CN1830005A - System and method for ground glass nodule (GGN) segmentation - Google Patents

Abstract

A system and method for ground glass nodule (GGN) segmentation is provided. The method comprises: selecting a point in a medical image, wherein the point is located in a GGN; defining a volume of interest (VOI) around the point, wherein the VOI comprises the GGN; removing a chest wall from the VOI; obtaining an initial state for a Markov random field; and segmenting the VOI, wherein the VOI is segmented using the Markov random field.

Description

The system and method that ground glass nodule (GGN) is cut apart

The mutual reference of related application

The application requires in the U.S. Provisional Application No.60/491 of submission on July 31st, 2003,650 interests, and the copy of this application is hereby incorporated by.

Background of invention

1. technical field

The present invention relates to nodule segmentation, and relate more specifically to adopt Markov random field in pulmonary computed tomographic (CT) volume, to carry out ground glass nodule (GGN) to cut apart.

2. the discussion of correlation technique

Ground glass nodule (GGN) be for example with the radiography outward appearance of the dim incoherent fuzzy pulmonary shadow of bottom vascular.GGN occurs with two kinds of forms, i.e. these two kinds of forms of " pure " as shown in fig. 1 and " mixing ".Pure GGN can't help any solid composition and forms, and is made up of some solid compositions and mix GGN.

GGN shows more clearly in than flat radiography in high resolution computer tomography (HRCT) image.In the HRCT image, GGN also is different from solid nodules ground and manifests, because solid nodules has higher contrast ratio and border clearly.In addition, the outward appearance of GGN in the HRCT image is extremely significant discovery because their often indication active with existence potential medicable process, such as the existence of bronchioalveolar carcinoma or diffusivity gland cancer.

Because GGN is relevant with active lung disease usually, so the existence of GGN usually causes further diagnostic evaluation, this diagnostic evaluation comprises for example lung biopsy.Therefore, computer based is cut apart and can be helped the medical expert to carry out the diagnosis and the treatment of some types of lung disease.Therefore, the system and method that needs a kind of computer based to cut apart, it can be used for exactly and as one man cutting apart GGN to carry out quick diagnosis.

Brief summary of the invention

The present invention is used for the system and method that ground glass nodule (GGN) cuts apart and overcomes the above-mentioned and other problem that runs in known teachings by providing a kind of.

In one embodiment of the invention, a kind ofly be used for the method that ground glass nodule (GGN) cuts apart and comprise: select a bit at medical image, wherein this point is positioned at GGN; Define volume of interest (VOI) around this point, wherein VOI comprises GGN; From VOI, remove the wall of the chest; Obtain the original state of Markov random field; And cut apart VOI, wherein adopt Markov random field to cut apart VOI.This method also comprises the acquisition medical image, wherein adopts computer tomography (CT) imaging technique to obtain this medical image.

This method also comprises: adopt computer assisted GGN detection technique to detect GGN; And manually detect GGN.Select this point automatically or manually.GGN is pure GGN and mixes a kind of among the GGN.This method also comprises the shape and one of size that defines VOI.Grow by execution area and to remove the wall of the chest.By after removing the wall of the chest on VOI execution area grow and obtain the original state of Markov random field.

Adopt the step that Markov random field is cut apart VOI to comprise: the posterior probability of definition VOI; And the maximal value that adopts posterior probability marks each pixel among the VOI, and wherein each pixel among the VOI is noted as one of GGN and background.Defined posterior probability is calculated by P (L|F) ∝ P (F|L) P (L).The step that marks each pixel by $l_{\stackrel{\‾}{x}}\left(i\right)=\underset{l\∈\{g,b\}}{\arg \min }\{U(g,i-1)+\frac{1}{{2\mathrm{\σ}}^2}[f\stackrel{\_}{\left(x\right)}-{\mathrm{\μ}}_g],U(b,i-1)+\frac{1}{{2\mathrm{\σ}}^2}[f\stackrel{\_}{\left(x\right)}-{\mathrm{\μ}}_b\left]\right\}$ Calculate, wherein mark comprises scanning VOI, up to reaching convergence.

