CN103745470A - Wavelet-based interactive segmentation method for polygonal outline evolution medical CT (computed tomography) image - Google Patents

Abstract

The invention discloses a wavelet-based interactive segmentation method for a polygonal outline evolution medical CT (computed tomography) image. The wavelet-based interactive segmentation method comprises the following steps: interactively acquiring the initial polygonal outline of a to-be-segmented region in the CT image (generating an initial outline and a direct and reactive rendering-based polygonal outline based on a longitudinal axis by aiming at object regions with different features by adopting a single seed point-based initial outline generation method); performing wavelet modulus maximum-based edge detection to obtain the global high-frequency information of the whole CT image; performing automatic evolution on the polygonal outline, and approaching the object regions; finally acquiring the segmentation result of the CT image. According to the wavelet-based interactive segmentation method for the polygonal outline evolution medical CT image disclosed by the invention, the defects that the applicability is poor and the flexibility is low in the prior art can be overcome so as to realize the advantages that the applicability is good and the flexibility is high.

Description

Polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo

Technical field

The present invention relates to technical field of image processing, particularly, relate to the polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo.

Background technology

the status of Interactive Segmentation method in medical image segmentation:

It is exactly certain specific character (as edge, intensity profile, texture etc.) according to image information that image is cut apart, and image is divided into different regions, and extracts technology and the process of target area.Medical image segmentation has himself special technical complexity, becomes focus and difficulties in graphical analysis in recent years.Along with popularizing of computer technology, image Segmentation Technology has a great development.People are cut apart in tradition on the basis of thinking and are constantly put into practice, optimize and improve, and have produced automanual Interactive Segmentation method and automatic division method.

But from current image Segmentation Technology applicable cases clinically, due to the complicacy of medical image, medical image segmentation generally will be used knowledge in medical domain and doctor's subjective experience.At the various automatic segmentation algorithms that field of machine vision is more ripe, for medical image, can not obtain desirable effect.For complicated medical image, Interactive Segmentation can make full use of doctor's subjective experience, and this artificial participation is essential to cutting apart of medical image.At present Interactive Segmentation algorithm becomes the research emphasis of domestic and international medical image segmentation because it has the features such as high precision, practicality are stronger.

the limitation of partitioning algorithm itself:

According to the different mechanism of dividing method, by image cut apart be roughly divided into based on the cutting apart of region similarity, based on border cut apart and based on cutting apart of physical characteristics (as texture) etc.According to image in recent years, cut apart progress and the trend in field, the comprehensive various traditional dividing methods of many researchers, learn from other's strong points to offset one's weaknesses, as the dividing method on calmodulin binding domain CaM and border ^[1].Simultaneously, as a comprehensive and very strong cross discipline of application, the Research Thinking in other field and theory have also incorporated in image Segmentation Technology more and more, as the partitioning algorithm based on wavelet neural network, based on the partitioning algorithm of mathematical morphology, based on the partitioning algorithm of partial differential equation, partitioning algorithm based on curve etc.

Make a general survey of domestic and international medical image segmentation area research and application achievements in recent years, up to the present, although medical image segmentation has been carried out to a large amount of research both at home and abroad, but due to the complicacy of human anatomic structure, scrambling and the interindividual otherness of histoorgan shape, also there is no a kind of general partitioning algorithm that is suitable for all images.In addition, because the Practical Performance of image segmentation algorithm itself is uneven, in the clinical practice of medical image segmentation, lack unified effective technical solution.

Some developed countries have obtained suitable achievement at image aspect cutting apart in the world, and the research of Medical Image Segmentation Techniques and application also maintain the leading position.For example, in the eighties in 20th century, the people such as Mallat by Wavelet Analysis Theory introduce image process and analysis field, obtained very ten-strike, along with its theory enrich constantly and perfect, application also constantly expand.The research field that the movable contour model being proposed by people such as Kass and the level set dividing method being proposed by Berkeley College Maths system are also cut apart at image is occupied important position.Siemens Company is provided with the theoretical method fundamental research of specialize medical image analysis and the research institute of software development in Princeton.

The U.S. has developed a set of algorithm platform (Insight Segmentation and Registration Toolkit, ITK) that is specifically designed to Medical Image Processing and analysis, and has successfully issued an opening, OO class libraries.ITK is a software development kit for the open source code of medical image segmentation and registration, provides some main flow algorithms, as multiple dividing methods such as region growing, Threshold segmentation, level sets.

In China, many R&D institutions have done a large amount of research to the analyzing and processing of medical image, and medical image visual analyzing treatment technology tentatively enters clinical practice at present.Some scientific research institutions have also developed the Medical Image Processing platform of oneself, the 3 D medical visual analyzing platform of being developed as the visual research of the human body cryosection image three-dimensional reconfiguration system of automation research institute of Chinese Academy of Sciences medical image research department, Tsing-Hua University exploitation, Northwest University etc.Medical Image Segmentation Techniques is an important component part in various Medical Image Processing platforms and visualization system.

The research that Domestic Medicine image is cut apart is mainly on the basis of the conventional thought such as gray level threshold segmentation method, rim detection split plot design, region growing segmentation method, in conjunction with specific theoretical tool, updates.The target extraction algorithm that for example border combines with region, the application of the medical image segmentation based on fuzzy set theory, the medicine CT image based on mathematical morphology and region merging is cut apart etc.

The at present technology of the medical image segmentation in clinical medical or be simply integrated to various algorithms, can not be mutual; Be the interactive operation of single poor efficiency, there is no elasticity, dirigibility is inadequate, can not bring into play the fine segmentation performance of highly effective algorithm.

1) although some image segmentation algorithm also likely obtains good closed outline (as cutting apart based on Canny operator), but the profile of these algorithm gained is of overall importance, be not for target area, thereby cannot meet the final demand of cutting apart in clinical practice meaning yet.

