and wherein 192 is y direction size, and 489 is x direction size, and 498 is z direction size.

In the present invention, step 5, because limited angle CT imaging and ultrasonic imaging are two kinds of different imaging patterns, so will inevitably there be certain difference in the result obtained, especially aspect the size of the two.Yet because these two kinds of methods are again the reconstructions to the same destination organization, these two kinds of results must all have certain linear ratio relation with original destination organization so, accordingly, also certainly existing the linear ratio relation between these two kinds of results, this is just the theoretical foundation that provided of registration work.

The core purpose of registration is size and the position of corresponding lump in two kinds of reconstructed results of adjustment.Known with reference to step 3 and step 4, in this example, the area size of limited angle CT reconstructed results is 1008 * 524 * 430, and the 3-D supersonic imaging result is of a size of 192 * 489 * 498, as can be seen here, in CT result and ultrasonic experiments in the y-direction in the much the same situation of physical length, ultrasonic reconstructed results is few more a lot of than the CT image in the number of plies of y direction, therefore ultrasonic experiments is not high at the y directional resolution, need to carry out interpolation to it, makes the number of plies be expanded.The first step of registration work has been the registration with the plane of xOz plane parallel, in the CT process of reconstruction, the height of each voxel is voxel in half of x direction and y direction length, therefore at first ultrasonic reconstructed results to be carried out to linear interpolation in the z direction, make it highly become original twice; Then by size and the position of limited angle CT and corresponding lump in ultrasonoscopy, determine a linear scaling of two kinds of reconstructed results, to ultrasonic imaging, each is parallel to the plane of xOz according to this proportional linearity stretching as a result.Because ultrasonic imaging is rebuild whole thing piece, therefore reconstruction regions is greater than the X ray reconstruction regions, so need to reject to the result after stretching the inactive area (being that the CT imaging does not have the zone rebuild) of surrounding, thereby obtain being of a size of 192 * 524 * 430 ultrasonic experiments.Second step be to the three-D ultrasonic result being expanded of y direction, according to size and the position of lump on y direction in limited angle CT imaging, determine the stretch ratio of ultrasonic experiments in the y direction, according to this ratio, ultrasonic experiments is stretched in the y direction; Similarly, stretch and finish the rear inactive area of suitably rejecting both sides, obtain good 1008 * 524 * 430 the three-dimensional ultrasound pattern that is of a size of of complete registration.

In the present invention, step 6, ultrasonic imaging with the parallel plane plane of xOz inward flange clear (with reference to Fig. 5), therefore this is also the advantage place of ultrasonic imaging, gets lump gradient information in these planes and adds in the middle of the SART algorithm after improvement as optimization information.While choosing gradient information, under the prerequisite of lump main information, the ultrasonic imaging result is carried out to some image processing (as medium filtering, full variation optimization process) in not affecting ultrasonic imaging, the edge image calculated has been carried out to medium filtering and morphology processing, thereby made extracted edge clear, level and smooth.

Calculate ultrasonic reconstructed image when the gradient of x direction and z direction, its computing method are consistent with the method for compute gradient in the CT reconstructed results.In this example, the computing method of gradient are: the voxel value that the Grad of a certain tissue points is expressed as this point deducts the voxel value that put on its right (x direction) or top (z direction), as shown in Figure 3, and some f (x ₀, z ₀) at the Grad of x direction, be expressed as:

G _x(x ₀，z ₀)=f(x ₀，z ₀)-f(x ₀+1，z ₀) (5)

Similarly, some f (x ₀, z ₀) at the Grad of z direction, can be expressed as:

G _z(x ₀，z ₀)=f(x ₀，z ₀)-f(x ₀，z ₀+1) (6)

In formula (5), (6), G means the gradient information of single voxel value, is a scalar, and subscript x and z mean the Grad of x direction and z direction.

The right side one row pixel and top line pixel for every picture that is parallel to the xOz plane, need to be set to its Grad zero.