This method also comprises: adopting after Markov random field cuts apart VOI, carry out shape analysis, be attached to that GGN goes up or near it blood vessel to remove; And the VOI that show to adopt Markov random field to cut apart.

In another embodiment of the present invention, the system that a kind of GGN of being used for is cut apart comprises: be used for stored program memory device; With the processor that memory device communicates, this processor is implemented with program: adopt the data definition GGN relevant with the medical image of lung volume of interest (VOI) on every side; From VOI, remove the wall of the chest; Obtain the original state of Markov random field; And cut apart VOI, wherein adopt Markov random field to cut apart VOI.This processor also implements to obtain medical image with program code, wherein adopts the CT imaging technique to obtain this medical image.

Grow by execution area and to remove the wall of the chest.Grow and obtain the original state of Markov random field by removing after the wall of the chest on VOI execution area.

When adopting Markov random field to cut apart VOI, this processor is also implemented with program code: the posterior probability of definition VOI; And the maximal value that adopts posterior probability marks each pixel among the VOI, and wherein each pixel among the VOI is noted as one of GGN and background.Defined posterior probability is calculated by P (L|F) ∝ P (F|L) P (L).The step that marks each pixel by $l_{\stackrel{\‾}{x}}\left(i\right)=\underset{l\∈\{g,b\}}{\arg \min }\{U(g,i-1)+\frac{1}{{2\mathrm{\σ}}^2}[f\stackrel{\_}{\left(x\right)}-{\mathrm{\μ}}_g],U(b,i-1)+\frac{1}{{2\mathrm{\σ}}^2}[f\stackrel{\_}{\left(x\right)}-{\mathrm{\μ}}_b\left]\right\}$ Calculate.

This processor is also implemented with program code: carry out shape analysis in the VOI that adopts Markov random field to cut apart, to remove the blood vessel that is attached to GGN; And show the VOI that adopts Markov random field to cut apart, wherein GGN is visible.

In another embodiment of the present invention, a kind of computer program comprises having and records the computer usable medium that is used for the computer program logic that GGN cuts apart on it, this computer program logic comprises: be used for selecting at medical image the program code of a bit, wherein this point is positioned near GGN or its; Be used for defining around this some the program code of VOI, wherein VOI comprises GGN; Be used for removing the program code of the wall of the chest from VOI; Be used to obtain the program code of the original state of Markov random field; And the program code that is used to cut apart VOI, wherein adopt Markov random field to cut apart VOI.

In another embodiment of the present invention, the system that a kind of GGN of being used for is cut apart comprises: be used for selecting at medical image the device of a bit, wherein this point is positioned at GGN; Be used for defining around this some the device of VOI, wherein VOI comprises GGN; Be used for removing the device of the wall of the chest from VOI; Be used to obtain the device of the original state of Markov random field; And the device that is used to cut apart VOI, wherein adopt Markov random field to cut apart VOI.

In another embodiment of the present invention, a kind ofly adopt Markov random field in the lung CT volume, to carry out the method that GGN cuts apart to comprise: from the data relevant, select GGN with the lung CT volume; Around GGN, define VOI; Grow by execution area on VOI and from VOI, remove the wall of the chest; By after removing the wall of the chest, cutting apart the original state that VOI obtains iterated conditional pattern (ICM) process; And adopt Markov random field to cut apart VOI, wherein this is cut apart and comprises: the posterior probability of definition VOI; And carry out the ICM process, and wherein the ICM process comprises that the maximal value that adopts posterior probability marks each pixel among the VOI, and wherein each pixel among the VOI is noted as one of GGN and background, and each pixel in VOI all is marked.