2) formation of some objective contour and evolutionary process principle complexity, calculates loaded down with trivial detailsly, often rests on experimental simulation and theoretical research aspect, as the partitioning algorithm based on partial differential equation ^[2]and movable contour model ^[3]deng.In actual clinical practice, particularly in the cutting apart of the larger medicine CT sequence image of data volume, these methods are also impracticable.

3) from the clinical practice angle of medical image, lack and both take into account whole efficiency (accuracy, robustness, time space complexity), there is again a whole set of technical scheme for medical image segmentation that has value for clinical application of better operating flexibility.Some algorithm itself is cut apart for medical image, can obtain desirable effect, but can not fully meet user's flexible and changeable clinical demand.

Realizing in process of the present invention, inventor finds at least to exist in prior art the defects such as poor for applicability and very flexible.

Summary of the invention

The object of the invention is to, for the problems referred to above, propose the polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo, to realize the advantage that applicability is good and dirigibility is good.

For achieving the above object, the technical solution used in the present invention is: the polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo, comprising:

A, by man-machine interaction, generate the initial profile in region to be split in CT image;

B, CT image is carried out to multi-scale morphology, obtain high frequency edge information of overall importance;

C, based on initial profile and high frequency edge information of overall importance, carry out initial profile orthogenic evolution, approach true

Real edge, obtains final target area.

Further, described step a, specifically comprises:

According to the space distribution of target area, select flexibly the profile generating algorithm based on single Seed Points, the profile generating algorithm based on longitudinal axis, the polygon that also can sketch the contours roughly a sealing is as initial profile.

Further, the described profile generating algorithm based on single Seed Points, is specially:

First user specifies a Seed Points O (x according to the concrete condition of target area to be split ₀, y ₀), then algorithm, centered by Seed Points, generates a simple regular polygon region P, automatically as seed region.

Further, the described profile generating algorithm based on longitudinal axis, is specially:

First, by user interactions, draw longitudinal axis;

Then, each section of axis extends (stop condition is determined by similarity criteria) by the direction perpendicular to axis automatically to two sides, and every section of longitudinal axis will obtain two frontier points like this;

Last end points that (clockwise or counterclockwise) connects each segment boundary point and longitudinal axis in certain sequence finally, forms a closed polygonal profile.

Further, in step b, described multi-scale morphology, specifically adopts the edge detection method of Wavelet Modulus Maxima, comprising:

(1) based on by the given decomposition scale of real needs, calculate the two-dimensional wavelet transformation of the CT sequence image of composition CT image to be split

(2) calculate the argument under different decomposition yardstick

and mould

(3) de-noising, asks respectively the modulus maximum of every a line and asks the modulus maximum of each row;

(4) ask row and column to obtain the point of maximum value simultaneously, be judged to be marginal point;

(5) press dual threshold method, the marginal point under each decomposition scale is connected into a continuous border.

Further, step (3) in, the operation of described de-noising, is specially:

Consider the extreme point that the impact of noise causes, adopt threshold denoising.

Further, (5) described step, is specially:

Adopt binarization method, the gray-scale value at modulus maximum point place is made as to 255, and other non-marginal point is set to 0, the bianry image of gained is exactly required edge image.

Further, described step c, specifically comprises:

(1) the profile of the initial regular polygon generating based on single Seed Points develops;

(2) the initial polygonal profile generating based on longitudinal axis develops;

(3) the rough initial polygonal profile directly sketching out based on user develops.

Further, (1) described step, is specially:

1. determine from its geometric center to each summit and the bearing of trend of each limit mid point;

2. carry out coarse segmentation, primary summit approach and for the first time mid point approach;

3. segment and cut, loop mid point and approach.

Further, (2) described step, is specially:

The initial polygonal profile generating based on longitudinal axis, carries out mid point and approaches, and profile is carried out to intense adjustment.

Further, (3) described step, is specially:

1. direction is approached on definite summit;

2. carry out coarse segmentation, primary summit approach and for the first time mid point approach;

3. segment and cut, loop mid point and approach.

Further, 1. described step, is specially:

Employing by the mid point of two line segments that mid point definite adjacent with each summit and the line direction on this summit as bearing of trend.

The polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo of various embodiments of the present invention, comprise: to a CT image, interactive mode is obtained the initial polygonal profile (adopting the initial profile generation method based on single Seed Points, the polygonal profile that generates initial profile and draw based on direct interaction based on longitudinal axis) in region to be split; Carry out the rim detection based on Wavelet Modulus Maxima, obtain the high-frequency information of overall importance of whole CT image; Polygonal profile develops respectively automatically, approaches target area; The workload that can reduce as far as possible doctor can incorporate again the Interactive Segmentation algorithm of doctor's subjective experience preferably, thereby meets medical image, the particularly application demand in medicine CT segmentation of sequence image field; Thereby can overcome the defect of poor for applicability in prior art and very flexible, to realize the advantage that applicability is good and dirigibility is good.

Because medicine CT image is generally comprised of one group of CT sequence image, the present invention embodiment subsequently plans polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo by the projecting method of adjacent layer, for one group of CT sequence image, form a medicine CT segmentation of sequence image system, give full play to the advantage of algorithm of the present invention, illustrate better and embody algorithm of the present invention applicability and dirigibility in actual applications.

To a set of CT sequence image, as long as select the key-course with obvious characteristic to carry out the Interactive Segmentation of single image, then successively its initial profile is projected to adjacent layer, last simultaneously carrying out develops based on the profile of Wavelet Modulus Maxima, just can obtain very soon the segmentation result of whole image sequence.

Other features and advantages of the present invention will be set forth in the following description, and, partly from instructions, become apparent, or understand by implementing the present invention.

Below by drawings and Examples, technical scheme of the present invention is described in further detail.

