CN101739661B - Method for structuring ultrasound long axis image quickly with high fidelity - Google Patents
Method for structuring ultrasound long axis image quickly with high fidelity Download PDFInfo
- Publication number
- CN101739661B CN101739661B CN2009102423419A CN200910242341A CN101739661B CN 101739661 B CN101739661 B CN 101739661B CN 2009102423419 A CN2009102423419 A CN 2009102423419A CN 200910242341 A CN200910242341 A CN 200910242341A CN 101739661 B CN101739661 B CN 101739661B
- Authority
- CN
- China
- Prior art keywords
- coordinate
- straight
- image
- intersection point
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 29
- 238000002604 ultrasonography Methods 0.000 title claims abstract description 9
- 238000002608 intravascular ultrasound Methods 0.000 claims abstract description 34
- 239000012634 fragment Substances 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 16
- 239000000047 product Substances 0.000 claims description 11
- 238000000926 separation method Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 239000012467 final product Substances 0.000 claims description 2
- 230000015654 memory Effects 0.000 abstract description 14
- 238000013507 mapping Methods 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 5
- 238000005070 sampling Methods 0.000 description 5
- 238000010276 construction Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 210000004204 blood vessel Anatomy 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000007667 floating Methods 0.000 description 3
- HPTJABJPZMULFH-UHFFFAOYSA-N 12-[(Cyclohexylcarbamoyl)amino]dodecanoic acid Chemical compound OC(=O)CCCCCCCCCCCNC(=O)NC1CCCCC1 HPTJABJPZMULFH-UHFFFAOYSA-N 0.000 description 2
- 238000005481 NMR spectroscopy Methods 0.000 description 2
- 208000029078 coronary artery disease Diseases 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000013146 percutaneous coronary intervention Methods 0.000 description 2
- 238000002203 pretreatment Methods 0.000 description 2
- 238000009877 rendering Methods 0.000 description 2
- 239000000523 sample Substances 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000001464 adherent effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 210000004351 coronary vessel Anatomy 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 230000007170 pathology Effects 0.000 description 1
- 238000012797 qualification Methods 0.000 description 1
- 238000001356 surgical procedure Methods 0.000 description 1
- 230000002792 vascular Effects 0.000 description 1
Images
Landscapes
- Image Generation (AREA)
Abstract
The invention discloses a method for structuring ultrasound long axis image quickly with high fidelity. The method is implemented by using a parallel computing architecture based on a graphics processor to perform efficient management to three-dimensional IVUS data with large data volume by quadtree data structure, performing recursion traverse to the quadtree and computing the intersecting point coordinate of the plane on which the long axis image is located and the quadtree data node, decomposing the long axis image into a series of small planes, then sending to the graphics processor for mapping and interpolation computing. The invention solves the problem that the data volume is too large to be loaded display memory of the graphics processor, improves utilization efficiency of texture cache of the graphics processor, and enhances computation speed and image quality by computing the three-dimensional spline interpolation by using single instruction multidata vector dot product of the graphics processor. The invention can achieve above 30 frame per second in structuring long axis image with high fidelity on common middle/high-grade display card for large scale IVUS image data of 512*512*4000*16bit.
Description
Technical field
The present invention relates to the medical image processing technical field, relate in particular to a kind of method of quickly with high fidelity structuring ultrasound long axis image.
Background technology
In worldwide, coronary cardiopathy (abbreviation coronary heart disease) is to cause one of human dead first cause at present, and the percutaneous coronary intervention operation is the main means of treatment coronary heart disease.New recently intravascular ultrasound (IVUS) image that rises is considered to the best a kind of image mode of auxiliary percutaneous coronary intervention operation.Obtaining of IVUS image is minimally-invasive, by on conduit, adding a very small-sized high frequency ultrasound probe, by patient's blood vessel it is placed into the doctor and suspects have in the coronary artery of pathology, control by motor, the probe of at the uniform velocity pulling back, thus a series of blood vessel endoscope image obtained.
The doctor mainly is the major axis imaging method for the sheet mode of readding of IVUS image at present, suppose to have obtained altogether N frame image, the doctor can be with mouse definition straight line section AB (please referring to accompanying drawing 1) on the first frame image, because the center of IVUS image generally is doctor's interesting areas such as blood vessel, the high echo area that adds upper conduit, so straight-line segment AB is by the central point C's of image, the doctor passes through mouse action, can be around central point C rotation straight-line segment AB, until obtaining satisfied angle, this moment, system can carry out image sampling along straight-line segment AB on the first frame image, obtain line data and be mapped on the Line 1 of long axis image, A wherein, B, C is mapped to A ' respectively, B ', on the C ', because the size of N frame image all equates, so A, the coordinate of B can be applied directly on the 2nd～N frame image, by same process, can obtain the view data of the 2nd～N line of long axis image.The doctor can obtain information such as vascular wall, patch, support placement by observing and measuring long axis image, thereby assisted surgery or evaluation operation after-poppet discharge adherent situation.
