CN100339872C - Systems and methods for providing automatic 3d lesion segmentation and measurements - Google Patents
Systems and methods for providing automatic 3d lesion segmentation and measurements Download PDFInfo
- Publication number
- CN100339872C CN100339872C CNB2004800062638A CN200480006263A CN100339872C CN 100339872 C CN100339872 C CN 100339872C CN B2004800062638 A CNB2004800062638 A CN B2004800062638A CN 200480006263 A CN200480006263 A CN 200480006263A CN 100339872 C CN100339872 C CN 100339872C
- Authority
- CN
- China
- Prior art keywords
- infringement
- dimensional
- coordinate space
- spherical
- volume
- 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
Images
Landscapes
- Image Processing (AREA)
- Apparatus For Radiation Diagnosis (AREA)
- Image Analysis (AREA)
Abstract
Systems and methods are provided for automatic 3D segmentation of abnormal anatomical structures such as colonic polyps, aneurisms or lung nodules, etc., in 3D medical imaging applications. For example, systems and methods for 3D lesion segmentation implement a centroid-based coordinate transformation (e.g., spherical transformation, ellipsoidal transformation, etc.) to transform a 3D surface of the lesion from an original volume space into, e.g., a spherical or ellipsoidal coordinate space, followed by interpolation of the transformed lesion surface to enable accurate determination of a boundary between a lesion and surrounding normal structures.
Description
The cross reference of related application
The sequence number that the application requires on March 11st, 2003 to submit to is 60/453573 U.S. Provisional Application No., by reference it intactly is attached to herein.
TECHNICAL FIELD OF THE INVENTION
In general, the present invention relates to three-dimensional (3D) medical image presents and visual system and method.More particularly, the present invention relates to be used to provide the medical imaging system and the method for the automatic three-dimensional segmentation of the abnormal anatomical structures (infringement) such as polyp of colon, aneurysm or lung brief summary, and be used to obtain for identification and the useful institute of classifying and cut apart the accurate dimension measurement of infringement and other describes the method for feature.
Background
In the imaging of medical field, the abnormal anatomical structures (infringement) such as polyp of colon, aneurysm or lung brief summary cut apart because of very variable shape, texture, density and the size of the infringement of this class and they are attached to becoming stubborn problem on the normal configuration on every side.For example, the problem of colonic polyp segmentation is difficult to consider the complicated shape of colon wall, and haustrum pleat that is wherein outstanding or that thicken and presence of residual feces be the shape and the density of similar polyp often.
Proposed the whole bag of tricks be provided cutting apart automatically of infringement in medical imaging system.For example, the automatic colon polyp segmentation method of having announced before having proposed, they adopt the surface segmentation of utilizing three-dimensional shape features, two-dimentional polyp segmentation technology or deformable model.More particularly, for example, title people such as H.Yoshida is that " computerize based on volume characteristic polyp of colon in the CT colon image detects: Primary Study " (Radiology 2002, disclose the polyp segmentation method that a kind of employing utilizes the surface segmentation of three-dimensional shape features in paper 222:327-336).This list of references discloses a kind of polyp candidate detection scheme, and the space connector that it has by extraction on the colon surface of particular shape characteristics usually adopts polyp segmentation.The condition morphological dilatation is as subsequent step.
In addition, for example, title people such as S.G kt ü rk is " the statistics three dimensional pattern disposal route of the computer aided detection of polyp in the CT colon image " (IEEE Trans.Med.Image., vol.20 (12), 1251-60 page or leaf, Dec calendar year 2001) a kind of two-dimentional polyp segmentation method is disclosed in the list of references.This list of references is described a kind of two-dimentional polyp segmentation technology, and it is applied to the some tlv triple to the vertical plane of the section of the sub-volumes around the polyp candidate.Cut apart and be intended to search the best square window that comprises candidate polyp.Quafric curve and fitting a straight line algorithm are used for searching the polypoid structure in the subwindow.
The defective of the two-dimentional polyp segmentation of the vertical plane tlv triple that is applied to the subimage that extracts from axial slices or is applied to the sub-volumes around the polyp candidate is cut into slices is, does not consider three-dimensional communication information.
Title people such as H.Yoshida is that " computerize based on volume characteristic polyp of colon in the CT colon image detects: Primary Study " (Radiology 2002, disclose the colonic polyp segmentation process of another kind of employing three-dimensional shape features in list of references 222:327-336).This list of references has been described a kind of three-dimensional polyp surface extraction method, and it realizes only cutting apart the polyp surface vertices.But, adopt the above-mentioned dividing method of two-dimentional polyp segmentation or three-dimensional polyp surface segmentation to be not suitable for the extraction of infringement continuously, also be not suitable for obtaining the accurate three-dimensional measurement of the density, texture and the shape that characterize whole infringement volume and describing feature.
Another kind of polyp segmentation is by people such as J.Yao proposition in " deformable model that instructs based on knowledge in the CT colon image polyp of colon cut apart automatically and detects " (Medical Imaging 2003, SPIE, vol.5031-41, in press).People such as Yao propose a kind of automatic polyp segmentation method of the combination based on fuzzy c-mean cluster and deformable model.The gradient of fuzzy membership functions is driven into polyp boundary as image force with the deformable surface around the seed.Therefore this method is at first considered Strength Changes, may have the segmentation result of misleading when not having visual boundary or Strength Changes between them in enteron aisle (loop) contact of colon.Under this class situation, paid close attention to that volume may comprise two colonic lumens being separated by tissue or one of them comprises two colon walls of polyp, and wherein the surface under the polyp belongs to another enteron aisle.It is the part on polyp surface that the method that is proposed may be thought the surface under the polyp by mistake, and it may cause extracting the volume bigger than actual polyp size.
