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, which can effectively improve the liver segmentation accuracy in the CT image. The method includes: reading the training image set and the image to be segmented; extracting the texture feature of each pixel in the image; using a classifier to classify the feature of each pixel in the image to be segmented to obtain an initial liver probability map; extracting the The context feature of each pixel in the image; combine the context feature with the texture feature, and learn a series of classifiers through iterative learning until convergence, and obtain the liver probability map; take the liver probability map as the prior information, as the prior constraint, and add random In the objective function of walking, a random walk model based on context constraints is obtained to realize the segmentation of the liver; the liver segmentation of the three-dimensional CT image is realized layer by layer on the two-dimensional slice of the image to be segmented, and the segmentation of the discontinuous area of the liver boundary is realized. Interpolation and completion to obtain a smooth and continuous liver surface.

Description

Liver segmentation method and system in CT image 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 technique

Medical image segmentation assists doctors in identifying patients' internal tissues, organs and lesion areas, and plays a vital role in computer-aided treatment and surgical planning. Therefore, the automatic segmentation of the liver is the basis for doctors to diagnose and treat liver diseases such as cirrhosis, liver tumors, and liver transplantation. In abdominal CT images, the gray value difference between the liver and adjacent organs is small, and the liver itself has uneven gray levels and different shapes, so it is difficult to automatically and accurately segment the liver. Therefore, clinicians urgently need a simple, fast and accurate liver segmentation method.

The existing random walk segmentation method has the advantages of being fast and simple, but it is not effective in segmenting areas with low contrast in CT images, especially the junctions between the liver and adjacent organs such as large blood vessels and stomach, which rely solely on the gray value Segmentation of the liver is difficult to achieve efficiently.

Contents of the invention

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

On the one hand, the embodiment of the present invention proposes a CT image liver segmentation method based on an automatic context model, including:

S101. Read the training image set and the image to be segmented, wherein the training image in the training image set and the image to be segmented are CT images of the liver;

S102. Extracting texture features of each pixel in the training image and the image to be segmented;

S103. Using a classifier to classify the features of each pixel of the image to be segmented to obtain an initial liver probability map;

S104. Extract the context features of each pixel in the training image and the image to be segmented;

S105. Combining the context feature with the texture feature, learn again to obtain a new classifier, obtain the liver probability map again, extract the context feature of each pixel in the training image and the image to be segmented again, repeat the above algorithm, and learn a series of The classifier converges until the probability of the pixel belonging to the liver area is obtained, and then the liver probability map is obtained;

S106. Using the liver probability map as prior information and as prior constraints, add it to the objective function of random walk, obtain a random walk model based on context constraints, and realize liver segmentation;

S107. Realize the liver segmentation of the three-dimensional CT image layer by layer on the two-dimensional slice of the image to be segmented, and realize the interpolation and completion of the discontinuous region of the liver boundary, thereby obtaining a smooth and continuous liver surface.

On the other hand, an embodiment of the present invention proposes a CT image liver segmentation system based on an automatic context model, including:

A reading module, configured to read the training image set and the image to be segmented, wherein the training image in the training image set and the image to be segmented are CT images of the liver;

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

The classification module is used to use a classifier to classify the features of each pixel of the image to be segmented to obtain an initial liver probability map;

The second extraction module is used to extract the context features of each pixel in the training image and the image to be segmented;

The iterative module is used to combine the context feature with the texture feature, learn again to obtain a new classifier, obtain the liver probability map again, extract the context feature of each pixel in the training image and the image to be segmented again, repeat the above algorithm, and learn A series of classifiers until convergence, to obtain the probability that the pixel belongs to the liver area, and then obtain the liver probability map;

The segmentation module is used to use the liver probability map as prior information and as prior constraints, add it to the objective function of random walk, obtain a random walk model based on context constraints, and realize liver segmentation;

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

The method and system for liver segmentation in CT images based on the automatic context model provided by the embodiments of the present invention, on the basis of texture feature classification, uses context information as a new feature and performs iterative classification to obtain a priori model of the liver, and use this model as a priori According to the experimental constraints, the energy function of the random walk algorithm is improved, and the final liver segmentation result is obtained. The invention greatly improves the segmentation results for areas with inconspicuous gray contrast, and effectively improves the segmentation accuracy of the liver in CT images.