Because the obtaining at the uniform velocity to pull back by motor control of IVUS image obtained, the speed of typically pulling back at present is 0.5mm/s, suppose to obtain the data of 5cm, needed so to gather 100 seconds, speed with 25 frames/s is calculated, obtain N=2500 frame image so altogether, the data volume of once obtaining in fact clinical often is N＞4000 frames.The resolution of every frame image generally is 512 * 512, each pixel 8bit or 16bit, so typical IVUS data volume is in clinical: 512 * 512 * 4000 * 16 ÷ 8=2000MB.So huge data volume has brought very big challenge for the quick structure of long axis image, existent method all is to have done compromise so that reach the purpose of real-time generation long axis image on picture quality at present, such as considering that N is far longer than present exploration on display resolution ratio, therefore when generating long axis image, can adopt the method for down-sampling, every some line computations once, can reduce calculated amount exponentially like this, a lot of but the quality of the long axis image that generates also descends thereupon.In addition, present algorithm is just sampled along straight line on a frame image, only adopted the two-dimensional interpolation of image, do not utilize consecutive frame to carry out three-dimensional interpolation, in order to pursue computing velocity, what image interpolation algorithm was used is arest neighbors interpolation algorithm or bilinear interpolation algorithm, and more the interpolation algorithm of high-order is too big because of operand, can't reach real-time computing velocity, therefore not be used at present.
Recent years, very big variation has taken place in the trend of high-performance calculation, Graphics Processing Unit (the GPU that the normal domestic video card is comprised, Graphics Processing Unit) computing power of chip is more and more stronger, the computing power that has little by little surmounted traditional central processing unit (CPU), and allow the developer to use the grammer of similar C language directly video card to be programmed, Cg, CUDA programming language as U.S. NVidia company, the HLSL programming language of Microsoft company, the GLSL programming language of OpenGL Internet Architecture Board etc.GPU is applied to general-purpose computations more and more at present, shown powerful computation capability in fields such as Flame Image Process, Video processing, 3d gamings, make the former algorithm that is very difficult to handle the in real time possibility that becomes of serving, such as the real-time volume drawing (Volume Rendering) of 3 D medical image data, speed that can 30 frames/s on present main flow video card is drawn the data set of 512 * 512 * 512 scales.The another one development trend is to utilize the GPU powerful computing ability, give elimination with the compromise of in order to improve render speed picture quality being done in traditional algorithm, use the computing method of full accuracy, thereby obtain the image of E.B.B., such as the Fovia company of the U.S. (
Http:// www.fovia.com/index.php) algorithm of high-fidelity volume drawing (High Definition Volume Rendering) was proposed in 2007, thus the picture quality that conventional bulk is drawn has improved a class.
Considering that GPU than the advantage of CPU calculating and in the success that obtains aspect the volume drawing, provides a kind of high-fidelity long axis image construction algorithm at big data quantity IVUS image based on GPU in IVUS image documentation equipment or software, is very significant.
Summary of the invention
The objective of the invention is to solve the not high enough problem of employed long axis image construction algorithm precision in present IVUS image documentation equipment or the software, for this reason, provide a kind of real-time big data quantity IVUS long axis image building method of high-fidelity.
