CN105957066A - CT image liver segmentation method and system based on automatic context model - Google Patents

Abstract

The invention discloses a CT image liver segmentation method and system based on an automatic context model and capable of increasing liver segmentation precision in a CT image. The method comprises steps of: reading a training image set and a to-be-segmented image; extracting the textural feature of each pixel in the images; classifying the feature of each pixel in the to-be-segmented image by using a classifier to obtain an initial liver probability graph; extracting the context feature of each pixel in the images; combining the context features with the textural features and learning a series of classifiers by means of iteration until convergence to obtain a liver probability graph; using the liver probability graph as prior information and adding the liver probability graph a prior constraint condition to a randomly walking objective function as to obtain a random walk model based on context constraint for segmenting the liver; achieving liver segmentation of a three-dimensional CT image layer by layer on a two-dimensional slice of the to-be-segmented image, and interpolation and complement of a liver bound discontinuous area so as to obtain a smooth continuous liver surface.

Description

CT image liver segmentation method and system based on automatic context model

Technical Field

The invention relates to the technical field of machine learning, in particular to a CT image liver segmentation method and system based on an automatic context model.

Background

Medical image segmentation assists doctors in identifying internal tissues, organs and lesion areas of patients, and plays a vital role in computer-aided therapy and surgical planning. Therefore, automatic liver segmentation is the basis for doctors to diagnose and treat liver diseases such as liver cirrhosis, liver tumors, liver transplantation, and the like. In an abdominal CT image, the difference between the gray values of the liver and the adjacent organs is small, the gray value of the liver is not uniform, the shape of the liver is different, and the liver is difficult to segment automatically and accurately. Therefore, a simple, fast and accurate liver segmentation method is urgently needed by clinicians.

The existing random walk segmentation method has the advantages of rapidness, simplicity and the like, but has poor segmentation effect on regions with low contrast in a CT image, and particularly, the liver is difficult to be effectively segmented by simply depending on gray values at the joints of the liver and adjacent organs such as large blood vessels, stomach and the like.

Disclosure of Invention

In view of this, embodiments of the present invention provide a method and a system for segmenting a liver of a CT image based on an automatic context model, which can effectively improve the segmentation accuracy of the liver in the CT image.

On one hand, the embodiment of the invention provides a CT image liver segmentation method based on an automatic context model, which comprises the following steps:

s101, reading a training image set and an image to be segmented, wherein the training image and the image to be segmented in the training image set are CT images of a liver;

s102, extracting texture features of each pixel in the training image and the image to be segmented;

s103, classifying the characteristics of each pixel of the image to be segmented by using a classifier to obtain an initial liver probability map;

s104, extracting the context characteristics of each pixel in the training image and the image to be segmented;

s105, combining the context features with the texture features, learning again to obtain a new classifier, obtaining a liver probability graph again, extracting the context features of each pixel in the training image and the image to be segmented again, repeating the algorithm, learning a series of classifiers until convergence, obtaining the probability that the pixel points belong to the liver region, and further obtaining the liver probability graph;

s106, adding the liver probability map serving as prior information serving as a prior constraint condition into a random walk objective function to obtain a context constraint-based random walk model and realize liver segmentation;

s107, realizing liver segmentation of the three-dimensional CT image layer by layer on the two-dimensional slice of the image to be segmented, and realizing interpolation and completion of discontinuous regions of a liver boundary so as to obtain a smooth and continuous liver surface.