Summary of the invention
Example embodiment of the present invention generally comprises three-dimensional medical imaging system and method, they provide the automatic three-dimensional segmentation of the abnormal anatomical structures (infringement) such as polyp of colon, aneurysm or lung brief summary, and the accurate dimension that obtains the three-dimensional segmentation infringement that can be used for discerning and classify measures and other describes feature.More particularly, example embodiment of the present invention generally comprises and is used to provide the accurate three-dimensional system and method that infringement is cut apart, wherein utilize coordinate transform (for example spherical transform, elliptic transformation etc.) based on barycenter with the three-dimensional surface of infringement from the initial volume spatial alternation to for example sphere or the elliptic coordinates space, then to the infringement after conversion surface interpolation, thereby allow accurately to determine infringement and the border between the normal configuration on every side.
In an example embodiment of the present invention, a kind of automatic 3D (three-dimensional) infringement dividing method comprises: the three-dimensional surface of determining the infringement that view data is concentrated in first coordinate space; Utilize the barycenter that damages in first coordinate space to carry out the barycenter conversion of three-dimensional surface, thereby the shift surface that produces three-dimensional surface in second coordinate space is represented; Handling shift surface represents to determine damage the infringement surface with normal configuration separation on every side; And, cut apart infringement thereby concentrate from view data with first coordinate space of remapping of the infringement surface in second coordinate space.
In another example embodiment of the present invention, a kind of automatic 3D (three-dimensional) infringement dividing method comprises: the three-dimensional surface of determining infringement in original three-D volumes space; The three-dimensional surface of infringement is transformed to spherical coordinate space; Handle three-dimensional surface determining the infringement surface in the spherical coordinate space in spherical coordinate space, it will damage with normal configuration on every side and separate; Infringement surface transformation in the spherical coordinate space is arrived original three-D volumes space; And utilize institute conversion infringement surface from original three-D volumes space, to extract the volume corresponding with infringement.
By below in conjunction with the detailed description of accompanying drawing to example embodiment, it is obvious that these and other example embodiment of the present invention, feature and advantage will be described or become.
Summary of drawings
Fig. 1 is a process flow diagram, and the automatic 3 d lesion dividing method according to an example embodiment of the present invention is described.
Fig. 2 is a process flow diagram, illustrates according to an example embodiment of the present invention, is used for determining the method for three-dimensional edges image, and it can be realized according to the method for Fig. 1.
Fig. 3 is a process flow diagram, illustrates according to an example embodiment of the present invention, is used to carry out the method for the spherical coordinate transformation of three-dimensional edges image, and it can be realized according to the method for Fig. 1.
Fig. 4 be according to an example embodiment of the present invention, the edge data that is used for handling spherical coordinate space paid close attention to the process flow diagram of the method for volume with extraction, it can be realized according to the method for Fig. 1.
Fig. 5 is according to an example embodiment of the present invention, is used to carry out from the inverse transformation of spherical coordinate space so that paid close attention to the process flow diagram of the method for volume from the original three-dimensional image extracting data, and it can be realized according to the method for Fig. 1.
Fig. 6 be according to an example embodiment of the present invention, be used for determining be not the wrong surface of a part on infringement surface with the process flow diagram of the method for the over-segmentation that prevents to damage volume, it can be realized according to the method for Fig. 1.
Fig. 7 A-7E is a synoptic diagram, and illustrative is according to the edge detection method of an example embodiment of the present invention.
Fig. 8 A and Fig. 8 B are synoptic diagram, and illustrative is according to the spherical coordinate transformation method of an example embodiment of the present invention.
Fig. 9 A and Fig. 9 B are synoptic diagram, and illustrative is according to the method for normalizing of an example embodiment of the present invention.
Figure 10 A and Figure 10 B are synoptic diagram, illustrative according to an example embodiment of the present invention, be used for expanding the image of spherical coordinate space so that damage the method for surperficial interpolation.
Figure 11 A and Figure 11 B are exemplary diagram, illustrate one may be because of not being the polyp segmentation process that the wrong surperficial detection of the part on infringement surface causes the polyp over-segmentation.
The detailed description of example embodiment
In general, example embodiment of the present invention as herein described is included in the used system and method for automatic three-dimensional segmentation of the abnormal anatomical structures such as polyp of colon, aneurysm or lung brief summary in the three-dimensional imaging of medical application.In an example embodiment of the present invention as herein described, be used for system and method that 3 d lesion cuts apart and realize being used for the automated process of the interpolation on the spherical coordinate transformation of three-dimensional edges image and infringement surface subsequently, its allows being paid close attention to infringement and border accurately definite between normal anatomical tissue and the structure on every side.
In addition, be provided for measuring the various sizes of three-dimensional segmentation infringement and the method for feature automatically, can realize that they are to be used for identification or the classification automatically based on extracted infringement volume according to demonstration system of the present invention and method.Specifically, can help user (radiologist) to obtain accurate infringement dimensional measurement automatically, distinguish infringement and other anatomical structure such as health tissues, remaining ight soil or streak artifacts according to system and method for the present invention.In addition, computer aided detection (CAD) system can expand to and comprise according to three-dimensional segmentation system and method for the present invention, thereby obtains the additional differentiation feature with characterize abnormalities infringement or as the input of assorting process.
Should be appreciated that according to system and method as herein described of the present invention and can realize with various forms of hardware, software, firmware, application specific processor or their combination.In an example embodiment of the present invention, system and method as herein described is embodied as application program with software, comprising programmed instruction, they are embodied in (for example flexible plastic disc, RAM, CD Rom, DVD, ROM and flash memory) in one or more program storage devices, and can be carried out by any device or the machine that comprise suitable architecture.
It is also understood that because construction system module shown in the drawings and method step can realize by software, so the actual connection between the system unit (or process steps flow process) can be according to the mode of application programming and different.Provide theory at this paper, those skilled in the art can design of the present invention these and reach similarly realization or configuration.