For reaching described purpose, the method for a kind of quickly with high fidelity structuring ultrasound long axis image provided by the invention comprises that step is as follows:
Step 1: load three-dimensional intravascular ultrasound data, its width and highly be respectively W pixel, H pixel, the image frame number is N, neighbor x is x to physical separation
sMillimeter, y is y to physical separation
sMillimeter, z is z to physical separation between the adjacent two frame images
sMillimeter;
Step 2: obtain the straight line section AB that the user defines on the first frame image, straight-line segment AB is by the central point C of image, and the angle of straight-line segment AB and x axle is θ;
Step 3: carry out quaternary tree along the plane, the first frame image place of three-dimensional intravascular ultrasound data and evenly divide, make that the size of leaf node is M * M;
Step 4: quaternary tree is begun the recurrence traversal from root node, and handle each node, comprising:
Judge that whether present node intersects with straight-line segment AB, if present node and straight-line segment AB intersect and present node is not a leaf node, then continues the recurrence traversal; If present node and straight-line segment AB intersect and present node is a leaf node, then execution in step 5; Directly return father node if present node and straight-line segment AB are non-intersect and continue traversal;
Step 5: the x, the y coordinate that calculate intersection point F, the G of leaf node and straight-line segment AB, and obtain x, the y coordinate of corresponding point intersection point F ', G ' on the M+1 frame image simultaneously, make the x of intersection point F ', x, the y coordinate that the y coordinate equals intersection point F, the x of intersection point G ', y coordinate equal x, the y coordinate of intersection point G, and the z coordinate that the z coordinate of intersection point F ', G ' equals intersection point F, G respectively adds M * z
sDraw plane FF ' G ' G, simultaneously the data block at leaf node place is transferred to Graphics Processing Unit in the mode of three-D grain and handle; Said process is repeated
Inferior, each only the need adds M * z respectively with the z coordinate of intersection point F, G, F ', G '
s, corresponding blocks of data upgrades thereupon and gets final product;
Step 6: each fragment that in the fragment shader of Graphics Processing Unit drafting plane FF ' G ' G is produced is carried out three-dimensional SPL interpolation.
Advantage of the present invention is: according to technical scheme provided by the invention, use is based on the computing architecture of GPU, by the quaternary tree data structure three-dimensional IVUS data being carried out piecemeal handles, solve on the one hand data volume and can not all be loaded into problem in the GPU video memory too greatly, improve utilization ratio in addition on the one hand to the GPU texture cache, by using three-dimensional SPL interpolation, improved the quality of image greatly in addition.The GPU computing architecture that the present invention uses present normal domestic video card to provide can reach the above high-fidelity long axis image desin speed of frame p.s.s 30 to the so large-scale IVUS image data of 512 * 512 * 4000 * 16bit.Than existent method, the present invention does not use any image down sampling method to reduce image resolution ratio, and in GPU, calculate three-dimensional SPL interpolation fully, arest neighbors interpolation or bilinear interpolation than two dimension, higher computational accuracy can be obtained, still real-time computing velocity can be reached simultaneously.In addition, the invention provides and a kind ofly new three-dimensional data is carried out the space block division method,, can significantly reduce the amount of extra memory that needs, also can quicken the speed of traverse tree in addition than traditional method based on Octree based on quaternary tree.The method provided by the present invention powerful computation capability that fully digging utilization present age, GPU was provided, thus traditional IUVS long axis image construction algorithm is upgraded to the algorithm of high-fidelity of future generation, picture quality is promoted a class.
Description of drawings
Fig. 1 is the synoptic diagram of conventional I VUS long axis image structure.
Fig. 2 is the synoptic diagram of structure global coordinate system and eye coordinates system.
Fig. 3 is the synoptic diagram that the quaternary tree piecemeal is handled.
Fig. 4 is the synoptic diagram of judge specifying straight-line segment whether to intersect with certain leaf node.
Embodiment
Describe each related detailed problem in the technical solution of the present invention in detail below in conjunction with drawings and Examples.Be to be noted that described embodiment only is intended to be convenient to the understanding of the present invention, and it is not played any qualification effect.
With traditional method each frame being loaded three-dimensional intravascular ultrasound (IVUS) image handles different respectively, the present invention has adopted the method based on the 3-D view interpolation, therefore the volume data of all N frame images as a three-dimensional, IVUS long axis image structure is regarded as the sampling process of a three-dimensional planar to three-dimensional data, and this needs suitable coordinate system could guarantee correct calculating.
Accompanying drawing 2 has provided the organigram of global coordinate system and eye coordinates system, for convenience's sake, the true origin O of global coordinate system overlaps with the lower left corner of the 1st frame image, the planes overlapping at the xoy plane of global coordinate system and the 1st frame image place, x axle forward from left to right, y axle forward from bottom to top, other N-1 frame image is perpendicular to z axle series arrangement, z axle forward points to the N frame by the 1st frame, and the z between adjacent two frames is to distance: z
s=motor speed/the frame frequency of pulling back.Though N frame image disperses, can reconstruct a continuous three-dimensional volume data by image interpolation algorithm, therefore do not have each discrete frame of picture in the drawings, but used a continuous rectangular parallelepiped to represent the volume data that this is three-dimensional.The straight-line segment AB of doctor's appointment on the 1st frame can determine a plane ABDE perpendicular to z axial projection, along the sampling of ABDE uniform plane, just can construct long axis image.