Defined posterior probability is calculated by P (L|F) ∝ P (F|L) P (L).During the ICM process mark each pixel step by $l_{\stackrel{\‾}{x}}\left(i\right)=\underset{l\∈\{g,b\}}{\arg \min }\{U(g,i-1)+\frac{1}{{2\mathrm{\σ}}^2}[f\stackrel{\_}{\left(x\right)}-{\mathrm{\μ}}_g],U(b,i-1)+\frac{1}{{2\mathrm{\σ}}^2}[f\stackrel{\_}{\left(x\right)}-{\mathrm{\μ}}_b\left]\right\}$ Calculate, wherein the ICM process is from original state.

Above-mentioned feature is represented embodiment, and is presented and helps understand the present invention.Should be appreciated that these features do not plan to be considered to restriction of the present invention defined by the claims, or to the restriction of the equivalent of claim.Therefore, this summary of feature should not be in when determining equivalent and is considered to conclusive.Additional feature of the present invention will be in the following description, become obvious from accompanying drawing and accessory rights requirement.

The accompanying drawing summary

Fig. 1 illustrates " pure " ground glass nodule (GGN) and " mixing " GGN;

Fig. 2 according to one exemplary embodiment of the present invention, be used for the block diagram of the system that GGN cuts apart;

Fig. 3 illustrates according to process flow diagram one exemplary embodiment of the present invention, that be used for the method that GGN cuts apart;

Fig. 4 illustrates according to connectivity types one exemplary embodiment of the present invention, that adopt during region growing;

Fig. 5 illustrates according to a series of groups one exemplary embodiment of the present invention, that adopt during region growing (clique);

Fig. 6 illustrates the order according to raster scanning one exemplary embodiment of the present invention, that adopted by iterated conditional pattern (ICM); And

Fig. 7 illustrate according to one exemplary embodiment of the present invention, carry out GGN cut apart before and carrying out the several GGNs of GGN after cutting apart.

The detailed description of one exemplary embodiment

Fig. 2 according to one exemplary embodiment of the present invention, be used for the block diagram of the system that ground glass nodule (GGN) cuts apart.As shown in Figure 2, this system especially comprises scanning device 205, personal computer (PC) 210 and operator's control desk and/or is connected virtual navigation terminal 215 on the ethernet network 220 for example.Scanning device 205 is high resolution computer tomography (HRCT) imaging device.

The PC 210 that can be portable or laptop computer, PDA(Personal Digital Assistant) etc. comprises CPU (central processing unit) (CPU) 225 and storer 230, and these devices are connected to input 255 and output 260.PC 210 is connected to volume of interest (VOI) selector switch 245 and comprises on the splitting equipment 250 of one or more ground glass nodule (GGN) dividing method.PC 210 also can be connected to and/or comprise diagnostic module, and this diagnostic module is used to carry out the automatic diagnosis or the Function of Evaluation of medical image.In addition, PC 210 also can be coupled on the lung volume examination device.

Storer 230 comprises random-access memory (ram) 235 and ROM (read-only memory) (ROM) 240.Storer 230 also can comprise the combination of database, disc driver, tape drive etc. or these equipment.RAM 235 is as data-carrier store, and it is stored in the data that adopted during the program of carrying out among the CPU225, and is used as the workspace.ROM 240 is as program storage, is used for being stored in the performed program of CPU 225.Input 255 is made up of keyboard, mouse etc., is made up of LCD (LCD), cathode ray tube (CRT) display, printer etc. and export 260.

The operation of system is controlled by operator's control desk 215, and this operator's control desk 215 comprises controller 270 (for example keyboard) and display 265 (for example CRT monitor).Operator's control desk 215 is communicated by letter with scanning device 205 with PC 210, also can observe on display 265 thereby can plot the 3D data by PC 210 by the 2D view data that scanning device 205 is collected.Be to be understood that, under the situation that does not have operator's control desk 215, adopt for example import 255 and output 260 equipment carry out by some performed task of controller 270 and display 265, PC 210 can be configured to operate and show the information that is provided by scanning device 205.