Fig. 1 is a process flow diagram, illustrates according to an example embodiment of the present invention, the method that provides automatic 3 d lesion to cut apart is provided.Fig. 1 also can be counted as the system that three-dimensional segmentation is provided, and wherein, the described method step of Fig. 1 is to carry out as herein describedly to be used to provide 3 d lesion to cut apart and the method measured and the parts or the module of function.In addition, be appreciated that demonstration system as herein described and method can maybe should be used for realizing for the three-dimensional imaging of medical and the CAD system of various imaging forms (CT, MRI etc.).In addition, demonstration system as herein described and method are well suited for automatic extraction and the measurement such as abnormal anatomical structures such as polyp of colon, aneurysm, lung brief summary or infringement.At this on the one hand, though example embodiment may specifically be described with reference to colonic polyp segmentation in this article, never should regard as to limit the scope of the invention.
Referring now to Fig. 1, exemplary three-dimensional infringement dividing method is generally to be paid close attention to the coordinate of infringement and begun (step 10) to cutting apart the module input.In the following description, suppose that specific anatomical in specific imaging form (for example CT, MRI etc.) (for example colon, lung, heart etc.) obtained original three-dimensional volume data collection (for example a plurality of two dimension slicing).In an example embodiment of the present invention, the infringement coordinate can be imported via GUI (graphic user interface), what wherein, the individual can use that mouse for example or indicating device select two dimension or 3-D display image (for example intracolic 3-D view) is subjected to region-of-interest (for example polyp).In another embodiment of the present invention, the infringement coordinate can be imported from automatic system or the method that can select/discern the candidate to damage position (input may need before or may not need the user to examine) automatically.
Subsequently, according to the input candidate, the part of original three-dimensional volume data collection (sub-volumes) is through handling (via edge detection method) to determine the three-dimensional edges (surface) (step 20) of selected infringement.More particularly, the realization edge detection process is determined the pixel in the three-dimensional sub-volume view data, and they are parts of the three-dimensional surface of the selected infringement in the original coordinates space.With reference to the process flow diagram of for example Fig. 2 and the synoptic diagram of Fig. 7 A-7E a demonstration methods carrying out the three-dimensional edges detection is described below.
In an example embodiment of the present invention, spherical coordinate transformation then is applied to three-dimensional edges view data (step 30).In general, selecting the centroid position of (or calculating automatically) according to spherical coordinate transformation process according to the present invention according to the user who is paid close attention to infringement, is that the surface of the selected infringement in the spherical coordinate space is represented with the 3 d lesion surface transformation.Should be appreciated that this paper will describe an exemplary spherical transformation process in order to describe.But understand, in general, any suitable centroid transformation process can realize according to the present invention, it is (for example cartesian coordinate space) extraction spatial data from the original coordinates space, and the centroid position of selecting or calculating according to the user in the original coordinates space, this spatial data is transformed into corresponding with spatial data from another coordinate space.The type of used centroid transformation process can be depending on the typical shape of being paid close attention to infringement.For example,, can adopt the elliptic transformation process to realize, in fact have oval shape because polyp is found polyp usually based on the conversion of barycenter for polyp.In this respect, this paper should not be counted as the type of the spendable conversion based on barycenter of restriction because those skilled in the art can be easy to imagine and understands can be according to theoretical other conversion based on barycenter for the three-dimensional segmentation realization of this paper.
The surface expression also through handling, so that accurately determine the surface of selected infringement, comprises the boundary/transition region (for example polyp neck) between the infringement (for example polyp), thereby selected infringement and surrounding tissue (intestines wall) is separated (step 40).Describe in more detail with reference to the process flow diagram of for example Fig. 3, Fig. 4 and Fig. 5 below and be used for three-dimensional edges data (infringement surface) are transformed to spherical coordinate space and handle institute's transform data to extract the of the present invention various example embodiment of selected infringement.
After in spherical coordinate space, having extracted selected infringement (step 40), use inverse transformation process so that the volume data that extracted in the spherical co-ordinate is converted to original coordinates (for example Cartesian coordinates), from original three-dimensional sub-volume view data, cut apart selected infringement (step 50) so that allow.For example, with reference to the process flow diagram of Fig. 5 demonstration inverse transformation process according to an example embodiment of the present invention is described below.
In another embodiment of the present invention, can realize automated procedure, be used to obtain that they can be used as the description feature of infringement identification and classification about cut apart of the various measurement results (step 60) of damaging volume.For example, can include but not limited to damage surface area from the measurement result that extract infringement volume obtains, tight ness rating, volume, average and Gaussian curvature, its mean value, minimum value, maximal value and STD, sphericity mean value, minimum value, maximal value and STD, minimum and maximum gauge, highly, neck area (for polyp), average strength, maximal value, minimum value and STD (being used to solve texture and pseudomorphism).
Fig. 2 is a process flow diagram, illustrates according to an example embodiment of the present invention, is used for determining the method for three-dimensional edges image.The method of Fig. 2 can realize for the step 20 of the demonstration methods of Fig. 1.In order to illustrate, also discuss the demonstration methods of Fig. 2 with reference to Fig. 7 A-7E, Fig. 7 A-7E illustrative according to the present invention, be used to detect the demonstration methods of the three-dimensional edges (surface) of being paid close attention to polyp of colon.
At first with reference to Fig. 2, initial step is the center-of-mass coordinate (step 21) of definite (or approximate treatment) selected infringement.In one embodiment of the invention, the center-of-mass coordinate of selected infringement can adopt any suitable method to determine automatically.In another example embodiment of the present invention, center-of-mass coordinate may be that the user selects, and the user selects to be considered on shown two-dimentional standard orthogonal view (axial, crown or sagittal) or the 3 d lesion (for example polyp) point of approximate center position thus.For example, Fig. 7 A is the exemplary three-dimensional chamber internal view of pedunculated polyp, and wherein, some C describes to select about the user of 3-D display polyp head the centroid position of (or determining automatically).
Then, concentrate extraction to comprise the sub-volumes data set (step 22) of selected infringement and surrounding environment from original three-dimensional volume data.In an example embodiment of the present invention, the sub-volumes that is extracted comprises that a plurality of spaces of the view data in the neighborhood that is in centroid position C are near two dimension slicing.For example, the sub-volumes that Fig. 7 B illustrative is extracted, it comprises 25 two dimension slicings that are arranged near the view data of the selected barycenter C shown in Fig. 7 A approx.In this example embodiment, suppose that two dimension slicing comprises the view data in the x-y plane of cartesian space.