On the other hand, the embodiment of the invention provides a system for segmenting a liver of a CT image based on an automatic context model, which comprises:

the reading module is used for reading a training image set and an image to be segmented, wherein the training image set and the image to be segmented are CT images of the liver;

the first extraction module is used for extracting the texture features of each pixel in the training image and the image to be segmented;

the classification module is used for classifying the characteristics of each pixel of the image to be segmented by using the classifier to obtain an initial liver probability map;

the second extraction module is used for extracting the context characteristics of each pixel in the training image and the image to be segmented;

the iteration module is used for combining the context features with the texture features, learning again to obtain a new classifier, obtaining a liver probability map again, extracting the context features of each pixel in the training image and the image to be segmented again, repeating the algorithm, learning a series of classifiers until convergence, obtaining the probability that the pixel points belong to the liver region, and further obtaining the liver probability map;

the segmentation module is used for taking the liver probability map as prior information and taking the liver probability map as a prior constraint condition, adding the prior information and the prior constraint condition into a random walk target function to obtain a context constraint-based random walk model and realize the segmentation of the liver;

and the filling module is used for realizing liver segmentation of the three-dimensional CT image on the two-dimensional slice of the image to be segmented layer by layer, and realizing interpolation and completion of discontinuous regions of the liver boundary so as to obtain a smooth and continuous liver surface.

According to the CT image liver segmentation method and system based on the automatic context model, on the basis of texture feature classification, context information is used as new features and iterative classification is carried out to obtain a prior model of the liver, the model is used as prior constraint to improve an energy function of a random walk algorithm to obtain a final liver segmentation result, the segmentation result is greatly improved for regions with unobvious gray contrast, and the segmentation precision of the liver in a CT image is effectively improved.

Drawings

FIG. 1 is a schematic flow chart illustrating an embodiment of a method for segmenting a liver in a CT image based on an automatic context model according to the present invention;

FIG. 2 is a schematic structural diagram of an embodiment of a system for segmenting a liver in a CT image based on an automatic context model according to the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some embodiments, but not all embodiments, of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

As shown in fig. 1, the present embodiment discloses a method for segmenting a liver in a CT image based on an automatic context model, which includes:

s101, reading a training image set and an image to be segmented, wherein the training image and the image to be segmented in the training image set are CT images of a liver;

s102, extracting texture features of each pixel in the training image and the image to be segmented;

s103, classifying the characteristics of each pixel of the image to be segmented by using a classifier to obtain an initial liver probability map;

s104, extracting the context characteristics of each pixel in the training image and the image to be segmented;

s105, combining the context features with the texture features, learning again to obtain a new classifier, obtaining a liver probability graph again, extracting the context features of each pixel in the training image and the image to be segmented again, repeating the algorithm, learning a series of classifiers until convergence, obtaining the probability that the pixel points belong to the liver region, and further obtaining the liver probability graph;

s106, adding the liver probability map serving as prior information serving as a prior constraint condition into a random walk objective function to obtain a context constraint-based random walk model and realize liver segmentation;

s107, realizing liver segmentation of the three-dimensional CT image layer by layer on the two-dimensional slice of the image to be segmented, and realizing interpolation and completion of discontinuous regions of a liver boundary so as to obtain a smooth and continuous liver surface.

In the method for segmenting the liver of the CT image based on the automatic context model, based on the texture feature classification, the context information is used as a new feature and is subjected to iterative classification to obtain a prior model of the liver, and the model is used as a prior constraint to improve the energy function of the random walk algorithm to obtain a final liver segmentation result.

Optionally, in another embodiment of the CT image liver segmentation method based on the automatic context model of the present invention, the texture feature extracted in S102 may be a Haar feature, a local binary pattern feature, a histogram of oriented gradients feature, or a co-occurrence matrix feature, and is not limited to the above four features.

Optionally, in another embodiment of the CT image liver segmentation method based on the automatic context model, the classifier in S103 is an AdaBoost classifier, a support vector machine classifier, a decision tree classifier, an artificial neural network classifier, a naive bayes classifier, or a random forest classifier using a support vector machine as a weak classifier.