Operator's control desk 215 also comprises any suitable image rendering (rendering) systems/tools/application program, it can handle the Digital Image Data of the image data set (or its part) that is obtained, to generate and demonstration 2D and/or 3D rendering on display 265.More specifically, the 2D/3D that image drawing system can provide medical image draws and visual application program, and it is carried out on universal or special computer workstation.In addition, image drawing system can navigate the user in 3D rendering or a plurality of 2D image slices.PC 210 also can comprise the image drawing system/tool/application of the Digital Image Data that is used to handle the image data set that is obtained, to generate and to show 2D and/or 3D rendering.

As shown in Figure 2, PC 210 also adopts splitting equipment 250 to receive and handle digital medical image data, the form that this digital medical image data can be taked as mentioned above is raw image data, 2D data reconstruction (for example axial slices) or 3D data reconstruction (such as volumetric image data or the reorganization of many planes), or the combination in any of these forms.Data processed result can output to image drawing system operator's control desk 215 from PC 210 by network 220, and the 2D and/or the 3D that are used for generating according to the view data of data processed result (such as the cutting apart of organ or anatomical structure, color or Strength Changes etc.) draw.

Should be appreciated that system and method that the GGN of being used for according to the present invention is cut apart may be implemented as the expansion of the conventional segmentation methods that is used to handle medical image or substitutes.In addition, should recognize, example system described here and method can be easy to utilize 3D medical image and computer-aided diagnosis (CAD) system or application program to implement, described system or application program are suitable for the imaging pattern (for example CT, MRI etc.) of wide region, and be suitable for diagnosis and estimate various unusual lung structures or damage, such as lung tubercle, tumour, narrow, inflamed areas etc.In this respect, though one exemplary embodiment can be described with reference to specific imaging pattern or particular anatomical features at this, these should be interpreted as limitation of the scope of the invention.

It is also understood that the present invention can various forms of hardware, software, firmware, application specific processor or its make up and implement.In one embodiment, the present invention can be used as the software that is tangibly embodied in the application program on the program storage device (for example magnetic floppy disc, RAM, CD ROM, DVD, ROM and flash memory) and implements.This application program can be downloaded to the machine that comprises any suitable construction and by its execution.

Fig. 3 illustrates according to process flow diagram one exemplary embodiment of the present invention, that be used for the operation of the method that GGN cuts apart.As shown in Figure 3, from lung or a pair of lung, obtain 3D data (step 31O).This is by adopting scanning device 205 (for example HRCT scanner) thereby scanning lung generates a series of 2D images relevant with this lung and finish.The 2D image of lung can be changed or be transformed into the 3D drawing image then, for example shown in (a) among Fig. 7 row.

After from lung, obtaining the 3D data, select GGN (step 320).This for example manually selects GGN by medical expert (such as the radiologist) or by adopting computer assisted GGN to detect and/or feature technology is finished from data.As an alternative, in step 320, can select a bit in GGN or near it.This process also can check that maybe this is manually carried out the relevant data of lung with this lung, is perhaps automatically performed by the computing machine that is programmed the point among the GGN that discerns in the medical image by the radiologist.

After selecting GGN, adopt VOI selector switch 245 to define VOI (step 330).In this step, the size of VOI and/or shape are defined as automatically comprises GGN.Indicate in the zone that is positioned at GGN square frame on every side in example VOI (a) row by Fig. 7.The zoomed-in view of VOI shown in (b) row of Fig. 7.Next, carry out the pre-service of VOI.Particularly, from VOI, remove the wall of the chest (step 340).Therefore, the part that for example belongs to the VOI of the wall of the chest is excluded from VOI.This is to finish with the zone of removing among the VOI that belongs to the wall of the chest by the execution area growth.Therefore, eliminated the potential impact of the wall of the chest, such as the influence of MRF being cut apart (as discussed below) to further treatment technology.

Next, (have removed the wall of the chest) VOI is cut apart (step 350).This for example adopts region growing to carry out, wherein the seed points of this region growing be among the VOI a bit, this point is in GGN or in its vicinity.The example of the connectivity types that can adopt during region growing as shown in Figure 4.For example, (be respectively X when pixel and slice spacings _ResAnd Z _Res) when satisfying the condition that the 10-shown in 1 as the following formula is communicated with (wherein d is predetermined distance constant), carry out the growth of 10-connected region, as shown in Figure 4.Similarly, when preset distance constant d satisfies the condition of the 18-connection shown in the formula 1, carry out the growth of 18-connected region, as shown in Figure 4.