The three-dimensional sub-volume of being extracted then adopts interpolation process to handle, so that present sub-volumes isotropy (step 23).More particularly, the two dimension slicing of sub-volumes comprises in the example embodiment of pixel in the x-y plane therein, carry out interpolation method in the z direction, so that make sub-volumes isotropy (that is, making the measure-alike of the size of the pixel in the x-y section and the pixel on the z direction).Interpolation process can adopt any suitable sampling process again to carry out.The two dimension slicing of the view data that the explanation of the exemplary diagram of Fig. 7 C produces from the two dimension slicing interpolation for Fig. 7 B.
Subsequently, the three-dimensional edges detection method is applied to isotropy sub-volumes data set so that determine the three-dimensional edges (surface) (step 24) of selected infringement.In an example embodiment, rim detection adopts three-dimensional Canny edge detector process to carry out, and it is the well-known process that is used to extract by the edge that obtain, a pixel thick that lags behind.For example, Fig. 7 D illustrative is applied to the result of exemplary edge detection process of the interpolation two dimension slicing of Fig. 7 C.As shown in the figure, each two dimension slicing comprises one or more lines on surface of the polyp of presentation graphs 7A.In addition, Fig. 7 E is the demonstration diagram of three-dimensional edges image, and it adopts the two-dimentional marginal date of Fig. 7 D to present.The exemplary diagram of Fig. 7 E has described to have the three-dimensional edges image of the cube volume of about 35 * 35 * 35 pixels.
Edge detection process be used for determining being included in the original three-dimensional sub-volume space paid close attention to infringement the coordinate of the pixel in the three-dimensional edges that detects (surface).The result of edge detection process outputs to spherical coordinate transformation process (step 25).
Fig. 3 is a process flow diagram, illustrates according to an example embodiment of the present invention, is used to carry out the method for the spherical coordinate transformation of three-dimensional edges view data.The method of Fig. 3 can realize in the step 30 of Fig. 1.In addition, Fig. 8 A-8B is an exemplary diagram, the illustrative spherical coordinate transformation process.With reference to Fig. 3, initial step is to be spherical co-ordinate (step 31) according to selected/centroid position of calculating with the coordinate transform of three-dimensional edges (surface) data.In an example embodiment, centroid transformation process comprise each point on the three-dimensional edges (surface) of selected infringement (x, y, z) with respect to the new coordinate of barycenter C (it is counted as the initial point in the original three-dimensional sub-volume space) (r, , θ), as follows:
Subsequently,, not any extra summit deleted (step 32) that only belongs to the infringement surface for example by checking the intensity gradient that makes progress from the footpath of the barycenter C of selected (or calculating).In addition, not that the remote surface that damages the part on surface is eliminated (step 33).This process produces shift surface and represents that (it comprises the two-dimensional representation on the infringement surface of detecting in the spherical coordinate space to r for , θ) (step 34), and ( θ) has only a value r for each therein.
Fig. 8 A and 8B are exemplary diagram, and illustrative is according to spherical coordinate transformation method of the present invention.More particularly, Fig. 8 A explanation comprises the two dimension slicing of the sub-volumes of the polyp that expands to colonic lumen L.Dotted line (N) expression " polyp neck ", wherein polyp invests on the colon wall.More particularly, " polyp neck " is the transition from the polyp head to the colon wall.When transition was elongated, polyp was called pedunculated polyp.
Shown in Fig. 8 A, spherical coordinate transformation is equivalent to send a plurality of " rays " (R) and determine the crossing position, edge (E) on infringement (polyp) surface each ray (R) and the original three-dimensional sub-volume space from centroid position (C), value wherein " r " expression ray from barycenter C to ray with the distance of the crossing position of edge E.Fig. 8 B is a synoptic diagram, the illustrative range conversion, and it produces all sections in the isotropy sub-volumes according to the process shown in Fig. 8 A.Fig. 8 B represents the surface seen from centroid position (C) wherein, to be expressed as surperficial height from barycenter (C) apart from r.
Shown in Fig. 8 A, the ray (R) that enters colon by polyp neck (N) not with the edge (E) of polyp on point intersect, but propagate into (through polyp neck N) in the colon.Therefore, adopt the method for eliminating remote surface (step 33).Describe an example embodiment that is used to eliminate remote surface in detail with reference to for example Fig. 6 and Figure 11 A-11B below.In general, in fact be used to eliminate is not that the method for remote surface of a part on infringement surface is based on for example can be applicable to for example various standards of the length of the ray of polyp neck N of restricted passage.
For example, can consider such as from the distance of barycenter with angle changes and two standards the intersection location of ray in the far surface between the ray in succession.Distance can be used near the mean distance of the ray difference distant place area (surface) and the barycenter.More particularly, for example, if barycenter is determined for the polyp of spherical shape, then raydist from (r) on average near the radius of the infringement of spherical shape.For example when the infringement such as polyp has neck (referring to Figure 11 B), situation is not so just.Under the situation of two adjacent rays of the object (for example polyp) of crossing over spherical shape, this class ray will be equidistant with barycenter approx with the point that intersect on the infringement surface.But when intersect and adjacent ray expands in the polyp neck and when intersecting with remote surface on a ray and polyp surface, the difference between the distance that the end points of these two adjacent rays that intersect with the surface faces toward will be greatly different.These also are detectable, and are expressed as discontinuous (for example referring to Figure 11 B) in the spherical face conversion.As mentioned above, will for example describe a demonstration program in the step 33 of Fig. 3, realizing in detail below with reference to Fig. 6 and Figure 11 A-11B.