In the present invention, step 7, be optimized traditional SART algorithm, and added on the basis of its loss function (2) and comprised the loss item of ultrasonic imaging result in x direction and z directional information, as formula (7):

$L\left(x\right)=\frac{1}{2}{\left|\right|b-\mathrm{Ax}\left|\right|}_{\mathrm{<figure-callout\; id="2"\; label="w"\; filenames="HSA0000095446480000041.png"\; state="\{\{state\}\}">w</figure-callout>}}^2+\frac{{\mathrm{\&theta;}}_1}{2}\left|\right|G_{\mathrm{us}1}-{\mathrm{<figure-callout\; id="1"\; label="G\; X"\; filenames="HSA0000095446480000041.png"\; state="\{\{state\}\}">G</figure-callout>}}_X{\left|\right|}^2+\frac{{\mathrm{\&theta;}}_3}{2}\left|\right|G_{\mathrm{us}3}-G_{X3}{\left|\right|}^2---\left(7\right)$

In formula (7), θ ₁, θ ₃be used for controlling the affect size of gradient information on reconstructed results.G means gradient, and overstriking is expressed as vector.For avoiding confusion, 1 in subscript, 3 means respectively x direction and z direction, and subscript us means ultrasonic imaging, and subscript X means limited angle CT imaging.G _us1, G _us3be in the ultrasonic reconstructed results calculated in step 6 a little at the Grad of x direction and z direction, G _x1, G _x3in calculated in advance, do not obtain, but calculate in iterative process, can be expressed as formula (8), (9):

G _x1=Bx (8)

G _x3=Cx (9)

Wherein, x is the voxel value of unknown voxel to be reconstructed, is the N dimensional vector, number of voxels that N will rebuild.Voxel in rebuild destination organization is carried out to label according to following rule: the voxel at initial point place of take is first voxel x ₁from initial point (x=0, y=0, z=0) start, first increase in the y-direction the sequence number of voxel, forward next line (x=1) while reaching end to and continue counting, by that analogy, move to next plane (z=1) continuation counting after counting an xOy plane, count so all tissue points, thereby the sequence number of each voxel in destination organization is converted into to a dimension coordinate from three-dimensional coordinate, as shown in Figure 5, wherein m means the size of reconstruction regions in the y direction, and n means the size of reconstruction regions in the x direction, and l means the size of reconstruction regions in the z direction.

Matrix B, C in formula (8), (9) is used for calculating the gradient of x direction and z direction, the same with the method for asking gradient in step 6, the Grad of voxel also is expressed as the poor of this voxel and its right (x direction) or top (z direction) adjacent voxels value herein, therefore, matrix B, C can be expressed as respectively:

$B=\left(\begin{array}{c}{B}_1\\ {\mathrm{<figure-callout\; id="2"\; label="B"\; filenames="HSA0000095446480000041.png"\; state="\{\{state\}\}">B</figure-callout>}}_2\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ B_j\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ B_N\end{array}\right)---\left(10\right)$

$C=\left(\begin{array}{c}{C}_1\\ {\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="HSA0000095446480000041.png"\; state="\{\{state\}\}">C</figure-callout>}}_2\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ C_j\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ C_N\end{array}\right)---\left(11\right)$

Wherein, B _j(j=1,2 ..., N), C _j(j=1,2 ..., be N) all N dimension row vector, meet respectively following two conditions:

1, B _j: when j ∈ plane CBEF (as Fig. 5), B _j=0; j

during the CBEF of plane, B _jj element be that the individual element of 1, the j+ (m+1) is-1, all the other elements are all 0, make B _jx=x _j-x _{j+ (m+1)}, namely try to achieve voxel x _jgradient in the x direction.

2, C _j: when j ∈ plane DEFG when (as Fig. 5), C _j=0; j

during the DEFG of plane, C _jj element be 1, the j+ (m+1) (n+1) individual element be-1, all the other elements are all 0, make C _jx=x _j-x _{j+ (m+1) (n+1)}, namely try to achieve voxel x _jgradient in the z direction.