Should be appreciated that and in step 350, can adopt various additional cutting techniques, such as based on histogram analysis get that threshold value, Gauss are level and smooth, rim detection and template matches.It is also understood that the initial segmentation state of iterated conditional pattern (ICM) process of carrying out in the step 360 that execution in step 350 obtains to be discussed below.

In step 350, cut apart after the VOI, adopt Markov random field (MRF) that VOI is cut apart (step 360) once more.For example, utilize MRF merging and organization space and temporal information by the general knowledge of introducing about feature to be estimated, this MRF has stipulated the nonlinear interaction between the phase Sihe different characteristic.For example, by using MRF in step 360, MRF provides prior probability by the space constraint of using from the adjacent voxels among the VOI.Then can be by considering the intensity and the space constraint of adjacent voxels, to each the voxel assigned tags among the VOI.Therefore, GGN can be given a kind of type, but not GGN or background information (for example pulmonary parenchyma, blood vessel, wall of the chest part etc.) are given another kind of type.Thereby the VOI that allow to adopt MRF to cut apart is shown discretely illustrative examples as GGN and background as the row (c) of Fig. 7 as shown in, and wherein the zone by the jagged edge indication of image central authorities illustrates GGN, and the perimeter is a background.

The MRF cutting procedure of step 360 is following to be derived and to carry out.At first, make Ω _ R ³Expression VOI, the intensity with VOI is thought of as random field F (x), wherein x ∈ Ω then.Next, make l represent the dividing mark of voxel x, and l ∈ L={GGN, background }.Adopt the Bayes' theorem shown in 2 as the following formula then, from condition intensity probability P (F|L) and prior probability P (L), obtain to be used to mark the posterior probability P (L/F) of GGN and background information.

P(L|F)∝P(F|L)P(L) [2]

Therefore, provide the best mark of statistics that MRF is cut apart by posterior maximal value (MAP).

The conditional probability P that illustrates above (F/L) GGN and background intensity obtain in distributing, and these distribute and are simulated by the Gaussian distribution shown in 3 as the following formula,

$P(F=f|L=1)=\frac{1}{\sqrt{2\mathrm{\π\σ}}}\exp [-\frac{1}{{2\mathrm{\σ}}^2}{(f-{\mathrm{\μ}}_l)}^2]---\left[3\right]$

μ wherein _lBe the mean value of GGN or background intensity.Suppose that dividing mark L is MRF, then prior probability P (L) is provided by the Gibbs Distribution shown in 4 as the following formula,

P(L＝l)∝exp[-U(l)]， [4]

Energy function wherein $U\left(l\right)=\underset{c\∈C}{\mathrm{\Σ}}{V}_c\left(l\right)$ Be the summation on the set C of all pixels of the 3D neighborhood definition that is communicated with by 26-and two pixels group, it is used for definition group (shown in Figure 5), after this discusses with reference to formula 7.The C of one pixel group ₁Potential function V _c(l) define by formula 5,

L wherein _xMark for current voxel x.α _lThe prior probability of indication specific markers l, just, less α _lMean that mark l is preferred by prior probability, and bigger α _lMean that mark l is not preferred.The c ∈ C of two pixels group ₂Potential function V _c(l) define by formula 6,

Wherein

Represent the adjacent voxels during for example two pixels are rolled into a ball, and β _kDesign based on the group's type shown in the formula 7,

Wherein (b), (c) ..., (n) be isoplanar (for example b-e) and crossing plane (for example f-n) group, as shown in Figure 5, β is the potential constant of being scheduled to, and w is a weighting constant.When the distance between two pixels of group is big, β _kLess; And the distance between two pixels of group hour, β _kBigger.