Fig. 4 is a process flow diagram, a kind of method is described, be used for handling the expression on the infringement surface of spherical coordinate space, so that accurately determine to comprise the surface of the selected infringement on the border (for example polyp neck) between the infringement (for example polyp), thereby make selected infringement can separate (for example definite polyp neck that polyp and colon wall are separated) with surrounding tissue.The step 40 that the method for Fig. 4 can be Fig. 1 realizes.Referring now to Fig. 4, initial step is to calculate normalization (stretching, extension) factor (step 41) of each discreet radius value.Thereby the normalization process provides the expression on the surface in the equilibrium spheroid coordinate space to give the means of pixel equal weight (being their bi-directional scalings).Normalization process (step 41) is optional, but can simplify surface interpolation process, as described below (otherwise can realize the weighting interpolation process).
Fig. 9 A and Fig. 9 B are exemplary diagram, and illustrative is according to normalization process of the present invention.In this example, normalized factor is by δ (r)=r
Max/ r determines, careful degree is by D=(the 2 π r that round up
Max) determine, and angle step is determined by Δ θ=2 π/D.Fig. 9 A is in different radii (r with the graphics mode explanation
1=1, r
2=2, r
Max=3) relation between the lip-deep point of infringement.These figure illustrate the reduced form of normalization process, and process is relevant with two dimensional form, thereby via expressing this process to polar conversion.In this case, when radius increased, given angle (as shown in the figure) will be facing to the wideer part of circumference.When being transformed into utmost point space, referring to Fig. 9 B, more near the center (at r
1The place) name a person for a particular job " stretching, extension ".By δ (r)=r
MaxThe determined normalized factor of/r is caught this stretching, extension relation, and according to pixels to its quantification.The explanation of value among Fig. 9 A at distance r=1, be rounded up to the circumference l of next round values=6
1And normalized factor δ
1Quantitative relationship between the pixel at=3 places.Normalized factor is caught and is made innermost circumference and the same long required span of excircle.Therefore, in the example that is provided, for along excircle r
MaxEach unit of length, 3 units of corresponding effectively the inside circumference.When passing the surperficial interpolation of different radii, importantly suitably the effect of interpolating function is weighted, realize by demonstration normalization process.Above example expands to spherical co-ordinate and surface.
Refer again to Fig. 4, after normalization, the medium filtering process is applied to damage surface expression r (, θ) (step 42).Filtering has been eliminated owing to any noise that conversion process produces (calculating noise of parasitic measurement has been eliminated in filtering).The result of normalization (step 41) and filtering (step 42) is that the quantification on the infringement surface in the spherical coordinate space presents.Because the actual damage surface is actually level and smooth and continuous, therefore carry out interpolation process so that produce level and smooth continuous surface from quantizing the infringement surface, this will be used for extracting the infringement volume from surrounding health tissue.In an example embodiment of the present invention, in order from surrounding health tissue, to extract infringement, infringement surface expression r (, θ) further processed, so that determine to be paid close attention to the position and the shape on the infringement surface that volume and its environment separate.This process is determined infringement and the border between the normal configuration (release surface) on every side.For example, in the example embodiment of polyp segmentation as herein described, the infringement release surface is called " polyp neck ".
With reference to Fig. 4, for example, be used for helping the initial step of the interpolation of " polyp neck " to comprise carrying out expansion infringement surface expression r (, method (step 43) θ), its mode be consider following true: two-dimensional surface is represented r (, edge θ) is connected, be r (, θ+2 π)=r (, θ) and r (+π, θ)=and r (, θ).In an example embodiment, expansion process comprise with r ( θ) is mapped to r ' ( ', θ '), and is as follows:
(i)r′(
π/
2...3
π/
2]，[2π...3π])＝r([0...π]，[[0...2π]])；
(ii)r′([
π/
2...3
π/
2]，[1...π])＝r([0...π]，[[π...2π]])；
(iii)r′([0...
π/
2]，[π...3π])＝r([
π/
2...π]，[[0...2π]])；
(iv)r′([
π/
2...3π]，[3π...4π])＝r([0...π]，[[0...π]])；
(v)r′([3
π/
2...π]，[π...3π])＝r([0...
π/
2]，[0...2π]).
Figure 10 A and Figure 10 B are exemplary diagram, and (wherein, r (, zones of different θ), and the result of the image of Figure 10 B explanation expander graphs 10A are represented in Figure 10 A explanation infringement surface for , method θ) to be used for the surperficial expression of explanation expansion infringement r.As shown in the figure, expansion process mainly extracts the latter half of image and it is copied on the image top, extract the first half of image and it is copied under the bottom of image, extract the left side of image and it is copied to the right side of image, and extract the right of image and it is copied to the left side." fold " image on this process nature, it helps smooth interpolation process.More particularly, carry out duplicating image so that solve any problem for the support of borderline region.The border duplicates, folding or cross replication is a kind of known technology in the computer vision, and is used to provide for the interpolation of boundary and/or the support of filtering at this.For example, when the interpolate value of the frontier point on the left-hand side on computed image (surface), only there is the value on the right side on surface.A kind of traditional approach is by mirror reflection value or as performed here around providing support.This is significant in the context of spherical transform.Therefore, duplicate be equivalent to around the center with the mirror reflection image in case comprise few extra, thereby extract overlappingly on a small quantity, it is transformed into spherical space then as duplicating.In case interpolation is finished fully, the expansion of image is left in the basket, and only considers original image on (surface).
Interpolation process then adopts expanded images to carry out (step 45).Image spreading (step 43) and interpolation (step 45) produce the determining of smooth three-dimensional infringement surface, it comprises infringement and the border between the normal anatomical structures (for example polyp neck) or separator bar on every side.In other words, this process produces the extraction of the level and smooth closure surfaces of the infringement in the spherical coordinate space, and its surface is a benchmark with barycenter C.Next procedure is the infringement volume of determining in the original three-dimensional image space.