Determined after the global coordinate system that also need to construct a suitable eye coordinates system, if inappropriate words are selected by eye coordinates system, the ABDE plane projection will be a parallelogram so, and does not reach the effect of traditional major axis structure to view plane.The initial point o ' that selection A point as eye coordinates is among the present invention, AB are as the x ' axle of eye coordinates system, and AE is as the y ' axle of eye coordinates system, and the z ' axle of eye coordinates system is according to left-hand rule, and selection is perpendicular to x ' o ' y ' plane and pass through o ' straight line.In addition, projection pattern is selected parallel projection, and view frustums is respectively by A, B, four summits of D, E.
On the basis of above-mentioned structure coordinate system, can there be a very simple method to realize the major axis structure of IVUS image: at first N frame IVUS image to be transferred to Graphics Processing Unit (GPU) video memory as a three-D grain, then respectively with the world coordinates of A, B, D, four points of E texture coordinate as them, by drawing plane ABDE, can utilize rasterization unit and trilinear texture interpolating unit in the GPU hardware to obtain the IVUS long axis image.The principle of multiplanar reconstruction (MPR) algorithm in the principle of this algorithm and traditional CT, MRI (nuclear magnetic resonance) image processing is identical, but it's a pity, also inapplicable in the IVUS of big data quantity long axis image structure, because the data volume of typical IVUS image is often greater than 2GB, and at present except the very expensive professional video card of minority price, the video memory of most of video card all is less than 2GB far away, at present the video memory of the middle and high end video card of main flow generally at 256MB between the 1GB, and can not all be used for three-D grain.
In order to address this problem, the fritter that among the present invention a big IVUS three-dimensional data is divided into several M * M * M, M is typically chosen in 2 integer power, and make the data of a fritter to be held by the three-D grain buffer memory of GPU fully as far as possible, because the texture cache of GPU hardware is often all smaller, therefore M can not be too big, if to such an extent as to the too big texture cache of M value can not comprise the data of a fritter, GPU is when doing calculating so, the slow texture video memory of access speed in large quantities, speed can be subjected to seriously influencing very much.In the present invention, M value 32, in the time of still concrete enforcement, the value that can suitably regulate M according to the size of used video card texture cache is to reach the fastest arithmetic speed.
After deblocking, can reduce a lot of unnecessary texture video memorys because in all data blocks and plane ABDE intersect only account for a seldom part, and just this part seldom need be transferred among the GPU, goes to do further calculating.Though solved the problem of video card video memory deficiency like this, but the another one problem is appeared in one's mind out again: face ABDE does to intersect and calculates because each data block all is desirous of peace, calculate according to M=32, data set for one 512 * 512 * 4000, the piece number that needs altogether is 32000, calculates asking of 32000 plane-rectangular parallelepipeds and hands over calculating to need very large calculated amount.The way that solves this class problem traditionally is to use Octree to carry out the piecemeal by different level of recurrence, at first whole data set is divided into equal-sized eight pieces from the space, each piece is the child node of his father's piece, to each piece eight equal parts recursively again, till block size is smaller or equal to M, just stop to divide then.When drawing, from root node, top-down traversal Octree, for each node, if it does not intersect with plane ABDE, its all child node also can not intersect with plane ABDE so, therefore can just discard incoherent on very high level very apace, significantly reduces to ask to hand over the piece number that calculates.
But there are some shortcomings in the method that is based on Octree: at first, Octree needs very big additional storage space, is example according to M=32 also, the leaf node of Octree is 32000, calculate with full Octree, add required intermediate node, need the storage space of 37000 nodes altogether approximately; Secondly, the recurrence of Octree traversal more complicated and consuming time; At last, use Octree just must carry out the cap of plane-rectangular parallelepiped, and this computing is a three-dimensional geometry computing, also needs certain complexity.
Provide among the present invention one new for quaternary tree recurrence block division method, quaternary tree can only be used for the recurrence of two-dimensional space and divide, why can be applicable to three-dimensional division, be owing to utilized the singularity of IVUS long axis image construction algorithm resulting fully.Please referring to accompanying drawing 3, can only on the 1st frame image, use recursively piecemeal of quaternary tree, the rectangular region of calculated line section AB and the pairing two dimension of each data block is asked friendship then, and the intersecting point coordinate of being tried to achieve can directly apply to the z next piece that makes progress, and need not the recurrence piecemeal that makes progress at z.Intersect at the F point as AB among the figure and certain piece, so vertical projection F o'clock to the 1+M frame, can obtain F ' point, its x, y coordinate and F point x, y coordinate are identical, only z coordinate difference: z
F '=z
F+ z
s* M, wherein z
F 'Be F ' z coordinate, z
FThe z coordinate of ordering for F.By such processing, the size of each piece still is M * M * M, only by the coordinate of traversal each intersection point that quaternary tree calculated, upwards can reuse at z
Inferior.