Adopting formula 3 to determine conditional probability and adopting formula 4 to determine prior probability P (L) afterwards, the data that draw from formula 3 and 4 are updated to the formula 2, to calculate the posterior probability P (L/F) that is used to mark GGN and background information.The optimization of MAP then becomes minimization process, as the following formula shown in 8.

$\mathrm{MAP}=\underset{L}{\arg \min }[U\left(L\right)+\frac{1}{2{\mathrm{\σ}}^2}(F-{\mathrm{\μ}}_L)]---\left[8\right]$

L={l|GGN wherein, background }.

Determined the MAP of formula 8 then by the ICM process, the MAP of this formula 8 is last segmentation result.The ICM process that starts from the original state (for example iteration 0) determined by the region growing in the step 350 is with mark l _x(i) be assigned to the voxel x on each iteration i in the formula 9,

Wherein U (g, i-1) and U (b, i-1) energy value for from the mark state of the iteration i-1 of GGN and context marker, calculating respectively.F (x) is the intensity level of voxel x, and μ _gAnd μ _bBe respectively the average intensity value of GGN and context marker.

During each ICM iteration, carry out the raster scanning of VOI, and therefore designated GGN of voxel or the context marker among the VOI.As shown in Figure 6, the voxel mark is upgraded by the raster scanning of carrying out with eight kinds of different modes: (1) from the left front angle (A) of going up of VOI to right back inferior horn (H); (2) from the right front angle (B) of going up of VOI to left back inferior horn (G); (3) from the left front inferior horn (C) of VOI to the right back angle (F) of going up; (4) from the right front inferior horn (D) of VOI to right back inferior horn (E); (5) from the right back inferior horn (H) of VOI to the left front angle (A) of going up; (6) from the left back inferior horn (G) of VOI to the right front angle (B) of going up; (7) from the right back angle (F) of going up of VOI to left front inferior horn (C); And (8) from the left back angle (E) of going up of VOI to right front inferior horn (D).

Repeat said process (wherein for each ICM iteration, with a kind of raster scanning of carrying out in eight kinds of modes), up to observing convergence.In other words, repeat said process, all voxels in VOI all are marked.Should be appreciated that and in said process, can adopt scanning sequence and/or interchangeable raster scanning process in proper order with varying number.

After the employing Markov random field was cut apart VOI in step 360, the VOI of being cut apart can be through further handling (step 370).Especially, being attached near the blood vessel that GGN goes up or it is is removed from the VOI of being cut apart by carrying out shape analysis.The example of this situation can be observed in Fig. 7 (c) row, has wherein removed the blood vessel that is attached to GGN.For example, at first getting threshold value and compactedness by execution measures blood vessel and GGN are made a distinction identify attached to GGN and goes up or near it blood vessel, remove then attached to GGN and go up or near blood vessel it and come level and smooth these results, from GGN and/or VOI, remove blood vessel thus by using a series of morphological operations.The technology of removing blood vessel in VOI from GGN is the U.S. Provisional Application No.60/503 of " Improved GGO Nodule Segmentation withShape Analysis " at the title of submission on September 17th, 2003, be disclosed in 602, the copy of this application is hereby incorporated by.Next, the display 265 by for example operator's control desk 215 shows GGN (step 380) to the user.The example of the GGN that shows after execution MRF according to the present invention is cut apart is shown in Fig. 7 (d) row, and wherein the dark part of image central authorities is GGN.

Therefore, cut apart by carrying out MRF according to the present invention, GGN in the medical image just can be by coming to be distinguished accurately and quickly from background information to their associated voxels assigned tags, thereby make the GGN can be to the medical expert as seen, to be used for diagnosis and to estimate some pulmonary disease.

Should be appreciated that because the system component and the method step of some formations shown in the accompanying drawing can be implemented by software, so the mode that the actual connection between the system component (or process steps) can be programmed according to the present invention and different.Give to fix on the instruction of the present invention that this provides, those of ordinary skills can expect these and similarly embodiment or structure of the present invention.