Accompanying drawing explanation

Accompanying drawing is used to provide a further understanding of the present invention, and forms a part for instructions, for explaining the present invention, is not construed as limiting the invention together with embodiments of the present invention.In the accompanying drawings:

Fig. 1 is the initial profile projection theory figure of polygonal profile evolution medicine CT image Interactive Segmentation method adjacent figure layer during for one group of CT sequence image of the present invention is based on small echo;

Fig. 2 the present invention is based on single Seed Points O in the polygonal profile evolution medicine CT image Interactive Segmentation method of small echo to generate regular polygon profile P figure;

Fig. 3 is the bearing of trend figure that the present invention is based on the initial profile generating based on single Seed Points in the polygonal profile evolution medicine CT image Interactive Segmentation method of small echo;

Fig. 4 the present invention is based on profile approximate algorithm flow process in the polygonal profile evolution medicine CT image Interactive Segmentation method of small echo;

Fig. 5 the present invention is based on profile approximation theory schematic diagram in the polygonal profile evolution medicine CT image Interactive Segmentation method of small echo;

Fig. 6 the present invention is based on summit approximate procedure process flow diagram in the polygonal profile evolution medicine CT image Interactive Segmentation method of small echo;

Fig. 7 is the definite figure that the present invention is based on summit bearing of trend in the polygonal profile evolutionary process generating based on direct interaction in the polygonal profile evolution medicine CT image Interactive Segmentation method of small echo;

Fig. 8 the present invention is based in the polygonal profile evolution medicine CT image Interactive Segmentation method of small echo to generate complex outline figure by multiple longitudinal axis reference mark; (a) longitudinal axis line chart of initial profile; (b) the bearing of trend figure of each section of longitudinal axis; (c) the polygonal profile figure generating;

Fig. 9 is the polygonal profile evolution medicine CT image Interactive Segmentation method overview flow chart that the present invention is based on small echo.

Embodiment

Below in conjunction with accompanying drawing, the preferred embodiments of the present invention are described, should be appreciated that preferred embodiment described herein, only for description and interpretation the present invention, is not intended to limit the present invention.

From the angle of medical science clinical practice, according to the embodiment of the present invention, as shown in Fig. 1-Fig. 9, polygonal profile evolution medicine CT image Interactive Segmentation based on small echo method is provided, and be used for one group of CT sequence image by the method for adjacent layer projection, be specially the workload that can reduce as far as possible doctor and can incorporate preferably again the Interactive Segmentation algorithm of doctor's subjective experience, thereby meet the application demand that medical image (particularly medicine CT sequence image) is cut apart field.Wherein:

Image is cut apart: image is cut apart exactly image is divided into several specific, to have the region of peculiar property and propose interesting target technology and processes; It is the committed step of being processed graphical analysis by image.

Similarity criteria: for judging whether image two sub regions (or single pixel) meet the regulation of certain similarity.As: based on the similarity criteria of gray scale difference, the similarity criteria based on subregion etc.

Polygonal profile evolution medicine CT image Interactive Segmentation method that should be based on small echo, can, according to concrete target area and picture quality flexible configuration, have very large elasticity, can meet medical image various different levels cut apart demand; Can be directly used in clinical diagnosis treatment, can be also more complicated research and processing, and Data support is provided.

As shown in Fig. 1-Fig. 9, about the polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo of the present embodiment, be used for the specific implementation process of one group of CT sequence image, referring to following explanation.

(i) the initial profile projection of adjacent figure layer:

For a set of CT sequence image, the single image data of each layer of section have certain similarity and continuity.According to this feature, the segmentation result of last layer can directly apply to current layer to be split sometimes, or the initial profile that its initial profile can be used as current layer directly further approaches and cuts apart.Therefore, the further polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo, for CT sequence image, form a medicine CT segmentation of sequence image system that can meet multiple demand and practicality and high efficiency, fully shown dirigibility, high efficiency and the practical value of Interactive Segmentation method.

To a set of CT sequence image, as long as select the key-course interactive mode with obvious characteristic, obtain initial polygonal profile, then successively its initial profile is projected to adjacent layer, finally carry out developing based on the profile of Wavelet Modulus Maxima, just can obtain very soon the segmentation result of whole image sequence.

As shown in Figure 1, at key-course, the initial profile P forming by user interactions, can be directly by projecting to the initial profile P' of adjacent layer as current layer, carry out profile and approach and cut apart.

(ii) the rim detection based on Wavelet Modulus Maxima:

Wavelet theory is the important achievement in signal time frequency analysis field in recent years, has more complete Fundamentals of Mathematics, in engineering practice, is also widely applied ^[4,5].High-frequency information can be identified and detect to many resolution characteristics of wavelet analysis well, therefore uses wavelet analysis also can carry out rim detection, obtains the multi-scale edge of image.For cutting apart of medicine CT image, its advantage is apparent.

Conventional Wavelet Edge Detection algorithm utilizes the modulus maximum of wavelet conversion coefficient to carry out rim detection.On mathematics, can prove, when the first order derivative of a certain smooth function is regarded as to wavelet function, the catastrophe point of the module maximum point respective signal of wavelet transformation.Therefore, can carry out by the method that detects wavelet coefficient modulus maximum the catastrophe point of detection signal.