Use has following advantage based on the recurrence method of partition of quaternary tree: at first, the needed additional storage space of quaternary tree is very little, also uses the example of above-mentioned M=32, and the leaf node of quaternary tree only needs (512/32)
2=256, calculate with full quaternary tree, add required intermediate node, need 341 nodes altogether; Secondly, because the shared storage space of 341 nodes is very little, can be loaded at an easy rate among the L1 Cache (level cache) of CPU, the degree of depth of quaternary tree has only 4 in addition, and these two factors make that the small-sized quaternary tree efficient of traversal is very high; At last, use quaternary tree to avoid complicated three-dimensional planar-rectangular parallelepiped to ask friendship, ask friendship and only need calculate fairly simple two-dimentional straight-line segment-rectangle.
The pairing rectangle of leaf node that accompanying drawing 4 has provided calculated line section AB and any one quaternary tree is asked the synoptic diagram of friendship, and the coordinate on rectangular four summits is without loss of generality by the leaf node institute record of quaternary tree, suppose to be respectively (x
0, y
0), (x
1, y
0), (x
1, y
1), (x
0, y
1).In addition, the width of supposing a frame IVUS image is a W pixel, highly is H pixel (use maximum images at present clinical, its W and H are 512 pixels), and neighbor x is x to physical separation
s, y is y to physical separation
s, the angle of straight-line segment AB and x axle is θ, then the image center C coordinate of ordering is: (C
x, C
y)=(Wx
s/ 2, Hy
s/ 2)), the parametric equation of straight-line segment AB can be expressed as:
Wherein t is a parameter, and x (t), y (t) go up arbitrarily x, the y coordinate of any for straight-line segment AB; Itself and straight line x=x
0, x=x
1Intersection point be respectively: (x
0, C
y+ (x
0-C
x) tan θ), (x
1, C
y+ (x
1-C
x) tan θ), with straight line y=y
0, y=y
1Intersection point be respectively: (C
x+ (y
0-C
y) cot θ, y
0), (C
x+ (y
1-C
y) cot θ, y
1), then judge following four conditions:
If four conditions do not satisfy, straight-line segment AB and this leaf node are non-intersect so, if four conditions all satisfy, then straight-line segment AB and this leaf node intersect, and intersecting point coordinate also can be obtained by aforementioned calculation.When straight-line segment and rectangle intersect and straight-line segment during without one of rectangular four summits, there are and have only two to satisfy in above-mentioned four Rule of judgment, two points of F, G among this moment Fig. 4 lay respectively on rectangular two different limits; Only one of by rectangular four summits, when all non-intersect, have two to satisfy with other limit in above-mentioned four Rule of judgment when straight-line segment, the summit of coincidence must be arranged, remove the summit of coincidence, two points of this moment F, G coincide with crossing summit; When having three to satisfy in above-mentioned four Rule of judgment, must be that straight-line segment passes through rectangular certain summit at this moment, but pass through rectangular certain bar limit simultaneously again that one and crossing summit coincidence are arranged in two points of this moment F, G; When above-mentioned four Rule of judgment satisfy simultaneously, only may be straight-line segment by rectangular two to angular vertex, two points of this moment F, G are two crossing summits.No matter under above-mentioned which kind of situation, can calculate the x of F, two points of G, the y coordinate, and x, the y coordinate of two points of F ', G ' (referring to accompanying drawing 3) they are the x that equals F, G respectively, y coordinate, the z coordinate that its z coordinate is respectively F, G adds M * z
sObtain, the size that will comprise plane FF ' G ' G this moment is that the blocks of data of M * M * M transfers to GPU texture video memory as three-D grain, and in three dimensions, draw plane FF ' G ' G, and just finished processing to a data block, also obtained the part of whole long axis image.The x of above-mentioned F, G, y coordinate will reuse
Inferior, each z coordinate adds M * z
s, corresponding blocks of data also upgrades thereupon, handles the leaf node of a quaternary tree like this and asks friendship to calculate, and just can upwards handle at z
Individual data block by traveling through whole quaternary tree, just can be handled all
Individual data block, plane FF ' G ' G that each data block comprised, therefore the good whole plane ABDE that formed that is stitched together can obtain complete long axis image.