It is also understood that top description only represents illustrative embodiment.For convenience of the reader, top description concentrates on the representative example of possible embodiment, and this example is an explanation principle of the present invention.This description is not to want to attempt exhaustive all possible modification.The alternative embodiment of specific part of the present invention may not present, or certain part can obtain the other refill of not describing, and these should not think to abandon those interchangeable embodiment.Under situation without departing from the spirit and scope of the present invention, can directly implement other application program and embodiment.Therefore, the present invention does not plan to be limited to specifically described embodiment, because can generate multiple above-mentioned conversion and combination and relate to the embodiment of not having creative replacement to above-mentioned, but the present invention will limit according to following claim.Should be understood that many embodiment that do not describe fall in the literal scope of following claim, and other is an equivalent.

Claims (31)

1, a kind ofly be used for the method that ground glass nodule (GGN) is cut apart, it comprises:

Select a bit in medical image, wherein this point is arranged in GGN;

Define volume of interest (VOI) around this point, wherein VOI comprises GGN;

From VOI, remove the wall of the chest;

Obtain the original state of Markov random field; And

Cut apart VOI, wherein adopt Markov random field to cut apart VOI.

2, method according to claim 1 also comprises:

Obtain described medical image.

3, method according to claim 2 wherein adopts computer tomography (CT) imaging technique to obtain described medical image.

4, method according to claim 1, wherein said point are to select automatically.

5, method according to claim 1, wherein said point are manually to select.

6, method according to claim 1 also comprises:

Adopt computer assisted GGN detection technique to detect GGN.

7, method according to claim 1 also comprises:

Manually detect GGN.

8, method according to claim 1, wherein GGN is pure GGN and mixes a kind of among the GGN.

9, method according to claim 1 is wherein grown by execution area and is removed the described wall of the chest.

10, method according to claim 1 also comprises:

The definition VOI shape and size in one.

11, method according to claim 1, wherein the original state of Markov random field by after removing the described wall of the chest on VOI execution area growth obtain.

12, method according to claim 1, wherein adopt the step that Markov random field is cut apart VOI to comprise:

The posterior probability of definition VOI; And

The maximal value of employing posterior probability marks each pixel among the VOI, and wherein each pixel among the VOI is noted as one of GGN and background.

13, method according to claim 12, wherein defined posterior probability is calculated by P (L|F) ∝ P (F|L) P (L).

14, method according to claim 12, the step that wherein marks each pixel by $l_{\stackrel{\‾}{x}}\left(i\right)=\underset{l\∈\{g,b\}}{\arg \min }\{U(g,i-1)+\frac{1}{2{\mathrm{\σ}}^2}[f\left(\stackrel{\‾}{x}\right)-{\mathrm{\μ}}_g],U(b,i-1)+\frac{1}{2{\mathrm{\σ}}^2}[f\left(\stackrel{\‾}{x}\right)-{\mathrm{\μ}}_b\left]\right\}$ Calculate.

15, method according to claim 12, wherein said mark comprises:

Scanning VOI is up to reaching convergence.

16, method according to claim 1 also comprises:

After the employing Markov random field is cut apart VOI, carry out shape analysis and be attached on the GGN or near the blood vessel it to remove.

17, method according to claim 1 also comprises:

Show the VOI that adopts Markov random field to cut apart.

18, a kind ofly be used for the system that ground glass nodule (GGN) is cut apart, it comprises:

Be used for stored program memory device;

With the processor that this memory device communicates, this processor is implemented with program:

Adopt the data definition GGN relevant volume of interest (VOI) on every side with the medical image of lung;

From VOI, remove the wall of the chest;

Obtain the original state of Markov random field; And

Cut apart VOI, wherein adopt Markov random field to cut apart VOI.

19, system according to claim 18, wherein processor is also implemented with program code:

Obtain described medical image, wherein adopt computer tomography (CT) imaging technique to obtain this medical image.

20, system according to claim 18 wherein grows by execution area and removes the described wall of the chest.