Optionally, in another embodiment of the CT image liver segmentation method based on the automatic context model, the step S103 specifically includes:

defining a training image set as V_iI is 1,2 … n, and the corresponding division golden standard image is V_siAnd i is 1,2 … n, selecting a training sample point set in the training image set, and extracting texture features of the point set, so that the training point set information can be expressed as:

S₀＝{(y_t,f⁰(N_t)),t＝1,2…T}，

wherein N is_tNeighborhood image block, f, centered around a pixel with index t⁰(N_t) Texture feature, y, representing the neighborhood of pixel points indexed by t_tThe method comprises the steps of obtaining a liver classifier based on texture feature classification by utilizing an AdaBoost algorithm if a category mark corresponding to a pixel point with an index of T is adopted, wherein T is the total index number, and then, for an image V to be segmented_uExtracting the texture features of the pixel point xAnd classify to obtain pairsInitial liver posterior probability to classify maps

$p_{ux}^0(y=1|x)=\frac{e^{H_0\left(f_{ux}^0\right)}}{e^{H_0\left(f_{ux}^0\right)}+e^{-H_0\left(f_{ux}^0\right)}},$

Wherein y is the category label corresponding to the pixel point x, y is 1, which indicates that the pixel point belongs to the liver,is the posterior probability that pixel point x belongs to the liver, H₀The classifier is obtained by learning in the texture feature space, and for the image to be segmented, the initial liver posterior probability of the classification model for classifying and mapping each pixel point can be obtained.

Optionally, in another embodiment of the method for segmenting liver of CT image based on automatic context model of the present invention, the S104 is based on the current classification result, for the pixel point, a plurality of rays with equal angular intervals are led out from the pixel point as the center, sparse sampling is performed on the rays, so as to obtain the classification probability of the corresponding position as the context feature, on the classification result graph corresponding to the CT image slice where the pixel point with index t is located, starting from the pixel point, a first number of rays are led out from the pixel point at intervals of 45 °, the context position is sampled at equal intervals on each ray, and the classification probability at the position is used as the context feature P of the pixel point with index t⁰(t):

$P^0\left(t\right)=(p_{t_1}^0,p_{t_2}^0,...,p_{t_m}^0),$

Wherein, t_mAn index indicating a pixel point corresponding to the mth context position around the pixel point having the index t,is indexed by t_mThe pixel points are classified based on the liver posterior probability value of the texture feature, and the contextual features of the pixel points in the image to be segmented can be obtained by the same method.

Optionally, in another embodiment of the automatic context model-based CT image liver segmentation method of the present invention, the first number is 8.

Optionally, in another embodiment of the CT image liver segmentation method based on the automatic context model, the step S105 is specifically: and integrating texture features and context features of the image to construct a new training point set, wherein the information can be expressed as:

S₁＝{(y_t,(f⁰(N_t),P⁰(t))),t＝1...T}，

wherein f is⁰(N_t) And P⁰(t) respectively representing texture features of pixel points with index of t and context features extracted based on classification mapping, learning to obtain a new classifier by using AdaBoost algorithm again based on new features of training image combination, repeating the algorithm, integrating the texture features of the image and the context information, learning a series of classifiers until convergence,

for an image V to be segmented_uThe probability that the pixel point x in the image belongs to the liver region is obtained through q times of classifier iterative learning

$p_{ux}^q(y=1|x)=\frac{e^{H_q\left(f_{ux}^q\right)}}{e^{H_q\left(f_{ux}^q\right)}+e^{-H_q\left(f_{ux}^q\right)}},$

Wherein,the posterior probability that a pixel point x obtained by iterative learning of the classifier belongs to the liver region, q is the iterative times of the classifier in convergence, H_qIs a groupThe q-th classifier obtained by learning in the texture feature and context feature combination space,representing an image V to be segmented_uAnd combining the texture features of the pixel points x and the context features obtained based on the q-1 th classification mapping to obtain a liver probability map of the image to be segmented.