Above-mentioned quaternary tree traversal and cap are all finished in CPU, drawing plane FF ' G ' G is the rasterisation hardware that utilizes GPU, plane grating is changed into pixel into series of discrete, each pixel is called Fragment, record the information such as coordinate, color of oneself, the fragment shader (Fragment Shader) that can give GPU further processes.The Fragment Shader of GPU is a programmable device, can use language such as GLSL, Cg, HLSL or CUDA that it is programmed, to reach own specific calculating purpose.There are a plurality of Fragment Shader simultaneously in the GPU hardware at present, can handle a plurality of fragments (Fragment) concurrently, therefore can improve computing velocity greatly.Use the Fragment Shader of GPU that each pixel after the plane gratingization is carried out three-dimensional spline interpolation among the present invention and calculate, with the high-quality that guarantees image simultaneously and the high-performance of computing.
In the Fragment of GPU Shader, can obtain screen coordinate (x when the pixel of pre-treatment
f, y
f, z
f), and viewpoint change projection matrix M
Mvp, can calculate as the pairing world coordinates (x of the pixel of pre-treatment by them
g, y
g, z
g), computing formula is as follows:
[x
g，y
g，z
g]
T＝M
mvp -1[x
f，y
f，z
f]
T
Because image interpolation need carry out, therefore also need the coordinate (x in computational data space in data space
v, y
v, z
v), computing formula is as follows:
[x
v，y
v，z
v]
T＝[x
g/x
s，y
g/y
s，z
g/z
s]
T
And satisfy:
The x that calculates by above-mentioned coordinate transform
v, y
v, z
vInteger not necessarily, but the IVUS image data only provides W * H * N discrete sampled value, therefore needs to use image interpolation algorithm obtain any floating number (x
v, y
v, z
v) sampled value at coordinate place.Traditional algorithm often adopts arest neighbors interpolation or bilinear interpolation algorithm in order to guarantee computing velocity, and the picture quality that obtains is poor, the higher three-dimensional SPL interpolation algorithm of service precision among the present invention, and its computing formula is:
Wherein I (x, y, z) be three-dimensional IVUS image data on the z frame (x, the y) pixel value at coordinate place, f are three-dimensional spline surface function, and be simple in order to calculate, and generally selects f to make it satisfy:
f(x，y，z)＝f(x)·f(y)·f(z)
Above-mentioned like this three-dimensional interpolation computing can be separated into three independent one dimension interpolation arithmetics, be shown below:
By such decouples computation, can be the SPL function evaluation of one dimension with the spline surface function evaluation abbreviation of three-dimensional, and can calculate the interpolation of one dimension according to the such order of the last frame of Row Column, significantly reduced calculated amount.Even but like this, interpolation calculation to a Fragment still needs multiplication 672 times, 64 internal storage access, if required data are not in the buffer memory of CPU, internal storage access is an operation that cost is very high so, since the high capacity of IVUS image data, and interpolation all needs to visit the interior data of continuous four frames at every turn, so cpu cache mechanism lost efficacy under most of situation.The size of supposing the IVUS image is 512 * 512 * 4000, in order with high fidelity to construct long axis image, its resolution at least should for:
That is to say and comprise at least 2896310 Fragment, and in order to reach real-time renewal speed, at least should be able to upgrade 30 one second, therefore the multiplication number that needs for 1 second is 672 * 2896310 * 30, greater than 58,000,000,000 times, the internal storage access number of times is greater than 5,000,000,000 times, and this can't realize on present PC, although the therefore superiority of its picture quality, existing algorithm is not all used three-dimensional SPL interpolation.
Make full use of single instruction multiple data (SIMD) arithmetic capability of GPU among the present invention, go to address this problem.GPU hardware inside is when the storage texture, often the texture processing for RGBA four-way form is the most effective, every passage is 32 a floating number, so each texture pixel (Texel) is made up of 128bit, can handle the multiplication and the addition of 4 floating numbers when calculating concurrently simultaneously.By examining formula (1), it can be write as the sum operation that length is 4 vector dot product again:
(x wherein
v, y
v, z
v) Fragment that is respectively pending is mapped to x, y, the z coordinate of data space, i, j, k is the summation index variables, for load three-dimensional intravascular ultrasound image data on the z frame (x, the y) pixel value at coordinate place, f are the SPL function,
Be that length is 4 vector:
＜, be the vector dot product computing.By same operation, also can be rewritten into length respectively to the summation of j and k in the formula (2) is that 4 vector dot product calculates, therefore want calculating formula (1), needing 21 length altogether is 4 vector dot product, use the single instruction multiple data vector dot product computing of Graphics Processing Unit, once finish a length and be 4 vector dot product, can in 21 clock period, finish by the SIMD instruction of GPU.About internal storage access, because data are divided into the fritter of M * M * M before, wherein each fritter can both be loaded in the texture cache of GPU, therefore in theory to each data block, only need once slow video memory visit, all later on texture visits are all directly obtained from buffer memory.