Fig. 5 is a process flow diagram, illustrates according to an example embodiment of the present invention, is used to carry out inverse transformation so that paid close attention to the method for volume from the original three-dimensional image extracting data.The method of Fig. 5 is applicable to the step 50 that realizes Fig. 1.At first, interpolation infringement surface data is input to inverse transformation process (step 51).Mapping is produced (for example look-up table) so that interpolation infringement surface (it is in spherical coordinate space) is remapped to original three-D volumes space (for example Cartesian coordinates).More particularly, in an example embodiment of the present invention, original sub-volume vertices (in Cartesian coordinates) is mapped to spherical coordinate space (step 52).
After this, all pixels in the original sub-volume are scanned (step 53).For each pixel, determine whether pixel has intensity (step 54).If this pixel does not have intensity level (for example pixel is in the colonic lumen) (negating in the step 54 determine), then pixel is not because of being a part of being paid close attention to volume be left in the basket (step 55).If this pixel has intensity (determining certainly in the step 54), but determine that (via mapping) pixel is not arranged within the interpolation infringement surface of spherical co-ordinate or under (step 56 negate to determine), then pixel is not because of being a part of being paid close attention to volume be left in the basket (step 55).
On the contrary, if given pixel has intensity (determining certainly in the step 54), and determine (via mapping) be arranged within the interpolation infringement surface of spherical co-ordinate or under (determining certainly of step 56), then pixel is marked as a part (step 57) of being paid close attention to volume.By to all the pixel repeating steps 54,55,56 and 57 in the original three-dimensional sub-volume, the volume of infringement can be cut apart from original three-dimensional sub-volume view data.
Cut apart in case pay close attention to volume, then can be carried out various measurements (step 60 of Fig. 1).For example, in an example embodiment, the volume of cutting apart infringement is included in the real surface (polyp surface) of cutting object by calculating and the number of the voxel (reaching part fully) between the following interpolation surface is determined.These voxels are converted into mm by actual x, y and the z spatial resolution of considering the data that obtain
3In addition, as mentioned above, another distance that maximum gauge (in polyp is measured is important) can extract by calculating between two extrema voxels in the volume is next definite.Intensity square (mean value, standard value etc.) can calculate by considering the intensity values of pixels that constitutes this volume.Curvature and other surface characteristics can directly derive from characterize this surperficial voxel, and adopt known curvature formulations to calculate, and wherein these moment also can be calculated.
In another embodiment of the present invention, can make the infringement cubing consistent and with irrelevant according to the initial selection centroid position of user's input or the infringement that couple candidate detection provided by carrying out iterative process.For example, an exemplary iterative procedure comprises for example determines new volume center, and repeats leaching process (for example step 30 of Fig. 1,40 and 50), till the bulking value convergence.
The method on remote/mistake surface by being provided for eliminating a part that does not in fact belong to infringement surface (as above with reference to as described in the step 33 of Fig. 3) is suitable for preventing to damage the over-segmentation of volume fully according to demonstration infringement segmenting system of the present invention and method.For example, for the three-dimensional segmentation of polyp, Figure 11 A and Figure 11 B are exemplary diagram, and the situation that wherein may occur over-segmentation for three-dimensional polyp segmentation is described.Specifically, Figure 11 A is the exemplary three-dimensional view of colon of area (A) with enteron aisle contact of colon.Shown in Figure 11 B, to be paid close attention to volume (V) and can be comprised two colonic lumens, they separate by tissue or by one of them border (B) that comprises between two adjacent colon walls being paid close attention to polyp.Shown in Figure 11 B, when spherical coordinate transformation by when the center projection radiation of polyp is used, the wrong surface (" false neck surface ") below the polyp may belong to another enteron aisle, it may be regarded as actual polyp neck by out of true ground.An explanation diagrammatic sketch of transformation results is shown in Figure 11 b, and wherein, transform data comprises " false neck surface ".When making up the polyp surface, this wrong surface should not considered, because if does not eliminate on wrong surface, then " extra " surface patches may form wrong polyp neck, so interpolation polyp surface may be far longer than actual polyp.Therefore, in order to prevent over-segmentation, can realize a kind of method of remote surface of a part of ignoring the surface that does not in fact belong to selected infringement according to the present invention.
Fig. 6 is a process flow diagram, illustrates according to an example embodiment of the present invention, is used to eliminate the method for wrong surface patches.In the method for Fig. 6, realize the remote surface patches that area growth process is followed the tracks of connection.Initial step is to determine one or more " seed point " (steps 70) of area growth process.In an example embodiment, the seed point of region growing is the point with the radius value that is higher than adaptive threshold.Adaptive threshold equals median radius value and adds the standard deviation of being paid close attention to the radius in the sub-volumes.
When the seed point was determined, area growth process adopted seed to put and carries out, thereby followed the tracks of the remote surface patches (step 71) that connects.For the set of surface patches of each connection, whether the surface area of definite set of patches that connects is less than appointed threshold (step 72).For example, in an example embodiment, threshold setting be the polyp surface in the spherical co-ordinate surface area 1/3.In other cases and when adopting different structure to carry out to cut apart, this ratio may be different.If the surface area of the set of patches that connects is defined as less than this thresholding (determining certainly in the step 72), then the set of patches of Lian Jieing will be left in the basket, and not regard the part (step 74) on infringement surface as.On the contrary, if the surface area of the set of patches that connects is not less than this thresholding (in the step 72 negates to determine), then the set of patches of Lian Jieing will be as the part involved (step 73) on infringement surface.
Everybody will appreciate that, as mentioned above, demonstration system as herein described and method can be by various application, realize as three-dimensional polyp segmentation.In this case, can realize as herein described according to demonstration methods of the present invention so that accurately extract polyp from colon wall on every side.Can realize demonstration system and method in addition so that measure, for example calculate that they will help to distinguish polyp and ight soil such as supplementary features such as Strength Changes, textures for user (radiologist) provides automatic polyp size.In addition, demonstration methods according to the present invention can be used to obtain accurate polyp shape characteristics, and they help to distinguish false-positive other source of polyp and for example main pleat and pseudomorphism and so on.