Optionally, in another embodiment of the CT image liver segmentation method based on the automatic context model, the S106 is specifically: for an image V to be segmented_uIt is converted into an undirected graph G ═ (V, E), and a set of nodes V ═ V₁,v₂...v_N}∪{l₀,l₁},v_iRepresenting graph nodes corresponding to the pixel points with index i, N being the total number of nodes, l₀And l₁End nodes representing non-liver regions and liver regions, respectively; edge set E is composed of_T-linkAnd E_N-linkIs composed of (A) a_T-linkFor two end nodes l₀、l₁And pixel node v_iThe weight of the edge of the middle edge set is respectively as follows:

$\left\{\begin{array}{c}{\mathrm{\ω}}_i^1=p_{ui}^q\\ {\mathrm{\ω}}_i^0=1-p_{ui}^q\end{array}\right.,$

wherein,is the end node l₁And pixel point v_iThe edge weight of (1) is the probability value of the pixel point with index i belonging to the liver region obtained by the context model q times of iterationIn the same way as above, the first and second,is the end node l₀And pixel point v_iThe edge weight value of (a) is the probability value that the pixel with index i does not belong to the liver region, so as to represent the prior information of the pixel in the image, E_N-linkRepresenting the connection relationship between adjacent pixels and the weight thereofThe method comprises the steps that gray values of pixel points in an image to be segmented are determined, and meanwhile, the pixel points which belong to a liver region and have a probability value of 1 are marked as seed points of the liver region; the pixel point with the probability value of 0 is marked as a seed point of a non-liver region,

based on the graph model, a new objective function with prior constraint is established by using the marked seed points to minimize:

$E_{aspatial}^s=\underset{e_{ij}\∈E}{\Σ}{\mathrm{\ω}}_{e_{ij}}{(x_i^s-x_j^s)}^2+\mathrm{\γ}(\underset{v_i\∈V}{\mathrm{\Σ}}{\mathrm{\ω}}_i^{1-s}{\left(x_i^s\right)}^2+\underset{v_i\∈V}{\mathrm{\Σ}}{\mathrm{\ω}}_i^s{\left(x_i^s-1\right)}^2),$

wherein,for an objective function with a priori constraints, e_ijThe first term of the above formula represents the edges of the pixel point with the connection index i and the pixel point with the connection index iRepresenting the original random walk objective function, the second term is a priori constraint term based on the context model, gamma is an adjusting parameter,the probability that a pixel point with index i in the image belongs to the category s, where s is {0,1}, and represents the liver category and the non-liver category, respectively, and the above formula is expressed by a matrix, so that the following can be obtained:

$E_{aspaital}^s=x^{sT}{\mathrm{Lx}}^s+\mathrm{\γ}\[x^{sT}{\mathrm{\Λ}}^{1-s}{x}^s+{(x^s-1)}^T{\mathrm{\Λ}}^s(x^s-1)\],$

wherein x is^sFor the probability that each pixel in the image node set belongs to different classes, the matrix L is the Laplacian matrix of the image to be segmented, Λ^sIs the value of the ith row on the diagonalThe diagonal matrix of (1) for solving the above formula, dividing all vertexes in a node set V of the undirected graph into a seed node set V_M(marked Point set) and unmarked Point set V_UTwo subsets, decompose the above equation and relate to x_UThe differential of (2) can be obtained:

$(L_U+{\mathrm{\γ\Λ}}_U^{1-s})x_U^s={\mathrm{\γ\Λ}}_U^S-{\mathrm{Bx}}_M^s,$

wherein L is_UA laplacian matrix of unlabeled point sets,the probability value belonging to the category s for an unmarked point,the value of the ith row on the diagonal for the unmarked point set isThe diagonal matrix of (a) is,the value of the ith row on the diagonal for the unmarked point set isA diagonal matrix ofIs obtained by mathematical derivation, B is a matrix,the probability value of the marking pixel point reaching the category s seed point for the first time is shown, namely if the pixel point with the index i is the category s seed point, the probability value is shownIf not, then,

based on the above-mentioned | V-containing_USolving the probability value from the unmarked point to the two kinds of seed points by the symmetrical positive definite linear equation set of | unknowns to obtain the maximum transition probabilityJudging the category label (i) of the pixel point with index i as a criterion, namely:

$label\left(i\right)=\left\{\begin{array}{cc}1,& x_i^1\&GreaterEqual;x_i^0\\ 0,& x_i^1<x_i^0\end{array}\right.,$

the expression shows that when the probability that the non-mark point reaches the liver seed point is obtained to be larger than or equal to the probability that the non-mark point reaches the liver seed point, the non-mark point belongs to the liver region, and therefore the final segmentation of the liver region in the image is achieved.