By formula (8), (9) substitution formula (7), obtain following loss function:

$L\left(x\right)=\frac{1}{2}{\left|\right|b-\mathrm{Ax}\left|\right|}_{\mathrm{<figure-callout\; id="2"\; label="w"\; filenames="HSA0000095446480000041.png"\; state="\{\{state\}\}">w</figure-callout>}}^2+\frac{{\mathrm{\&theta;}}_1}{2}\left|\right|G_{\mathrm{us}1}-\mathrm{Bx}{\left|\right|}^2+\frac{{\mathrm{\&theta;}}_3}{2}{\left|\right|G_{\mathrm{us}3}-\mathrm{Cx}\left|\right|}^2---\left(12\right)$

Utilize gradient descent method to ask the solution x that makes L (x) minimum meeting under the condition of Ax=b, to formula (12), ask gradient to obtain:

Known according to gradient descent method, separating x can be obtained by following formula iteration:

According to the SART algorithm, ω is taken as λ, and adds coefficient V ^-1, wherein, V is a diagonal matrix:

Weighting matrix W is also diagonal matrix:

A in W and V _{i ,+}a _{+, j}as shown in formula (4).

In the present invention, by ω θ ₁, ω θ ₃be set as constant λ ₁, λ ₃, in order to guarantee convergence, in this example, by these two setting parameters, be 0.03, also should be according to concrete Grad size to λ in other implementation condition ₁, λ ₃carry out some tests and adjust.Therefore, can obtain following formula:

x ^(t+1)=x ^(t)+λV ^-1A ^TW(b-Ax ^(t))+λ ₁B ^T(G _us1-Bx ^(t))+λ ₃C ^T(G _us3-Cx ^(t))(17)

Formula (17) is launched, just obtains the iterative formula for each tissue points:

$x_j^{(t+1)}=x_j^{\left(t\right)}+\mathrm{\&lambda;}\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\frac{A_{\mathrm{ij}}}{A_{+,j}}\&CenterDot;\frac{b_i-\left({\mathrm{Ax}}^{\left(t\right)}\right)}{A_{i,+}}+{\mathrm{\&lambda;}}_1{P}_j\left(x_j^{\left(t\right)}\right)+{\mathrm{\&lambda;}}_3{Q}_j\left(x_j^{\left(t\right)}\right)---\left(18\right)$

Wherein,

shown in (19), (20):

Wherein, Gu _{s1, j}mean j tissue points Grad in the x-direction in ultrasonic imaging, G _{us3, j}mean j tissue points Grad in the z-direction in ultrasonic imaging.G _{x1, j}the j tissue points Grad in the x-direction that means reconstructed results after the t time iteration, G _{x3, j}mean the j tissue points Grad in the z-direction of reconstructed results after the t time iteration, can be expressed as:

${\mathrm{<figure-callout\; id="1"\; label="G\; X"\; filenames="HSA0000095446480000041.png"\; state="\{\{state\}\}">G</figure-callout>}}_X=x_j^{\left(t\right)}-x_{j+(m+1)}^{\left(t\right)}---\left(21\right)$

$G_{X3,j}=x_j^{\left(t\right)}-x_{j+(m+1)(n+1)}^{\left(t\right)}---\left(22\right)$

Gradient information is updated in the middle of the SART algorithm after improvement, again rebuilds and just can obtain the limited angle CT reconstructed results that z directional image quality improves.It is to be noted, in the process of again rebuilding, because limited angle CT rebuilds, need to carry out iteration 3 times, in each iterative process, each tissue points be carried out to 21 times and upgrade, namely each tissue points is carried out altogether to 63 times and upgrade, this can not make the gradient information convergence added.Make the gradient information convergence, need to carry out the approximately iteration of 1000 times.Therefore, in the present invention the iterative process of gradient information is separated with the SART iterative process, after each group projection information is carried out to the SART iteration, usining this iteration result carries out the iteration of 15 subgradient information again as initial value, as the iterative process to one group of data for projection, all 21 groups of data for projection are completed to calculate as once complete iterative process and finish like this.After carrying out like this 3 iteration, each subpoint has carried out the iteration of 63 * 15=945 subgradient information, can guarantee the convergence of reconstructed results.