Though this paper is described illustrative embodiment of the present invention with reference to accompanying drawing, but everybody is appreciated that, the invention is not restricted to these specific embodiments, those skilled in the art can carry out various other changes and modification to it, and does not deviate from scope of the present invention or spirit.All of these changes and modifications all are included in the scope of the present invention that defines as claims.
Claims (32)
1. one kind is used for the method that automatic 3 d lesion is cut apart, and may further comprise the steps:
Determine the three-dimensional surface of the infringement in the original three-D volumes space;
The three-dimensional surface of described infringement is transformed to spherical coordinate space;
Handle the three-dimensional surface in the described spherical coordinate space, so that determine the infringement surface that described infringement and normal configuration are on every side separated in the described spherical coordinate space;
Described infringement surface transformation in the described spherical coordinate space is arrived described original three-D volumes space; And
Adopt the infringement surface of described conversion from described original three-D volumes space, to extract and the corresponding volume of described infringement.
2. the method for claim 1 is characterized in that, determines that the described step of three-dimensional surface comprises:
From described original three-D volumes space, extract three-dimensional sub-volume around described infringement;
With image data interpolation in described three-dimensional sub-volume, so that make described three-dimensional sub-volume isotropy; And
Determine the three-dimensional edges of the described infringement in the described isotropy three-dimensional sub-volume.
3. method as claimed in claim 2 is characterized in that, determines that the described step of the three-dimensional edges of described infringement adopts three-dimensional Canny edge detection process to carry out.
4. the method for claim 1 is characterized in that, the described step that the three-dimensional surface of described infringement is transformed to spherical coordinate space comprises:
Determine the centroid position of the described infringement in the described original three-D volumes space; And
Determine the spherical co-ordinate of each pixel of described three-dimensional surface according to described centroid position; And
Adopt described spherical co-ordinate in described spherical coordinate space, to produce the two-dimensional representation of described three-dimensional surface.
5. method as claimed in claim 4 is characterized in that, the described centroid position of described infringement is determined by automated procedure.
6. method as claimed in claim 4 is characterized in that, the described centroid position of described infringement is selected by the user.
7. method as claimed in claim 4 is characterized in that also comprising that the described two-dimensional representation to the described three-dimensional surface in the described spherical coordinate space carries out normalization.
8. method as claimed in claim 7 is characterized in that also comprising described normalized two-dimensional representation is carried out medium filtering.
9. method as claimed in claim 4 is characterized in that, the described step of handling the described three-dimensional surface in the described spherical coordinate space may further comprise the steps:
Expand the two-dimensional representation of the described three-dimensional surface in the described spherical coordinate space; And
The release surface of the two-dimensional representation of the described expansion of interpolation to determine the anatomical structure that described infringement and described infringement are adhered to is separated.
10. the method for claim 1, it is characterized in that, the described infringement surface transformation in the described spherical coordinate space is comprised to the described step in described original three-D volumes space the summit in the described original three-D volumes space is mapped to described spherical coordinate space.
11. method as claimed in claim 10, it is characterized in that the described step that adopts the infringement surface of described conversion to extract the volume corresponding with described infringement from described original three-D volumes space may further comprise the steps: as the part of described infringement volume comprise within the described infringement surface that is arranged in spherical co-ordinate in the described original three-D volumes space or under all pixels.
12. the method for claim 1 is characterized in that, described infringement is a polyp of colon.
13. method as claimed in claim 12, it is characterized in that, handle described three-dimensional surface in the described spherical coordinate space comprises definite polyp neck with the described step on the infringement surface determining in the described spherical coordinate space described infringement and the surface of normal configuration infringement on every side to be separated step.
14. the method for claim 1 is characterized in that also comprising the step of measuring the one or more parameters related with the volume of described extraction.
15. the method for claim 1 is characterized in that also comprising with iterative manner and repeats described method step to obtain the convergence of bulking value.
16. the method for claim 1 is characterized in that further comprising the steps of:
Determine whether the described infringement surface expression in the described spherical coordinate space comprises the wrong surface of a part that in fact is not described infringement; And
Elimination is confirmed as being included in the wrong surface in the described infringement surface expression.
17. one kind is used for the method that automatic 3 d lesion is cut apart, may further comprise the steps:
In first coordinate space, in image data set, determine the three-dimensional surface of infringement;
Adopt the barycenter of the described infringement in described first coordinate space to carry out the barycenter conversion of described three-dimensional surface, thereby the shift surface that produces described three-dimensional surface in second coordinate space is represented;
Handle described shift surface and represent, thereby determine infringement surface that described infringement and normal configuration are on every side separated; And
With described first coordinate space of remapping of the described infringement surface in described second coordinate space, cut apart described infringement so that concentrate from described view data.
18. method as claimed in claim 17 is characterized in that, determines that the described step of three-dimensional surface comprises:
Concentrate the three-dimensional sub-volume of extracting around the view data of described infringement from described view data;
Image data interpolation is arrived described three-dimensional sub-volume, so that make described three-dimensional sub-volume isotropy; And
Determine the three-dimensional edges of the described infringement in the described isotropy three-dimensional sub-volume.
19. method as claimed in claim 18 is characterized in that, determines that the described step of the three-dimensional edges of described infringement adopts three-dimensional Canny edge detection process to carry out.
20. method as claimed in claim 17 is characterized in that, the described step of carrying out the barycenter conversion comprises the execution spherical transform.
21. method as claimed in claim 20 is characterized in that, carries out spherical transform and comprises:
Determine the spherical co-ordinate of each pixel of described three-dimensional surface according to described centroid position; And
Adopt described spherical co-ordinate in described spherical coordinate space, to produce the two-dimensional representation of described three-dimensional surface.
22. method as claimed in claim 17 is characterized in that also comprising the step of determining the centroid position of the described infringement in described first coordinate space automatically.
23. method as claimed in claim 17 is characterized in that also comprising that the user selects the step of the coordinate of described barycenter.
24. method as claimed in claim 17 is characterized in that also comprising described shift surface is represented to carry out normalization.