Optionally, in another embodiment of the CT image liver segmentation method based on the automatic context model, the step S107 specifically includes: the algorithm provided by the invention is to realize liver segmentation of a three-dimensional CT image layer by layer on a two-dimensional slice of the image, and the phenomenon of discontinuous boundary or even overlapped boundary can occur on the obtained liver boundary, and the invention realizes interpolation and completion of discontinuous regions of the liver boundary so as to obtain a smooth and continuous liver surface, and the specific calculation process is as follows:

inputting a liver segmentation result, carrying out octree decomposition on the segmentation result by utilizing a Scan-conversion algorithm, and decomposing the segmentation result into a finer subspace;

in the octree decomposition process, when the intersection points of all the decomposition lines and the original model are positioned at the leaves of the octree, stopping decomposition;

marking boundaries with intersections as "intersecting edges";

randomly selecting a vertex P from an original model, and marking the vertex P as 0; expanding the label along the boundary of the octree, changing the label into '1' when the 'intersecting edge' passes once, and so on, changing the label once when the 'intersecting edge' passes once until the traversal of the whole octree is finished;

and precisely reconstructing the vertexes only containing '0' and '1' by using a Dual containment algorithm to obtain the model after the cavity is filled.

As shown in fig. 2, the present embodiment discloses a system for segmenting liver in CT image based on automatic context model, which includes:

the system comprises a reading module 1, a calculating module and a calculating module, wherein the reading module 1 is used for reading a training image set and an image to be segmented, and the training image set and the image to be segmented are CT images of the liver;

the first extraction module 2 is used for extracting the texture features of each pixel in the training image and the image to be segmented;

the classification module 3 is used for classifying the characteristics of each pixel of the image to be segmented by using a classifier to obtain an initial liver probability map;

the second extraction module 4 is used for extracting the context characteristics of each pixel in the training image and the image to be segmented;

the iteration module 5 is used for combining the context features with the texture features, learning again to obtain a new classifier, obtaining a liver probability map again, extracting the context features of each pixel in the training image and the image to be segmented again, repeating the algorithm, learning a series of classifiers until convergence, obtaining the probability that the pixel points belong to the liver region, and further obtaining the liver probability map;

the segmentation module 6 is used for adding the liver probability map serving as prior information serving as a prior constraint condition into a random walk objective function to obtain a context constraint-based random walk model and realize the segmentation of the liver;

and the filling module 7 is used for realizing liver segmentation of the three-dimensional CT image on the two-dimensional slice of the image to be segmented layer by layer, and realizing interpolation and completion of discontinuous regions of the liver boundary so as to obtain a smooth and continuous liver surface.

The automatic context model-based liver segmentation system for the CT image provided by the embodiment uses context information as new features and performs iterative classification on the basis of texture feature classification to obtain a prior model of the liver, and uses the model as prior constraint to improve an energy function of a random walk algorithm to obtain a final liver segmentation result.

Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations fall within the scope defined by the appended claims.

Claims (10)

1. A CT image liver segmentation method based on an automatic context model is characterized by comprising the following steps:

s101, reading a training image set and an image to be segmented, wherein the training image and the image to be segmented in the training image set are CT images of a liver;

s102, extracting texture features of each pixel in the training image and the image to be segmented;

s103, classifying the characteristics of each pixel of the image to be segmented by using a classifier to obtain an initial liver probability map;

s104, extracting the context characteristics of each pixel in the training image and the image to be segmented;

s105, combining the context features with the texture features, learning again to obtain a new classifier, obtaining a liver probability graph again, extracting the context features of each pixel in the training image and the image to be segmented again, repeating the algorithm, learning a series of classifiers until convergence, obtaining the probability that the pixel points belong to the liver region, and further obtaining the liver probability graph;

s106, adding the liver probability map serving as prior information serving as a prior constraint condition into a random walk objective function to obtain a context constraint-based random walk model and realize liver segmentation;

s107, realizing liver segmentation of the three-dimensional CT image layer by layer on the two-dimensional slice of the image to be segmented, and realizing interpolation and completion of discontinuous regions of a liver boundary so as to obtain a smooth and continuous liver surface.