Description of drawings

Fig. 1 is the schematic flow chart of an embodiment of the CT image liver segmentation method based on the automatic context model of the present invention;

FIG. 2 is a schematic structural diagram of an embodiment of the liver segmentation system for CT images based on the automatic context model of the present invention.

detailed description

In order to make the purpose, 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 in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are the Some, but not all, embodiments are invented. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

As shown in FIG. 1, the present embodiment discloses a liver segmentation method for CT images based on an automatic context model, including:

S101. Read the training image set and the image to be segmented, wherein the training image in the training image set and the image to be segmented are CT images of the liver;

S102. Extracting texture features of each pixel in the training image and the image to be segmented;

S103. Using a classifier to classify the features of each pixel of the image to be segmented to obtain an initial liver probability map;

S104. Extract the context features of each pixel in the training image and the image to be segmented;

S105. Combining the context feature with the texture feature, learn again to obtain a new classifier, obtain the liver probability map again, extract the context feature of each pixel in the training image and the image to be segmented again, repeat the above algorithm, and learn a series of The classifier converges until the probability of the pixel belonging to the liver area is obtained, and then the liver probability map is obtained;

S106. Using the liver probability map as prior information and as prior constraints, add it to the objective function of random walk, obtain a random walk model based on context constraints, and realize liver segmentation;

S107. Realize the liver segmentation of the three-dimensional CT image layer by layer on the two-dimensional slice of the image to be segmented, and realize the interpolation and completion of the discontinuous region of the liver boundary, thereby obtaining a smooth and continuous liver surface.

The CT image liver segmentation method based on the automatic context model provided in this embodiment, on the basis of texture feature classification, uses context information as a new feature and iteratively classifies to obtain a priori model of the liver, using this model as a priori constraint, The energy function of the random walk algorithm is improved to obtain the final liver segmentation result. The invention has a large improvement in the segmentation result for areas with inconspicuous gray contrast, and effectively improves the segmentation accuracy of the liver in CT images.

Optionally, in another embodiment of the liver segmentation method for CT images based on the automatic context model of the present invention, the texture features extracted in S102 may be Haar features, local binary pattern features, oriented gradient histogram features or co-occurrence matrix features , 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 of the present invention, the classifier in S103 is an AdaBoost classifier, a support vector machine classifier, a decision Tree classifiers, artificial neural network classifiers, naive Bayes classifiers, or random forest classifiers.

Optionally, in another embodiment of the automatic context model-based CT image liver segmentation method of the present invention, the S103 is specifically:

Define the training image set as V _i , i=1,2...n, and its corresponding segmentation gold standard image is V _si , i=1,2...n, select the training sample point set in the training image set, and extract the texture of the point set feature, the training point set information can be expressed as:

S ₀ ={(y _t ,f ⁰ (N _t )),t=1,2...T},

Among them, N _t is the neighborhood image block centered on the pixel with index t, f ⁰ (N _t ) represents the texture feature of the pixel neighborhood with index t, and y _t is the corresponding Class mark, T is the total number of indexes, use the AdaBoost algorithm to obtain the liver classifier based on texture feature classification, then for the pixel point x in the image V _u to be segmented, extract its texture feature And classify, obtain the initial liver posterior probability corresponding to the classification map

${\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; e</span>}}^{{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span>}}{{\mathrm{<span\; class="notranslate">\; e</span>}}^{{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; e</span>}}^{<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span>}}<span\; class="notranslate">\; ,</span>$

Among them, y is the category label corresponding to the pixel point x, and y=1 means that the pixel point belongs to the liver, is the posterior probability that the pixel _x belongs to the liver, and H0 is the classifier learned in the texture feature space. For the image to be segmented, the initial liver posterior probability of the classification model for each pixel can also be obtained.

Optionally, in another embodiment of the liver segmentation method for CT image based on the automatic context model of the present invention, said S104 is based on the current classification result, for the pixel point, take it as the center, and draw out several equiangular intervals The rays are sparsely sampled on these rays, and the classification probability of the corresponding position is obtained as the context feature. On the classification result map corresponding to the CT image slice where the pixel point with index t is located, starting from the pixel point, the interval is 45° Draw the first number of rays outward, sample the context position at equal intervals on each ray and use the classification probability at this position as the pixel context feature P ⁰ (t) with index t:

${\mathrm{<span\; class="notranslate">\; P</span>}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; ...</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$

Among them, t _m represents the index of the pixel corresponding to the mth context position around the pixel with index t, is the liver posterior probability value of the pixel with index t _m classified based on the texture feature. For the pixel in the image to be segmented, the same method can be used to obtain its context feature.