The above; only be the embodiment among the present invention; but protection scope of the present invention is not limited thereto; anyly be familiar with the people of this technology in the disclosed technical scope of the present invention; can understand conversion or the replacement expected; all should be encompassed in of the present invention comprising within the scope, therefore, protection scope of the present invention should be as the criterion with the protection domain of claims.
Claims (3)
1. the method for a quickly with high fidelity structuring ultrasound long axis image is characterized in that, step is as follows:
Step 1: load three-dimensional intravascular ultrasound data, its width and highly be respectively W pixel, H pixel, the image frame number is N, neighbor x is x to physical separation
sMillimeter, y is y to physical separation
sMillimeter, z is z to physical separation between the adjacent two frame images
sMillimeter;
Step 2: obtain the straight line section AB that the user defines on the first frame image, straight-line segment AB is by the central point C of image, and the angle of straight-line segment AB and x axle is θ;
Step 3: carry out quaternary tree along the plane, the first frame image place of three-dimensional intravascular ultrasound data and evenly divide, make that the size of leaf node is M * M;
Step 4: quaternary tree is begun the recurrence traversal from root node, and handle each node, comprising:
Judge that whether present node intersects with straight-line segment AB, if present node and straight-line segment AB intersect and present node is not a leaf node, then continues the recurrence traversal; If present node and straight-line segment AB intersect and present node is a leaf node, then execution in step 5; Directly return father node if present node and straight-line segment AB are non-intersect and continue traversal;
Step 5: the x, the y coordinate that calculate intersection point F, the G of leaf node and straight-line segment AB, and obtain x, the y coordinate of corresponding point intersection point F ', G ' on the M+1 frame image simultaneously, make the x of intersection point F ', x, the y coordinate that the y coordinate equals intersection point F, the x of intersection point G ', y coordinate equal x, the y coordinate of intersection point G, and the z coordinate that the z coordinate of intersection point F ', G ' equals intersection point F, G respectively adds M * z
sDraw plane FF ' G ' G, simultaneously the data block at leaf node place is transferred to Graphics Processing Unit in the mode of three-D grain and handle; Said process is repeated
Inferior, each only the need adds M * z respectively with the z coordinate of intersection point F, G, F ', G '
s, corresponding blocks of data upgrades thereupon and gets final product;
Step 6: each fragment that in the fragment shader of Graphics Processing Unit drafting plane FF ' G ' G is produced is carried out three-dimensional SPL interpolation.
2. the method for quickly with high fidelity structuring ultrasound long axis image according to claim 1 is characterized in that, the computing method that described leaf node and straight-line segment AB intersect comprise:
Rectangular four apex coordinates that obtain its correspondence from leaf node are: (x
0, y
0), (x
1, y
0), (x
1, y
1) and (x
0, y
1);
The coordinate C that calculated line section AB central point C is ordered
x, C
yFor: (C
x, C
y)=(Wx
s/ 2, Hy
s/ 2));
The parametric equation of calculated line section AB:
Wherein t is a parameter, and x (t), y (t) go up arbitrarily x, the y coordinate of any for straight-line segment AB;
Calculated line section AB and straight line x=x
0Intersection point: (x
0, C
y+ (x
0-C
x) tan θ);
Calculated line section AB and straight line x=x
1Intersection point: (x
1, C
y+ (x
1-C
x) tan θ);
Calculated line section AB and straight line y=y
0Intersection point: (C
x+ (y
0-C
y) cot θ, y
0);
Calculated line section AB and straight line y=y
1Intersection point: (C
x+ (y
1-C
y) cot θ, y
1);
Judge following four conditions:
Whether satisfy, if four conditions do not satisfy, straight-line segment AB and this leaf node are non-intersect so, if four conditions have at least one to satisfy, then straight-line segment AB and this leaf node intersect, and whenever satisfy a condition, just can obtain an intersecting point coordinate,, the summit of coincidence must be arranged when the condition that satisfies during greater than 2, remove the summit of coincidence, can obtain the coordinate of two intersection point F, G.