Thinking is above extended to two-dimentional situation.For image f (x, y), establishing θ (x, y) is a smooth function, order:

${\mathrm{\ψ}}^1(x,y)=\frac{\∂}{\∂x}\mathrm{\θ}(x,y)---\left(1\right);$

${\mathrm{\ψ}}^2(x,y)=\frac{\∂}{\∂y}\mathrm{\θ}(x,y)---\left(2\right);$

With ψ ¹(x, y) and ψ ²(x, y) is as wavelet function, at yardstick 2 ^jtime wavelet transformation be:

$W_{2^j}^{1}f(x,y)=f*{\mathrm{\ψ}}_{2^j}^1(x,y),W_{2^j}^{2}f(x,y)=f*{\mathrm{\ψ}}_{2^j}^2(x,y);$

Two components of wavelet transformation with

respectively with image f (x, y) warp

after level and smooth, along the partial derivative of horizontal and vertical direction, be directly proportional.Like this, two-dimensional wavelet transformation is just equivalent to gradient, and its argument and mould are respectively:

$W_{2j}f(x,y)=\sqrt{{\left|W_{2^j}^{1}f(x,y)\right|}^2+{\left|W_{2^j}^{2}f(x,y)\right|}^2}---\left(3\right);$

$A_{2j}f(x,y)=\arctan \left(\frac{W_{2^j}^{2}f(x,y)}{W_{2^j}^{1}f(x,y)}\right)---\left(4\right);$

Conventionally θ (x, y) is taken as Gaussian function, and the details component of wavelet decomposition reflects the local gray level catastrophe point of image truly, and wavelet local maxima point has been determined the position of picture signal catastrophe point.So, as long as detect the Local modulus maxima of wavelet conversion coefficient mould along gradient direction, just can detect the edge of image.Utilize the basic step of Wavelet Modulus Maxima rim detection to be summarized as follows:

To under dimensioning, the two-dimensional wavelet transformation of computed image

decomposition scale can be determined according to specific needs;

(2) calculate the argument under different scale

and mould

(3) de-noising, asks respectively the modulus maximum of every a line and asks the modulus maximum of each row;

(4) ask row and column to obtain the point of maximum value simultaneously, be judged to be marginal point.

(5) by dual threshold method, the marginal point under each yardstick is connected into a continuous border.

Step (3) in, consider the extreme point that the impact of noise causes, generally adopt threshold denoising.(5) step also can adopt binarization method, the gray-scale value at modulus maximum point place is made as to 255, and other non-marginal point is set to 0, and the bianry image of gained is exactly required edge image.

(iii) the image segmentation algorithm basic step that the interactive profile based on Wavelet Modulus Maxima rim detection develops

With above-mentioned, based on Wavelet Modulus Maxima edge detection method, calculate the Wavelet Modulus Maxima of each yardstick, and carry out multi-scale morphology with this, finally can obtain binaryzation edge image.On this basis, algorithm can approach thought further combined with interactive profile cuts apart the data that extract target area, forms the image segmentation algorithm that a complete interactive profile based on Wavelet Modulus Maxima rim detection develops.In general, the performing step of algorithm is as follows:

(1) by man-machine interaction, generate seed region;

(2) calculate the Wavelet Modulus Maxima of each yardstick, and carry out multi-scale morphology with this, finally obtain binaryzation edge image;

(3) seed region orthogenic evolution, approaching to reality border.

From cutting apart flow process, can obviously find out: it can be also multiple that target area can be one, and unrestricted; Algorithm has been given full play to the guide effect on border and the innate advantage of outline line evolution thought, and whole cutting procedure is that a user carries out the mutual cutting procedure of controlling targetedly.

The process that profile evolution approaches target area is an interactive process.First doctor is aided with manual operation in various degree according to the complexity of image, thereby generates an initial profile.Initial profile auto-feeding obtains final target area.

According to the characteristic of spatial distribution of target area, can be divided into different types by the complicacy that generates initial profile: the profile generating algorithm based on single Seed Points, profile generating algorithm based on longitudinal axis, also can directly sketch the contours roughly a polygon as initial profile.Describe respectively below.

1) the evolution cutting procedure of the initial profile generation method based on single Seed Points

(1) single Seed Points generates initial regular polygon profile

The straightforward procedure of introducing initial profile is chosen exactly a unique point and is generated seed region.First doctor specifies a Seed Points O (x according to the concrete condition of target area to be split ₀, y ₀), then algorithm, centered by Seed Points, generates a simple regular polygon region P, automatically as seed region.Choose regular polygon as initial profile, how much distribution rules, are convenient to calculate.As Fig. 2, illustrated to have generated a foursquare profile P by Seed Points O.

(2) profile orthogenic evolution direction

After seed region is determined by initial profile, the direction that profile orthogenic evolution is required, can be determined by two particular points on profile.For this regular polygon, its geometric center, to each summit, in the direction of each limit mid point, can be determined a growth bearing of trend.Fig. 3 is a foursquare profile bearing of trend.

Like this, bearing of trend can be controlled the evolution shape of this initial polygonal profile.According to certain growth criterion, the growth that initial profile can carry out from its geometric center to each summit and the bearing of trend of each limit mid point is developed.

(3) polygonal profile evolution principle

The process of the object boundary of outline line evolution approaching to reality is two processes substantially: summit is approached with mid point and approached.The method that adopts the gray scale difference based on single pixel with mid point approximate procedure is approached on summit.Algorithm basic procedure as shown in Figure 4.

Analyze process flow diagram above known, profile approximate procedure shows as one on the whole by slightly to smart cutting procedure.Seed region develops by directed deformation, approaches the border of target area.

Coarse segmentation process: primary summit approach and for the first time mid point approach.This process has been sketched the contours the shape of target area substantially, and the formation of general outline is had to decisive meaning.Coarse segmentation process forms the new outline polygon of two times that a limit number is initial polygon limit number while finishing.

Thin cutting procedure: mid point approaches the cycle stage.Further adjust shape and the precision of profile.When thin cutting procedure finishes, each limit of polygonal profile should be evolved into a pixel.

Fig. 5 is using an equilateral triangle as initial profile, simply illustrated the profile whole process of approaching that develops.In Fig. 5, (a) illustrated the result of choosing of seed region; (b) illustrated the result that summit is approached; (c) illustrated the evolution direction that mid point approaches; (d) illustrated coarse segmentation result (primary summit approach and for the first time mid point approach) result; (e) be the result after n mid point approaches.