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 liver segmentation method for CT images based on the automatic context model of the present invention, the S105 is specifically: integrating the texture features and context features of the image, and constructing a new training point set, the information of which can represent for:

S ₁ ={(y _t ,(f ⁰ (N _t ),P ⁰ (t))),t=1...T},

Among them, f ⁰ (N _t ) and P ⁰ (t) respectively represent the texture feature of the pixel point with index t and the context feature extracted based on the classification map. Based on the new features of the training image combination, the AdaBoost algorithm is used to learn again to obtain a new classifier, and repeat the above algorithm, integrate the texture features of the image with context information, learn a series of classifiers until convergence,

For the pixel point x in the image V _u to be segmented, the probability that the pixel point belongs to the liver area after q classifier iterative learning is

${\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; e</span>}}^{{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; )</span>}}{{\mathrm{<span\; class="notranslate">\; e</span>}}^{{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; e</span>}}^{<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; )</span>}}<span\; class="notranslate">\; ,</span>$

in, is the posterior probability that the pixel point x obtained by iterative learning of the classifier belongs to the liver area, q is the number of iterations of the classifier when it converges, and H _q is the qth classifier learned in the combined space based on texture features and context features, Indicates the combination of the pixel point x texture feature in the image V _u to be segmented and the context feature obtained based on the q-1th classification mapping, and the liver probability map of the image to be segmented can be obtained.

Optionally, in another embodiment of the liver segmentation method for CT image based on the automatic context model of the present invention, the S106 is specifically: for the image V _u to be segmented, convert it into an undirected graph G=(V,E ), the node set V={v ₁ ,v ₂ ...v _N }∪{l ₀ ,l ₁ }, v _i represents the graph node corresponding to the pixel with index i, N is the total number of nodes, l ₀ and l ₁ represent the end nodes of the non-liver area and the liver area respectively; the edge set E is composed of E _T-link and E _N-link , where E _T-link is the two end nodes l ₀ , l ₁ and the pixel node v For the edge set between _i , the weights of the edges are:

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; q</span>}}\\ {\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; q</span>}}\end{array}\right.<span\; class="notranslate">\; ,</span>$

in, is the edge weight between end node l ₁ and pixel v _i , and its value is the probability value that the pixel with index i belongs to the liver area obtained after q iterations of the context model same, is the edge weight between end node l ₀ and pixel v _i , and its value is the probability value that the pixel with index i does not belong to the liver area, which represents the prior information of the pixel in the image, and E _N-link represents the relative The connection relationship between adjacent pixels, its weight It is determined by the gray value of the pixel in the image to be segmented. At the same time, for the pixel with a probability value of 1 belonging to the liver area, it is marked as the seed point of the liver area; the pixel with a probability value of 0 is marked as a non-liver area the seed point,

Based on the above graphical model, using the marked seed points, a new objective function with prior constraints is established to minimize it:

${\mathrm{<span\; class="notranslate">\; E.</span>}}_{\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; p</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; l</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; =</span>\underset{{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; E.</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&gamma;</span>}<span\; class="notranslate">\; (</span>\underset{{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; V</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\underset{{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; V</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$

in, is the objective function with prior constraints, e _ij represents the edge connecting the pixel point with index i and the pixel point with index i, the first term of the above formula Represents the original random walk objective function, the second item is the prior constraint item based on the context model, γ is the adjustment parameter, is the probability that the pixel with index i in the image belongs to category s, s={0,1}, which respectively represent the liver category and non-liver category, and the above formula can be expressed as a matrix, which can be obtained:

${\mathrm{<span\; class="notranslate">\; E.</span>}}_{\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; p</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; l</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; T</span>}}{\mathrm{<span\; class="notranslate">\; Lx</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&gamma;</span>}<span\; class="notranslate">\; \&lsqb;</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; T</span>}}{\mathrm{<span\; class="notranslate">\; \&Lambda;</span>}}^{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}}{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; T</span>}}{\mathrm{<span\; class="notranslate">\; \&Lambda;</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; ,</span>$

Among them, x ^s is the probability that each pixel in the image node set belongs to different categories, matrix L is the Laplacian matrix of the image to be segmented, and Λ ^s is the value of the i-th row on the diagonal Diagonal matrix of , in order to solve the above formula, all vertices in the node set V of the undirected graph are divided into two subsets, the seed node set V _M (marked point set) and the unmarked point set V _U , and the above formula is decomposed And take the differential with respect to x _U , we can get:

$<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&gamma;\&Lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}}^{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&gamma;\&Lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}}^{\mathrm{<span\; class="notranslate">\; S</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Bx</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; ,</span>$

Among them, L _U is the Laplacian matrix of the unlabeled point set, is the probability value of unmarked points belonging to category s, The value of the i-th row on the diagonal of the unmarked point set is the diagonal matrix, The value of the i-th row on the diagonal of the unmarked point set is The diagonal matrix of Mathematical derivation, B is a matrix, Indicates the probability value of the marked pixel point reaching the seed point of category s for the first time, that is, if the pixel point with index i itself is the seed point of category s, then otherwise,

Based on the above-mentioned symmetric positive definite linear equations containing |V _U | unknowns, the probability values from non-marked points to two types of seed points are solved, and the maximum transition probability As a criterion to judge the category label(i) of the pixel with index i, namely:

$\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; b</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; \&Greater\; Equal;</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}\\ \mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}\end{array}\right.<span\; class="notranslate">\; ,</span>$

The above formula indicates that when the probability of reaching the liver seed point of the non-marking point is greater than or equal to the probability of reaching the non-liver seed point, the non-marking point belongs to the liver region, thereby realizing the final segmentation of the liver region in the image.

Optionally, in another embodiment of the automatic context model-based CT image liver segmentation method of the present invention, the S107 is specifically: the proposed algorithm implements the liver segmentation of the three-dimensional CT image layer by layer on the two-dimensional slice of the image , the boundary discontinuity or even boundary overlap will appear on the obtained liver boundary. The present invention realizes the interpolation and completion of the discontinuous area of the liver boundary, so as to obtain a smooth and continuous liver surface. The specific calculation process is as follows:

Input the results of liver segmentation, use the Scan-conversion algorithm to decompose the segmentation results into octrees, and decompose the segmentation results into finer subspaces;

During the octree decomposition process, when the intersection points of all decomposition lines and the original model are at the leaves of the octree, stop the decomposition;

mark boundaries with intersections as "intersecting edges";

Randomly select a vertex P from the original model, and mark it as "0"; extend it along the boundary of the octree, when it passes through the "intersecting edge", the label will change to "1", and so on, every time After one "intersecting edge", the label changes once until the entire octree traversal ends;

The vertices containing only "0" and "1" are accurately reconstructed using the Dual Contouring algorithm to obtain a model after hole filling.

As shown in Fig. 2, the present embodiment discloses a CT image liver segmentation system based on an automatic context model, including:

A reading module 1, configured to read a training image set and an image to be segmented, wherein the training image in the training image set and the image to be segmented are CT images of the liver;

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

The classification module 3 is used to use a classifier to classify the features of each pixel of the image to be segmented to obtain an initial liver probability map;

The second extraction module 4 is used to extract the context feature of each pixel in the training image and the image to be segmented;

The iterative module 5 is used to combine the context feature with the texture feature, learn again to obtain a new classifier, obtain the liver probability map again, extract the context feature of each pixel in the training image and the image to be segmented again, and repeat the above algorithm, Learn a series of classifiers until convergence, get the probability that the pixel belongs to the liver area, and then obtain the liver probability map;

The segmentation module 6 is used to use the liver probability map as prior information and as prior constraints, add it to the objective function of random walk, obtain a random walk model based on context constraints, and realize liver segmentation;

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

The CT image liver segmentation system based on the automatic context model provided in this embodiment, on the basis of texture feature classification, uses context information as a new feature and iteratively classifies to obtain a priori model of the liver, using this model as a priori constraint, The energy function of the random walk algorithm is improved to obtain the final liver segmentation result. The invention has a large improvement in the segmentation result for areas with inconspicuous gray contrast, and effectively improves the segmentation accuracy of the liver in CT images.

Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present invention. within the bounds of the requirements.

Claims (10)