The invention provides a kind of method that improves limited angle CT image quality in conjunction with ultrasonoscopy; should be understood that; for those skilled in the art; under the premise without departing from the principles of the invention; can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.In addition, in the present embodiment not clear and definite each ingredient all available prior art realized.

Claims (8)

1. a method that improves limited angle CT (Computed Tomography, computed tomography) image quality in conjunction with ultrasonoscopy, is characterized in that, comprises the following steps:

Step 1, at upper and lower X ray transmitter and the receiver placed respectively of destination organization;

Step 2 makes the transmitter winding knit to move and launches X ray, recording projection data simultaneously in limited angle;

Step 3, utilize SART (Simultaneous Algebraic Reconstruction Technique, synchronous algebraic reconstruction technique) algorithm to rebuild for the first time;

Step 4, obtain the three-dimensional ultrasound pattern of tissue;

Step 5, in conjunction with first CT reconstructed results, to ultrasound image registration;

Step 6, the effective gradient information of extraction ultrasonoscopy;

Step 7, used improved SART algorithm again to rebuild, and obtains the limited angle CT imaging results that quality improves.

2. a kind of method that improves limited angle CT image quality in conjunction with ultrasonoscopy according to claim 1, it is characterized in that, in described step 1, X ray transmitter distance X ray receiver will reach certain distance, makes the X ray of transmitter emission can substantially cover whole destination organization.

3. a kind of method that improves limited angle CT image quality in conjunction with ultrasonoscopy according to claim 1, it is characterized in that, in described step 2, the X ray transmitter moves as arc in the plane perpendicular to receiver around receiver, in limited angular range, launch X ray every certain angle to receiver, receiver receives the several times data and is some data for projection.

4. a kind of method that improves limited angle CT image quality in conjunction with ultrasonoscopy according to claim 1, it is characterized in that, described step 3 is utilized the SART algorithm, and the data for projection collected is carried out to three-dimensional reconstruction, obtains the first limited angle CT reconstructed results to destination organization.

5. a kind of method that improves limited angle CT image quality in conjunction with ultrasonoscopy according to claim 1, it is characterized in that, 3-D supersonic imaging in described step 4 is used the Bi-directional Ultrasonic image-forming mechanism, respectively destination organization is carried out to three dimentional reconstruction from the upper and lower surface of destination organization, then this twice reconstructed results carried out to mark and fusion, thereby obtain the three dimentional reconstruction result to destination organization integral body.

6. a kind of method that improves limited angle CT image quality in conjunction with ultrasonoscopy according to claim 1, it is characterized in that, not of uniform size the causing of lump in described step 5 in the three-dimensional dimension of 3-D supersonic imaging result and limited angle CT imaging results, effective reconstruction regions and respective regions, by the linearity adjustment to the ultrasonoscopy size, obtain the ultrasonoscopy after an assembly standard, the characteristics of this image are basically identical at the aspects such as three-dimensional dimension, effective coverage, lump position, lump size and CT reconstructed results.

7. a kind of method that improves limited angle CT image quality in conjunction with ultrasonoscopy according to claim 1, it is characterized in that, in described step 6, the ultrasonic imaging result after registration is extracted to effective gradient information, refer to each gradient information perpendicular to the picture on X ray receiver plane that extracts three-dimensional ultrasound pattern, this gradient information should comprise the gradient information of both direction in this plane.

8. a kind of method that improves limited angle CT image quality in conjunction with ultrasonoscopy according to claim 1, it is characterized in that, in described step 7, improved SART algorithm is on the basis of original SART algorithm, optimize its loss function, add the loss item that utilizes 3-D supersonic imaging gradient information optimized reconstruction result, and derivation obtains corresponding iterative formula.

Images