2. The automatic context model-based CT image liver segmentation method of claim 1, wherein the texture features extracted in S102 are Haar features, local binary pattern features, histogram of oriented gradients features or co-occurrence matrix features.

3. The method for liver segmentation of CT images based on automatic context model as claimed in claim 1, wherein the classifier in S103 is AdaBoost classifier, support vector machine classifier, decision tree classifier, artificial neural network classifier, naive Bayes classifier or random forest classifier using support vector machine as weak classifier.

4. The method for segmenting the liver of the CT image based on the automatic context model according to claim 3, wherein the step S103 is specifically as follows:

defining a training image set as V_iI is 1,2 … n, and the corresponding division golden standard image is V_siI is 1,2 … n, inSelecting a training sample point set from the training image set, extracting the texture features of the point set, and expressing the information of the training point set as follows:

S₀＝{(y_t,f⁰(N_t)),t＝1,2…T}，

wherein N is_tNeighborhood image block, f, centered around a pixel with index t⁰(N_t) Texture feature, y, representing the neighborhood of pixel points indexed by t_tThe method comprises the steps of obtaining a liver classifier based on texture feature classification by utilizing an AdaBoost algorithm if a category mark corresponding to a pixel point with an index of T is adopted, wherein T is the total index number, and then, for an image V to be segmented_uExtracting the texture features of the pixel point xAnd classifying to obtain initial liver posterior probability of corresponding classification mapping

$p_{ux}^0(y=1|x)=\frac{e^{H_0\left(f_{ux}^0\right)}}{e^{H_0\left(f_{ux}^0\right)}+e^{-H_0\left(f_{ux}^0\right)}},$

Wherein y is the category label corresponding to the pixel point x, y is 1, which indicates that the pixel point belongs to the liver,is the posterior probability that pixel point x belongs to the liver, H₀The classifier is obtained by learning in the texture feature space, and for the image to be segmented, the initial liver posterior probability of the classification model for classifying and mapping each pixel point can be obtained.

5. The automatic context model-based liver segmentation method for CT images according to claim 4, wherein S104 is based on the current classification result, for a pixel point, with the pixel point as the center, a plurality of rays with equal angular intervals are extracted outwards, sparse sampling is performed on the rays, classification probability of the corresponding position is obtained as the context feature, on the classification result graph corresponding to the CT image slice where the pixel point with index t is located, starting from the pixel point, a first number of rays are extracted outwards at intervals of 45 degrees, the context position is sampled at equal intervals on each ray, and the classification probability at the position is used as the context feature P of the pixel point with index t⁰(t):

$P^0\left(t\right)=(p_{t_1}^0,p_{t_2}^0,...,p_{t_m}^0),$

Wherein, t_mAn index indicating a pixel point corresponding to the mth context position around the pixel point having the index t,is indexed by t_mThe pixel points are classified based on the liver posterior probability value of the texture feature, and the contextual features of the pixel points in the image to be segmented can be obtained by the same method.

6. The method of claim 5, wherein the first number is 8.

7. The method for liver segmentation in CT images based on automatic context model according to claim 5, wherein the step S105 is specifically as follows: and integrating texture features and context features of the image to construct a new training point set, wherein the information can be expressed as:

S₁＝{(y_t,(f⁰(N_t),P⁰(t))),t＝1...T}，

wherein f is⁰(N_t) And P⁰(t) respectively representing texture features of pixel points with index of t and context features extracted based on classification mapping, learning to obtain a new classifier by using AdaBoost algorithm again based on new features of training image combination, repeating the algorithm, integrating the texture features of the image and the context information, learning a series of classifiers until convergence,

for an image V to be segmented_uThe probability that the pixel point x in the image belongs to the liver region is obtained through q times of classifier iterative learning