(4) key step and correlation parameter

1. the limit number of initial regular polygon

The initial regular polygon generating can be that positive triangle can be also square.In theory, the limit number of this regular polygon is more, and determined bearing of trend is also just more, initial coarse segmentation (for the first time summit approach and for the first time mid point approach) effect just better.Here adopt the initial profile line of regular pentagon as seed region.

2. the size of seed region

Due to the driving factor that has adopted the similar characteristic of subregion to develop as profile, the size of subregion is an important factor that affects subregion similar features.Subregion should not be too large, but too little subregion is not enough to express the gray feature in region.Whether the size of subregion can express the gray feature in region more exactly, is along with different images degree of strength affected by noise is determined.Generally should, for different targets to be split, first use measuring, the method that provides a reference size is relatively applicable to the clinical practice that medicine CT image is cut apart.

3. the storage of polygonal profile

In order further to carry out profile, develop and process and the generation of final outline data, seed region can adopt the storage of endless chain coding mode.Adopt chain code to represent that the technology of lines is widely used in computer graphics.It utilizes a series of short lines sections with length-specific and direction to describe object boundary.Because the length of short lines is fixed and direction number limited (conventional have neighbours territory chain code and eight neighborhood chain codes), therefore, whole chain code only needs the absolute coordinates of record start point, and other point can represent by direction side-play amount.Chain representation does not need to record the coordinate of point in it again, has saved to a certain extent storage space.

But chain representation is also improper for a movable outline line that has evolution properties.Here adopted a kind of scheme of compromise.Adopt the dynamic coordinate chain that (counter clockwise direction or clockwise direction) arranges in certain sequence to represent polygonal profile.

In the incipient stage, dynamic coordinate chain generates a regular polygon automatically around Seed Points.As long as this regular polygon is the current apex coordinate of storage.

Along with the evolution of polygonal profile, each apex coordinate value of outline polygon also constantly changes.In fact, after the process of approaching at beginning mid point, the differentiation on mid point and summit disappears---and each limit mid point of initial profile carries out orthogenic evolution as the summit of new polygonal profile.If initial profile P is quadrilateral ABCD, its each limit mid point is M _aB, M _bC, M _cD, M _dA.After mid point approaches for the first time, profile P is evolved into AM _aBbM _bCcM _cDdM _dA.Then, can calculate new middle point coordinate.Like this, in evolutionary process, increase along with the process developing on the summit of initial shape changeable.Obviously, the limit number of outline polygon is dynamic change, and the limit number of final outline line is uncertain.Along with profile more and more approaches object boundary, polygonal each limit is finally a pixel size.In order to guarantee the order of the Coordinate Chain in evolutionary process, new apex coordinate must be inserted to suitable position according to the order of sequence.This process can be encapsulated as a function InsertMidPt(& P, & M), wherein P is the Coordinate Chain in developing, the Coordinate Chain that M is new summit.

4. the calculating of bearing of trend

In computing machine, what the geometric center of regular polygon was determined to arbitrfary point is a line segment.Obtain the initial profile extension point approaching in direction that develops and also need a computation process.The bearing of trend of two end points definite is equivalent to a straight line generative process.Coordinate by two-end-point is determined bearing of trend, is in fact a process of finding the pixel sequences (having demonstration can insert if desired color data) of best fit straight line.Conventional algorithm has DDA algorithm, mid point Bresenham algorithm and improvement algorithm thereof.

5. summit approximate procedure

Summit was approached in the coarse segmentation stage, and the Approximation effect of objective contour is had to material impact.

The limit number of supposing polygon P is n, and each apex coordinate is V _i(x _i, y _i), wherein 0<i≤n.From Seed Points O (x ₀, y ₀) to each summit, determined a vectorial K _i(x _i-x ₀, y _i-y ₀).From computer graphical, gained knowledge, from summit V _i(x _i, y _i) along vectorial K _isensing, polygonal profile P can extend outward evolution.For K _im in direction and m+1 pixel, can compare their gray-scale value.Because the rim detection stage has been considered the impact of noise, in order to raise the efficiency, the similarity criteria that adopts the method for the gray scale difference based on single pixel to approach as profile.When satisfy the demand similarity criteria time, K _idirection growth point of arrival S _i(x _i, y _i) stop growing, revise summit V _i(x _i, y _i) coordinate be S _i(x _i, y _i).At this moment, this vector K _ithe boundary of corresponding arrival target area, summit.To K _ithe summit approximate procedure of direction can be encapsulated as function Vf (& P, a K _i), its quick-reading flow sheets is as shown in Figure 6.

6. mid point approximate procedure

Mid point approaches and belongs to coarse segmentation process for the first time, the same with summit approximate procedure, and the formation of objective contour is had to decisive influence.After this approximate procedure is that mid point approaches completely.Be the adjustment to coarse contour line in essence, improve approximation accuracy.When summit approximate procedure finishes, in general initial polygonal profile can not be still a regular figure.At this moment, determine along the direction of growth of mid point and should adopt the method when being different from summit and approaching.

If polygon vertex is arranged in the direction of the clock in turn, any limit L of polygon _itwo-end-point be V _i(x _i, y _i) and V _i+1(x _i+1, y _i+1), at limit L _iin perpendicular bisector direction and point to the vector M of outside of polygon _i, the vector that can consist of two-end-point is around its mid point V _mbeing rotated counterclockwise 90 ° obtains.Like this, polygonal profile P can be along vector M _iby similarity criterion (gray scale difference), extend outward evolution.With summit approach similar, when mid point approximate procedure finishes, the mid point V of corresponding edge _marrive the boundary in region to be split.This process also can be encapsulated as mid point approximating function Vf (& P, M _i).