21, system according to claim 18, wherein by after removing the described wall of the chest on VOI execution area growth obtain the original state of Markov random field.

22, system according to claim 18, wherein, when adopting Markov random field to cut apart VOI, this processor is also implemented with program code:

The posterior probability of definition VOI; And

The maximal value of employing posterior probability marks each pixel among the VOI, and wherein, each pixel among the VOI is noted as one of GGN and background.

23, system according to claim 22, wherein defined posterior probability is calculated by P (L|F) ∝ P (F|L) P (L).

24, system according to claim 22, the step that wherein marks each pixel by $l_{\stackrel{\‾}{x}}\left(i\right)=\underset{l\∈\{g,b\}}{\arg \min }\{U(g,i-1)+\frac{1}{2{\mathrm{\σ}}^2}[f\left(\stackrel{\‾}{x}\right)-{\mathrm{\μ}}_g],U(b,i-1)+\frac{1}{2{\mathrm{\σ}}^2}[f\left(\stackrel{\‾}{x}\right)-{\mathrm{\μ}}_b\left]\right\}$ Calculate.

25, system according to claim 18, wherein this processor is also implemented with program code:

In the VOI that adopts Markov random field to cut apart, carry out shape analysis, to remove the blood vessel that is attached to GGN.

26, system according to claim 18, wherein this processor is also implemented with described program code:

Show the VOI that adopts Markov random field to cut apart, wherein GGN is visible.

27, a kind of computer program, it comprises having and records the computer usable medium that is used for the computer program logic that ground glass nodule (GGN) cuts apart on it that this computer program logic comprises:

Be used for selecting at medical image the program code of a bit, wherein this point is arranged near GGN or its;

Be used for defining around this some the program code of volume of interest (VOI), wherein VOI comprises GGN;

Be used for removing the program code of the wall of the chest from VOI;

Be used to obtain the program code of the original state of Markov random field; And

Be used to cut apart the program code of VOI, wherein adopt Markov random field to cut apart VOI.

28, a kind ofly be used for the system that ground glass nodule (GGN) is cut apart, it comprises:

Be used for selecting at medical image the device of a bit, wherein this point is arranged in GGN;

Be used for defining around this some the device of volume of interest (VOI), wherein VOI comprises GGN;

Be used for removing the device of the wall of the chest from VOI;

Be used to obtain the device of the original state of Markov random field; And

Be used to cut apart the device of VOI, wherein adopt Markov random field to cut apart VOI.

29, a kind of employing Markov random field carries out the method that ground glass nodule (GGN) is cut apart in pulmonary computed tomographic (CT) volume, and it comprises:

From the data relevant, select GGN with the lung CT volume;

Around GGN, define volume of interest (VOI);

Grow by execution area on VOI and from VOI, remove the wall of the chest;

By after removing the wall of the chest, cutting apart the original state that VOI obtains iterated conditional pattern (ICM) process; And

Adopt Markov random field to cut apart VOI, wherein this is cut apart and comprises:

The posterior probability of definition VOI; And

Carry out the ICM process, wherein the ICM process comprises that the maximal value that adopts posterior probability marks each pixel among the VOI, and wherein each pixel among the VOI is noted as one of GGN and background, and each pixel in VOI all is marked.

30, method according to claim 29, wherein defined posterior probability is calculated by P (L|F) ∝ P (F|L) P (L).

31, method according to claim 29, wherein the step of each pixel of mark is calculated by following formula during described ICM process, promptly

$l_{\stackrel{\‾}{x}}\left(i\right)=\underset{l\∈\{g,b\}}{\arg \min }\{U(g,i-1)+\frac{1}{2{\mathrm{\σ}}^2}[f\left(\stackrel{\‾}{x}\right)-{\mathrm{\μ}}_g],U(b,i-1)+\frac{1}{2{\mathrm{\σ}}^2}[f\left(\stackrel{\‾}{x}\right)-{\mathrm{\μ}}_b\left]\right\}.$

32, method according to claim 29, wherein said ICM process is from described original state.

Images