$p_{ux}^q(y=1|x)=\frac{e^{H_q\left(f_{ux}^q\right)}}{e^{H_q\left(f_{ux}^q\right)}+e^{-H_q\left(f_{ux}^q\right)}},$

Wherein,the posterior probability that a pixel point x obtained by iterative learning of the classifier belongs to the liver region, q is the iterative times of the classifier in convergence, H_qIs the q-th classifier obtained by learning in the combined space based on the texture features and the context features,representing an image V to be segmented_uThe texture feature of the pixel point x in (1) and the context feature obtained based on the classification mapping of the (q-1) th time.

8. The method for liver segmentation in CT images based on automatic context model according to claim 7, wherein the step S106 is specifically as follows: for an image V to be segmented_uIt is converted into an undirected graph G ═ (V, E), and a set of nodes V ═ V₁,v₂...v_N}∪{l₀,l₁},v_iRepresenting graph nodes corresponding to the pixel points with index i, N being the total number of nodes, l₀And l₁End nodes representing non-liver regions and liver regions, respectively; edge set E is composed of_T-linkAnd E_N-linkIs composed of (A) a_T-linkFor two end nodes l₀、l₁And pixel node v_iThe weight of the edge of the middle edge set is respectively as follows:

$\left\{\begin{array}{c}{\mathrm{\&omega;}}_i^1=p_{ui}^q\\ {\mathrm{\&omega;}}_i^0=1-p_{ui}^q\end{array}\right.,$

wherein,is the end node l₁And pixel point v_iThe edge weight of (1) is the probability value of the pixel point with index i belonging to the liver region obtained by the context model q times of iterationIn the same way as above, the first and second,is the end node l₀And pixel point v_iThe edge weight value of (a) is the probability value that the pixel with index i does not belong to the liver region, so as to represent the prior information of the pixel in the image, E_N-linkRepresenting the connection relationship between adjacent pixels and the weight thereofThe method comprises the steps that gray values of pixel points in an image to be segmented are determined, and meanwhile, the pixel points which belong to a liver region and have a probability value of 1 are marked as seed points of the liver region; the pixel point with the probability value of 0 is marked as a seed point of a non-liver region,

based on the graph model, a new objective function with prior constraint is established by using the marked seed points to minimize:

$E_{aspatial}^s=\underset{e_{ij}\&Element;E}{\&Sigma;}{\mathrm{\&omega;}}_{e_{ij}}{(x_i^s-x_j^s)}^2+\mathrm{\&gamma;}\left(\underset{v_i\&Element;V}{\&Sigma;}{\mathrm{\&omega;}}_i^{1-s}{\left(x_i^s\right)}^2+\underset{v_i\&Element;V}{\&Sigma;}{\mathrm{\&omega;}}_i^s{\left(x_i^s-1\right)}^2\right),$

wherein,for an objective function with a priori constraints, e_ijThe first term of the above formula represents the edges of the pixel point with the connection index i and the pixel point with the connection index iRepresenting the original random walk objective function, the second term is a priori constraint term based on the context model, gamma is an adjusting parameter,the probability that a pixel point with index i in the image belongs to the category s, where s is {0,1}, and represents the liver category and the non-liver category, respectively, and the above formula is expressed by a matrix, so that the following can be obtained:

$E_{aspaital}^s=x^{sT}{\mathrm{Lx}}^s+\mathrm{\&gamma;}\&lsqb;x^{sT}{\mathrm{\&Lambda;}}^{1-s}{x}^s+{(x^s-1)}^T{\mathrm{\&Lambda;}}^s(x^s-1)\&rsqb;,$

wherein x is^sFor the probability that each pixel in the image node set belongs to different classes, the matrix L is the Laplacian matrix of the image to be segmented, Λ^sIs the value of the ith row on the diagonalThe diagonal matrix of (1) for solving the above formula, dividing all vertexes in a node set V of the undirected graph into a seed node set V_M(marked Point set) and unmarked Point set V_UTwo subsets, decompose the above equation and relate to x_UThe differential of (2) can be obtained:

$(L_U+{\mathrm{\&gamma;\&Lambda;}}_U^{1-s})x_U^s={\mathrm{\&gamma;\&Lambda;}}_U^S-{\mathrm{Bx}}_M^s,$

wherein L is_UA laplacian matrix of unlabeled point sets,the probability value belonging to the category s for an unmarked point,the value of the ith row on the diagonal for the unmarked point set isThe diagonal matrix of (a) is,the value of the ith row on the diagonal for the unmarked point set isA diagonal matrix ofIs obtained by mathematical derivation, B is a matrix,the probability value of the marking pixel point reaching the category s seed point for the first time is shown, namely if the pixel point with the index i is the category s seed point, the probability value is shownIf not, then,

based on the above-mentioned | V-containing_USolving the probability value from the unmarked point to the two kinds of seed points by the symmetrical positive definite linear equation set of | unknowns to obtain the maximum transition probabilityJudging the category label (i) of the pixel point with index i as a criterion, namely:

$label\left(i\right)=\left\{\begin{array}{cc}1,& x_i^1\&GreaterEqual;x_i^0\\ 0,& x_i^1<x_i^0\end{array}\right.,$

the expression shows that when the probability that the non-mark point reaches the liver seed point is obtained to be larger than or equal to the probability that the non-mark point reaches the liver seed point, the non-mark point belongs to the liver region, and therefore the final segmentation of the liver region in the image is achieved.

9. The method for liver segmentation in CT images based on automatic context model according to claim 1, wherein the step S107 is specifically as follows: the algorithm provided by the invention is to realize liver segmentation of a three-dimensional CT image layer by layer on a two-dimensional slice of the image, and the phenomenon of discontinuous boundary or even overlapped boundary can occur on the obtained liver boundary, and the invention realizes interpolation and completion of discontinuous regions of the liver boundary so as to obtain a smooth and continuous liver surface, and the specific calculation process is as follows:

inputting a liver segmentation result, carrying out octree decomposition on the segmentation result by utilizing a Scan-conversion algorithm, and decomposing the segmentation result into a finer subspace;

in the octree decomposition process, when the intersection points of all the decomposition lines and the original model are positioned at the leaves of the octree, stopping decomposition;

marking boundaries with intersections as "intersecting edges";

randomly selecting a vertex P from an original model, and marking the vertex P as 0; expanding the label along the boundary of the octree, changing the label into '1' when the 'intersecting edge' passes once, and so on, changing the label once when the 'intersecting edge' passes once until the traversal of the whole octree is finished;

and precisely reconstructing the vertexes only containing '0' and '1' by using a Dual containment algorithm to obtain the model after the cavity is filled.

10. An automatic context model-based liver segmentation system for CT images, comprising:

the reading module is used for reading a training image set and an image to be segmented, wherein the training image set and the image to be segmented are CT images of the liver;

the first extraction module is used for extracting the texture features of each pixel in the training image and the image to be segmented;

the classification module is used for classifying the characteristics of each pixel of the image to be segmented by using the classifier to obtain an initial liver probability map;

the second extraction module is used for extracting the context characteristics of each pixel in the training image and the image to be segmented;

the iteration module is used for combining the context features with the texture features, learning again to obtain a new classifier, obtaining a liver probability map again, extracting the context features of each pixel in the training image and the image to be segmented again, repeating the algorithm, learning a series of classifiers until convergence, obtaining the probability that the pixel points belong to the liver region, and further obtaining the liver probability map;

the segmentation module is used for taking the liver probability map as prior information and taking the liver probability map as a prior constraint condition, adding the prior information and the prior constraint condition into a random walk target function to obtain a context constraint-based random walk model and realize the segmentation of the liver;

and the filling module is used for realizing liver segmentation of the three-dimensional CT image on the two-dimensional slice of the image to be segmented layer by layer, and realizing interpolation and completion of discontinuous regions of the liver boundary so as to obtain a smooth and continuous liver surface.