7. the end condition of algorithm

On summit for the first time, approach with mid point and approach while finishing, the mid point on each limit can be used as the summit of new polygonal profile.On new outline polygon basis, algorithm loops mid point approximate procedure.If when the polygonal current length of side is a pixel, the mid point approximate procedure on this limit finishes; Otherwise repeat mid point, approach, until polygonal each limit is all a pixel.At this moment the outline line forming is exactly the target area that will cut apart.

8. the scope of application of algorithm

Initial profile is a convex polygon all the time in the process of approaching target area, and therefore target area also should have certain convexity-preserving.Due in the target area of medical image except blood vessel, the extraorgan of the long and narrow types of minority such as enteron aisle, generally has the good and more level and smooth feature in edge of convexity-preserving, with this, goes for the demand of cutting apart of the most of target areas of medicine CT image.

2) the evolution cutting procedure of the polygonal profile generating based on direct interaction

By choosing multiple reference mark as a polygonal summit, directly generate more complicated initial profile, thereby can widen the scope of application of algorithm.

(1) algorithm idea

Mutual by necessity, user directly sketches out a rough initial polygon as seed region.On this basis, the advantage of performance profile approximatioss, just can make in theory algorithm be applicable to convexity-preserving is not cutting apart of good various complex regions, but needs more man-machine interaction.

(2) extend the expansion of mode

To a polygonal profile, in fact there are two summit bearing of trends: outwards with inside.Thus, can obtain two kinds and approach scheme: the one, by man-machine interaction, obtain one or more reference mark, algorithm calculates and generates an initial closed outline, and initial profile inside is exactly seed region, and then initial profile stretches out and approaches object boundary; The one, by interactive operation, obtain a coarse contour that surrounds on the whole target area, then object boundary is inwardly approached on each summit, obtains connecing in one and the profile of target area, follows each limit mid point and still outwards develops and approach.

These two kinds different extension modes can be used in conjunction with: as an example of the mode of approaching from inside to outside example, illustrate here.

(3) the calculating of bearing of trend

Due to polygonal profile at this moment, be no longer a regular polygon, its summit is approached direction and need to be taked different modes to determine, and definite mode of bearing of trend mid point and mid point still remains unchanged.Can adopt by two mid points adjacent with each summit and determine summit bearing of trend, its principle as shown in Figure 7.

In Fig. 7, M1, M2 is two mid points that arbitrary summit V is adjacent (can be determined by polygonal apex coordinate), M is the mid point (can be determined by the coordinate of M1 and M2) of line segment M1M2.This sampling point M and summit V have determined the bearing of trend on summit.

(4) algorithm flow

Only generate the process of initial polygonal profile different from the algorithm flow that the profile based on single Seed Points approaches, repeat no more here.

3) the evolution cutting procedure based on longitudinal axis generation initial profile

Adopt the profile approximate algorithm of single Seed Points too brief, and the medical image cutting method approaching based on interactive polygonal profile may can be too complicated.The present invention proposes one and guides as topology using multiple reference mark, generates a polygonal profile that is applicable to long and narrow region.How by possible multiple spot distribution, to generate rational initial profile is the key issue of algorithm.

(1) the guiding of multiple reference mark generates polygonal profile

The main process of algorithm: first, draw longitudinal axis by user interactions.Then each section of axis is by extending (pressing subregion similarity criteria) perpendicular to the direction of axis to two sides, and every section of longitudinal axis will obtain two frontier points like this.Finally, last end points connecting in certain sequence on each segment boundary point and longitudinal axis forms a closed polygonal profile.This process is equivalent to a coarse segmentation process.

In order to illustrate algorithm thinking, suppose, by user interactive, to have obtained longitudinal axis A ₁a ₂a _n, this is a multi-section-line, n is longitudinal axis end points number.For longitudinal axis A ₁a ₂section, according to the knowledge of computer graphics, can be by A ₁a ₂turn clockwise 90 ° and obtain an A ₁', by A ₁a ₂be rotated counterclockwise 90 ° and obtain an A ₁' ', like this, vectorial A ₁a ₁' and A ₁a ₁' ' can determine terminal A ₁two bearing of trends.Same mode can obtain A ₂, A ₃..., A _n-1bearing of trend.Then, by similarity criteria, the end points of longitudinal axis is extended take pixel as unit step-length.If terminal A ₁press subregion similarity criteria respectively at a M ₁, M ₁' stop M ₁and M ₁' two summits of the polygonal profile that will construct exactly.According to the same manner, can obtain M ₂, M ₂' ..., M _n-1, M _n-1'.To last terminal A _n, according to vectorial A _n-1a _nthe determined direction of section, extends to M by subregion similarity criteria _n.Finally, storage apex coordinate chain M ₁m ₂m _n-1m _nm _n-1' ... M ₂' M ₁', construct desired polygonal profile.Said process can be referring to Fig. 8 (establishing longitudinal axis be ABCDEFG).

(2) the line segment length between 2 of longitudinal axis

In user interaction process, the length of the longitudinal axis line segment of painting will guarantee to form convex polygon.For more complicated target area, in order to guarantee the convexity-preserving of profile, in the not so good place of object boundary convexity-preserving, should increase number of endpoint, make the limit of the polygonal profile forming not exceed as far as possible object boundary, or remain in certain permissible range; Meanwhile, end points number again can not be overstocked, to avoid bearing of trend overlapping.

(3) mid point approximate procedure

The process that generates outline line according to longitudinal axis is a coarse segmentation process, therefore, and in this case, no longer there is independently summit approximate procedure, but directly call mid point approximating function, and profile is carried out to intense adjustment, its mechanism and single Seed Points partitioning scheme are roughly the same.

(4) the scope of application

Profile based on longitudinal axis approaches medical image cutting method for long and narrow histoorgan, more can give play to advantage.For long and narrow target area, longitudinal axis can be together in series banded region compactly.