3. the method for quickly with high fidelity structuring ultrasound long axis image according to claim 1 is characterized in that, described three-dimensional SPL interpolation algorithm comprises:
Computing formula with three-dimensional SPL interpolation algorithm:
Be rewritten as based on length is the computing of 4 vector dot product:
(x wherein
v, y
v, z
v) fragment that is respectively pending is mapped to x, y, the z coordinate of data space, i, j, k is the summation index variables, I (x, y, z) for the three-dimensional intravascular ultrasound image data that loads on the z frame (x, the y) pixel value at coordinate place, f are the SPL function,
Be that length is 4 vector:
Also be that length is 4 vector:
＜, be the vector dot product computing;
Use the single instruction multiple data vector dot product computing of Graphics Processing Unit, once finish a length and be 4 vector dot product, finish described three-dimensional SPL interpolation by 21 vector dot product computings.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009102423419A CN101739661B (en) | 2009-12-09 | 2009-12-09 | Method for structuring ultrasound long axis image quickly with high fidelity |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009102423419A CN101739661B (en) | 2009-12-09 | 2009-12-09 | Method for structuring ultrasound long axis image quickly with high fidelity |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101739661A CN101739661A (en) | 2010-06-16 |
| CN101739661B true CN101739661B (en) | 2011-11-09 |
Family
ID=42463113
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2009102423419A Expired - Fee Related CN101739661B (en) | 2009-12-09 | 2009-12-09 | Method for structuring ultrasound long axis image quickly with high fidelity |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101739661B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102455885A (en) * | 2010-10-19 | 2012-05-16 | 李笑非 | Network display card with external access unit |
| EP3418923A1 (en) * | 2017-06-19 | 2018-12-26 | Dassault Systèmes | Computer-implemented method of designing a supporting structure for the packaging of a solid object |
| CN109035353B (en) * | 2018-06-27 | 2022-09-20 | 河南科技大学 | Blood vessel straightening curved surface recombination method based on CT image multi-plane reconstruction |
| CN113409436B (en) * | 2021-06-02 | 2023-10-13 | 东北大学 | Volume rendering method for diamond pixel arrangement |
-
2009
- 2009-12-09 CN CN2009102423419A patent/CN101739661B/en not_active Expired - Fee Related
Also Published As
| Publication number | Publication date |
|---|---|
| CN101739661A (en) | 2010-06-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101606854B (en) | High-precision real-time ultrasonic image scan conversion method | |
| Dai et al. | Real-time visualized freehand 3D ultrasound reconstruction based on GPU | |
| CN103971396A (en) | OpenGL ES (open graphics library for embedded system) implementation method for ray casting algorithm under ARM+GPU (advanced RISC machine+graphic processing unit) heterogeneous architecture | |
| CN101739661B (en) | Method for structuring ultrasound long axis image quickly with high fidelity | |
| JP2007537770A (en) | A dynamic crop box determination method for display optimization of luminal structures in endoscopic images | |
| JP5976003B2 (en) | Method and system for visualizing volume datasets | |
| KR101353303B1 (en) | Graphics processing unit and image processing apparatus having graphics processing unit and image processing method using graphics processing unit | |
| CN110462677A (en) | Single-pass flexible screen/scale rasterisation | |
| CN110060337B (en) | Carotid artery ultrasonic scanning three-dimensional reconstruction method and system | |
| JP5996870B2 (en) | Processing and analysis of cardiac images | |
| TWI601096B (en) | Method and apparatus for direct and interactive ray tracing of a subdivision surface | |
| CN101324962A (en) | Parallel processing method drawn by pre-projection light ray projection body | |
| CN109643463A (en) | Stagewise depth logic in preset pattern processor | |
| JP2012515598A5 (en) | ||
| CN102397082B (en) | Method and device for generating direction indicating diagram and ultrasonic three-dimensional imaging method and system | |
| Zhang et al. | Dynamic real-time 4D cardiac MDCT image display using GPU-accelerated volume rendering | |
| CN108805876A (en) | Using biomechanical model magnetic resonance and ultrasonoscopy can deformable registration | |
| CN108064148A (en) | The embryo transfer guided for image in vitro fertilization | |
| Wen et al. | GPU-based volume reconstruction for freehand 3D ultrasound imaging | |
| CN101174331B (en) | Maximum density projection generating method for medical image | |
| JP2014206931A (en) | Display processing program, display processing method, and display processing device | |
| Mazzeo et al. | In situ ray tracing and computational steering for interactive blood flow simulation | |
| Zhang et al. | Real-time visualization of 4D cardiac MR images using graphics processing units | |
| JP4299187B2 (en) | Ultrasonic diagnostic apparatus and echo data processing method | |
| JP4653536B2 (en) | Ultrasonic diagnostic equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20111109 Termination date: 20211209 |