25. method as claimed in claim 24 is characterized in that also comprising described normalization shift surface is represented to carry out medium filtering.
26. method as claimed in claim 17 is characterized in that, handles described step that described shift surface represents and comprises that the described shift surface of interpolation represents, so that determine release surface that the anatomical structure that described infringement and described infringement are attached to is separated.
27. method as claimed in claim 17, it is characterized in that, described first coordinate space is remapped so that concentrate the described step of cutting apart described infringement to comprise that the summit that the described view data described first coordinate space is concentrated is mapped to described second coordinate space from described view data in the described infringement surface in described second coordinate space.
28. method as claimed in claim 27, it is characterized in that also comprising by the part as the volume of described infringement comprise within the described infringement surface that is arranged in described second coordinate space in the described image data set or under all pixels, cut apart described infringement.
29. method as claimed in claim 28 is characterized in that also comprising the step of measuring the one or more parameters related with the described infringement of cutting apart.
30. method as claimed in claim 28 is characterized in that also comprising with iterative manner and repeats described method step to obtain the convergence of bulking value.
31. method as claimed in claim 17 is characterized in that, described infringement is a polyp of colon.
32. method as claimed in claim 31 is characterized in that, handles described shift surface and represents to comprise definite polyp neck with the described step on the infringement surface determining described infringement and normal configuration are on every side separated.
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US45357303P | 2003-03-11 | 2003-03-11 | |
| US60/453,573 | 2003-03-11 | ||
| US10/796,863 | 2004-03-09 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1759417A CN1759417A (en) | 2006-04-12 |
| CN100339872C true CN100339872C (en) | 2007-09-26 |
Family
ID=36704032
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2004800062638A Expired - Fee Related CN100339872C (en) | 2003-03-11 | 2004-03-11 | Systems and methods for providing automatic 3d lesion segmentation and measurements |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100339872C (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5825908A (en) * | 1995-12-29 | 1998-10-20 | Medical Media Systems | Anatomical visualization and measurement system |
| US5836872A (en) * | 1989-04-13 | 1998-11-17 | Vanguard Imaging, Ltd. | Digital optical visualization, enhancement, quantification, and classification of surface and subsurface features of body surfaces |
| CN1246944A (en) * | 1997-12-10 | 2000-03-08 | Ge医疗系统有限公司 | Semiautomatic segmentation method for estimating three-D volumes |
| CN1395713A (en) * | 2000-01-18 | 2003-02-05 | 芝加哥大学 | Method, system and computer readable medium for two-dimensional and three-dimensional detection of lungs nodules in computed tomography image scans |
-
2004
- 2004-03-11 CN CNB2004800062638A patent/CN100339872C/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5836872A (en) * | 1989-04-13 | 1998-11-17 | Vanguard Imaging, Ltd. | Digital optical visualization, enhancement, quantification, and classification of surface and subsurface features of body surfaces |
| US5825908A (en) * | 1995-12-29 | 1998-10-20 | Medical Media Systems | Anatomical visualization and measurement system |
| CN1246944A (en) * | 1997-12-10 | 2000-03-08 | Ge医疗系统有限公司 | Semiautomatic segmentation method for estimating three-D volumes |
| CN1395713A (en) * | 2000-01-18 | 2003-02-05 | 芝加哥大学 | Method, system and computer readable medium for two-dimensional and three-dimensional detection of lungs nodules in computed tomography image scans |
Also Published As
| Publication number | Publication date |
|---|---|
| CN1759417A (en) | 2006-04-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP4347880B2 (en) | System and method for performing automatic three-dimensional lesion segmentation and measurement | |
| Valente et al. | Automatic 3D pulmonary nodule detection in CT images: a survey | |
| Song et al. | Lung lesion extraction using a toboggan based growing automatic segmentation approach | |
| JP6267710B2 (en) | System and method for automatically detecting pulmonary nodules in medical images | |
| EP1436771B1 (en) | Computer-aided detection of three-dimensional lesions | |
| US7929748B2 (en) | Characteristic path-based colon segmentation | |
| Pu et al. | A computational geometry approach to automated pulmonary fissure segmentation in CT examinations | |
| JP4388121B2 (en) | Polyp detection method in 3D image volume | |
| Alilou et al. | A comprehensive framework for automatic detection of pulmonary nodules in lung CT images | |
| US7492968B2 (en) | System and method for segmenting a structure of interest using an interpolation of a separating surface in an area of attachment to a structure having similar properties | |
| US20020006216A1 (en) | Method, system and computer readable medium for the two-dimensional and three-dimensional detection of lesions in computed tomography scans | |
| Yoshida et al. | CAD in CT colonography without and with oral contrast agents: progress and challenges | |
| Singadkar et al. | Automatic lung segmentation for the inclusion of juxtapleural nodules and pulmonary vessels using curvature based border correction | |
| CN101014977A (en) | Lesion boundary detection | |
| Zhu et al. | Increasing computer‐aided detection specificity by projection features for CT colonography | |
| CN102842132A (en) | CT pulmonary nodule detection method | |
| WO2005112769A1 (en) | Nodule detection | |
| US7412084B2 (en) | Method of analysis of local patterns of curvature distributions | |
| Sundaram et al. | Colon polyp detection using smoothed shape operators: preliminary results | |
| US20100260390A1 (en) | System and method for reduction of false positives during computer aided polyp detection | |
| Yao et al. | Employing topographical height map in colonic polyp measurement and false positive reduction | |
| CN100339872C (en) | Systems and methods for providing automatic 3d lesion segmentation and measurements | |
| Wei et al. | Automated teniae coli detection and identification on computed tomographic colonography | |
| Yao et al. | Computer aided detection of lytic bone metastases in the spine using routine CT images | |
| Tan et al. | Detection of breast cancer in automated 3D breast ultrasound |
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 | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20190826 Address after: Erlangen Patentee after: SIEMENS AG Address before: American Pennsylvania Patentee before: American Siemens Medical Solutions Inc. |
|
| 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: 20070926 Termination date: 20210311 |