4) the polygonal profile evolution medicine CT image Interactive Segmentation method brief summary based on small echo

Known by above-mentioned labor and description, to any CT image, the first interactive initial polygonal profile in region to be split that obtains of user.Be directed to the target area of different characteristic, the initial profile generation method based on single Seed Points that can adopt the present invention to propose, the polygonal profile that generates initial profile and draw based on direct interaction based on longitudinal axis.Then, CT image is carried out to the rim detection based on Wavelet Modulus Maxima, obtain the high-frequency information of overall importance of whole CT image.Finally, polygonal profile develops respectively automatically, approaches target area.The final segmentation result that obtains CT sequence image.The overall procedure of algorithm can be concluded as shown in Figure 9.

(iv) medicine CT sequence image Interactive Segmentation system

Further above-mentioned algorithm is used for to whole CT sequence image, can forms a medicine CT segmentation of sequence image system that can meet multiple demand, fully show dirigibility, high efficiency and the practical value of Interactive Segmentation method.

System adopts layer-management mechanism to realize the management to user interactive data and partition data.After the initial profile of system by user interactions control acquisition key-course, call corresponding algorithm and realize cutting apart of whole CT sequence image.The data of cutting apart are still retained in figure layer object separately.

(v) the operating process of medicine CT sequence image Interactive Segmentation system

By the thinking of key-course projection, medicine CT segmentation of sequence image serial algorithm is integrated, it is as follows that the main behaviour of the medicine CT sequence interactive image segmentation scheme obtaining does step:

(1) the pre-service of CT sequence image.Opening after CT sequence image, first it is carried out to pre-service.Pretreated object is the feature for concrete segmentation object and CT image, and it is carried out necessary denoising, strengthens and process, and strengthens the display effect of image, outstanding target area.The view data of DICOM form comprises abundant information, before segmentation algorithm, if can make full use of these information, adjusts the data display effect of image, can effectively strengthen the segmentation effect of whole CT sequence image.

(2) obtain the initial profile of key-course.In CT sequence, choose those objective contours roughly similar section carry out interactive operation, obtain an initial profile.Better for convexity-preserving, border is more level and smooth, and the region to be split of simple shape can adopt single-point (single Seed Points method), multi-section-line (longitudinal axis collimation method), initial polygon (manually coarse segmentation method) to obtain.Before the details of this reciprocal process, done detailed introduction, repeated no more here.

(3) according to the initial profile projection theory of adjacent layer section, the initial profile of key-course is projected to adjacent layer, thereby obtain the initial profile of whole each figure layer of CT sequence image.

(4) cut apart whole CT sequence image.Successively calling related function (polygonal profile evolution algorithmic) realizes cutting apart whole CT sequence.

(5) preserve partition data.Finally, can browse overall segmentation effect with three dimensional constitution, or successively browse the Approaching Results of gained.To corresponding undesirable profile, can revise last save data.

Adopt the dynamic coordinate chain that (counter clockwise direction or clockwise direction) arranges in certain sequence to represent polygonal profile.As long as this polygon is the coordinate on the summit constantly generating in storage profile evolutionary process.

The polygonal profile evolution medicine CT image Interactive Segmentation method of the various embodiments described above of the present invention, i.e. the interactive polygonal profile evolution medicine CT image partitioning algorithm based on small echo, performed well in the cutting apart of a set of medicine CT sequence image.Explanation using this as an exemplary embodiments, has embodied applicability and the dirigibility of algorithm of the present invention well.

User (can adopt the initial profile generation method based on single Seed Points alternately by the key-course with the meaning of representing being carried out to the simple of necessity, the polygonal profile that generates initial profile and draw based on direct interaction based on longitudinal axis), generate an initial polygonal profile.Then according to the principle of adjacent layer projection, obtain the initial profile of each layer in CT sequence image.Then, algorithm carries out the rim detection based on Wavelet Modulus Maxima automatically, obtains each layer of high-frequency information of overall importance, and on this basis, the polygonal profile of each layer develops respectively automatically, approaches target area.

In an embodiment, calculating ratio juris and do not become, is in fact that each layer of CT sequence image is used respectively to algorithm of the present invention.For improving the interactive mode of each layer, obtain the process of initial profile, improve the service efficiency of algorithm in CT sequence image, adopted the method for general adjacent layer projection, and the primitive data of individual figure layer is carried out to layer-management.The principle of these introductions is only the auxiliary mechanism in the preferred embodiments of the present invention, is not limited to specifically use flexibly of the present invention.

This polygonal profile evolution medicine CT image Interactive Segmentation method, has following characteristics:

1) both work in coordination rim detection and profile to be developed, and combine together.Taken into account the overall permanence in region to be split and the local feature of image border in image, the two large theories that image is cut apart have been incorporated into together.Many resolution characteristics of Wavelet Modulus Maxima edge detection method, make edge detection process have good anti-noise ability; Meanwhile, the result of rim detection provides good support for profile develops to approach, and can obtain good segmentation result.Algorithm not only can extract the outline data of one or more target areas, and practical flexibly, can adapt to the demand of cutting apart of different complexities.

2) can further algorithm be used for to CT sequence image, form one and can on different complicated levels, effectively bring into play algorithm performance medicine CT segmentation of sequence image system, fully show dirigibility, high efficiency and the practical value of Interactive Segmentation method.

The polygonal profile evolution medicine CT image Interactive Segmentation method of the various embodiments described above of the present invention, also has following characteristics:

1), according to the basic ideas of algorithm, the similarity criteria of polygonal profile evolutionary process, can adopt more generally criterion flexibly, in these thoughts, is contained among this algorithm.

2) adopt the similarity criteria based on gray scale difference.Because many resolution characteristics of Wavelet Modulus Maxima edge detection method, make edge detection process consider noise factor common in medical image segmentation.Therefore, on this basis, adopt the simple similarity criteria based on gray scale difference to raise the efficiency.

In sum, the polygonal profile evolution medicine CT image Interactive Segmentation method of the various embodiments described above of the present invention, at least can reach following beneficial effect:

(1) from cutting apart theory, this algorithm rim detection and profile are developed both work in coordination, combine together.The result of rim detection provides good support for profile develops to approach, and more easily obtains desirable segmentation result.

(2) many resolution characteristics of Wavelet Modulus Maxima edge detection method, make edge detection process have good anti-noise ability, make algorithm be more suitable for complicated medical image.

(3) operate nature simple and direct, there is very large elasticity, can meet different complexities medical image cut apart demand, completely can be in clinical treatment, also can be more complicated research and processing, as the three-dimensional reconstruction of histoorgan, sham operated, Visual Teaching etc., provide basic platform and Data support.

List of references:

[1] Wang Jianqing, Guo Min, Xu Qiuping. the target extraction algorithm that border combines with region. computer utility .2009,29(9): 2411-2413.

[2] Liu Junwei. the Research on Method of Image Segmentation .(doctorate paper based on level set). Hefei: China Science & Technology University, 2009.

[3] Zhu Guopu. the image based on movable contour model is cut apart .(doctorate paper). Harbin: Harbin Institute of Technology, 2007.

[4] Liu Tao, Zeng Xiangli, Zeng Jun etc. practical wavelet analysis introduction. Beijing: National Defense Industry Press, 2006.

[5] Ingrid Daubechies.Ten Lectures on Wavelets. Beijing: National Defense Industry Press, 2004.

Finally it should be noted that: the foregoing is only the preferred embodiments of the present invention, be not limited to the present invention, although the present invention is had been described in detail with reference to previous embodiment, for a person skilled in the art, its technical scheme that still can record aforementioned each embodiment is modified, or part technical characterictic is wherein equal to replacement.Within the spirit and principles in the present invention all, any modification of doing, be equal to replacement, improvement etc., within all should being included in protection scope of the present invention.

Claims (12)

1. the polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo, is characterized in that, comprising:

A, by man-machine interaction, generate the initial profile in region to be split in CT image;

B, CT image is carried out to multi-scale morphology, obtain high frequency edge information of overall importance;

C, based on initial profile and high frequency edge information of overall importance, carry out initial profile orthogenic evolution, approaching to reality border, obtains final target area.

2. the polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo according to claim 1, is characterized in that, described step a, specifically comprises:

According to the space distribution of target area, select flexibly the profile generating algorithm based on single Seed Points, the profile generating algorithm based on longitudinal axis, the polygon that also can sketch the contours roughly a sealing is as initial profile.

3. the polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo according to claim 2, is characterized in that, the described profile generating algorithm based on single Seed Points, is specially:

First user specifies a Seed Points according to the concrete condition of target area to be split

, then algorithm, centered by Seed Points, generates a simple regular polygon region automatically

, as seed region.

4. the polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo according to claim 2, is characterized in that, the described profile generating algorithm based on longitudinal axis, is specially:

First, by user interactions, draw longitudinal axis;

Then, each section of axis extends to two sides automatically by the direction perpendicular to axis, and it is definite that the stop condition of extension is pressed similarity criteria, and every section of longitudinal axis will obtain two frontier points like this;

Finally, by clockwise or counterclockwise last end points being linked in sequence on each segment boundary point and longitudinal axis, form a closed polygonal profile.

5. the polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo according to claim 2, is characterized in that, described multi-scale morphology specifically adopts the edge detection method of Wavelet Modulus Maxima, comprising:

(1) based on by the given decomposition scale of real needs, calculate the two-dimensional wavelet transformation of the CT sequence image of composition CT image to be split

;

(2) calculate the argument under different decomposition yardstick

and mould ;

(3) de-noising, asks respectively the modulus maximum of every a line and asks the modulus maximum of each row;

(4) ask row and column to obtain the point of maximum value simultaneously, be judged to be marginal point;

(5) press dual threshold method, the marginal point under each decomposition scale is connected into a continuous border.

6. the polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo according to claim 5, is characterized in that, step (3) in, the operation of described de-noising, is specially:

Consider the extreme point that the impact of noise causes, adopt threshold denoising.

7. the polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo according to claim 5, is characterized in that, (5) described step, is specially:

Adopt binarization method, the gray-scale value at modulus maximum point place is made as to 255, and other non-marginal point is set to 0, the bianry image of gained is exactly required edge image.

8. the polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo according to claim 1, is characterized in that, described step c, specifically comprises:

(1) the profile of the initial regular polygon generating based on single Seed Points develops;

(2) the initial polygonal profile generating based on longitudinal axis develops;

(3) the rough initial polygonal profile directly sketching out based on user develops.

9. the polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo according to claim 8, is characterized in that, (1) described step, is specially:

(1) determine from its geometric center to each summit and the bearing of trend of each limit mid point;

(2) carry out coarse segmentation, primary summit approach and for the first time mid point approach;

(3) segment and cut, loop mid point and approach.

10. the polygonal profile evolution medicine CT image Interactive Segmentation method based on small echo according to claim 8, is characterized in that, (2) described step, is specially:

The initial polygonal profile generating based on longitudinal axis, carries out mid point and approaches, and profile is carried out to intense adjustment.

The 11. polygonal profile evolution medicine CT image Interactive Segmentation methods based on small echo according to claim 8, is characterized in that, (3) described step, is specially:

(1) direction is approached on definite summit;

(2) carry out coarse segmentation, primary summit approach and for the first time mid point approach;

(3) segment and cut, loop mid point and approach.

The 12. polygonal profile evolution medicine CT image Interactive Segmentation methods based on small echo according to claim 11, is characterized in that, (1) described step, is specially:

Employing by the mid point of two line segments that mid point definite adjacent with each summit and the line direction on this summit as bearing of trend.

Images