1. A CT image liver segmentation method based on an automatic context model, characterized in that, comprising: S101. Read the training image set and the image to be segmented, wherein the training image in the training image set and the image to be segmented are CT images of the liver; S102. Extracting texture features of each pixel in the training image and the image to be segmented; S103. Using a classifier to classify the features of each pixel of the image to be segmented to obtain an initial liver probability map; S104. Extract the context features of each pixel in the training image and the image to be segmented; S105. Combining the context feature with the texture feature, learn again to obtain a new classifier, obtain the liver probability map again, extract the context feature of each pixel in the training image and the image to be segmented again, repeat the above algorithm, and learn a series of The classifier converges until the probability of the pixel belonging to the liver area is obtained, and then the liver probability map is obtained; S106. Using the liver probability map as prior information and as prior constraints, add it to the objective function of random walk, obtain a random walk model based on context constraints, and realize liver segmentation; S107. Realize the liver segmentation of the three-dimensional CT image layer by layer on the two-dimensional slice of the image to be segmented, and realize the interpolation and completion of the discontinuous region of the liver boundary, thereby obtaining a smooth and continuous liver surface. 2. The CT image liver segmentation method based on an automatic context model according to claim 1, wherein the texture features extracted by the S102 are Haar features, local binary pattern features, oriented gradient histogram features or co-occurrence matrix features. 3. the CT image liver segmentation method based on automatic context model according to claim 1, is characterized in that, classifier is support vector machine as the AdaBoost classifier of weak classifier, support vector machine classifier, decision tree among the described S103 Classifier, Artificial Neural Network Classifier, Naive Bayes Classifier, or Random Forest Classifier. 4. The CT image liver segmentation method based on the automatic context model according to claim 3, wherein said S103 is specifically: Define the training image set as V _i , i=1,2...n, and its corresponding segmentation gold standard image is V _si , i=1,2...n, select the training sample point set in the training image set, and extract the texture of the point set feature, the training point set information can be expressed as: S ₀ ={(y _t ,f ⁰ (N _t )),t=1,2...T}, Among them, N _t is the neighborhood image block centered on the pixel with index t, f ⁰ (N _t ) represents the texture feature of the pixel neighborhood with index t, and y _t is the corresponding Class mark, T is the total number of indexes, use the AdaBoost algorithm to obtain the liver classifier based on texture feature classification, then for the pixel point x in the image V _u to be segmented, extract its texture feature And classify, obtain the initial liver posterior probability corresponding to the classification map ${\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; e</span>}}^{{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span>}}{{\mathrm{<span\; class="notranslate">\; e</span>}}^{{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; e</span>}}^{<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span>}}<span\; class="notranslate">\; ,</span>$ Among them, y is the category label corresponding to the pixel point x, and y=1 means that the pixel point belongs to the liver, is the posterior probability that the pixel _x belongs to the liver, and H0 is the classifier learned in the texture feature space. For the image to be segmented, the initial liver posterior probability of the classification model for each pixel can also be obtained. 5. The CT image liver segmentation method based on the automatic context model according to claim 4, characterized in that, based on the current classification result, the S104 draws out several equiangular intervals for the pixel point with it as the center The rays are sparsely sampled on these rays, and the classification probability of the corresponding position is obtained as the context feature. On the classification result map corresponding to the CT image slice where the pixel point with index t is located, starting from the pixel point, the interval is 45° Draw the first number of rays outward, sample the context position at equal intervals on each ray and use the classification probability at this position as the pixel context feature P ⁰ (t) with index t: ${\mathrm{<span\; class="notranslate">\; P</span>}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; ...</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$ Among them, t _m represents the index of the pixel corresponding to the mth context position around the pixel with index t, is the liver posterior probability value of the pixel with index t _m classified based on the texture feature. For the pixel in the image to be segmented, the same method can be used to obtain its context feature. 6. The CT image liver segmentation method based on an automatic context model according to claim 5, wherein the first number is 8. 7. The CT image liver segmentation method based on the automatic context model according to claim 5, characterized in that, said S105 is specifically: synthesizing the texture features and context features of the image to construct a new training point set, the information of which can represent for: S ₁ ={(y _t ,(f ⁰ (N _t ),P ⁰ (t))),t=1...T}, Among them, f ⁰ (N _t ) and P ⁰ (t) respectively represent the texture feature of the pixel point with index t and the context feature extracted based on the classification map. Based on the new features of the training image combination, the AdaBoost algorithm is used to learn again to obtain a new classifier, and repeat the above algorithm, integrate the texture features of the image with context information, learn a series of classifiers until convergence, For the pixel point x in the image V _u to be segmented, the probability that the pixel point belongs to the liver area after q classifier iterative learning is ${\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; e</span>}}^{{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; )</span>}}{{\mathrm{<span\; class="notranslate">\; e</span>}}^{{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; e</span>}}^{<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; q</span>}}<span\; class="notranslate">\; )</span>}}<span\; class="notranslate">\; ,</span>$ in, is the posterior probability that the pixel point x obtained by iterative learning of the classifier belongs to the liver area, q is the number of iterations of the classifier when it converges, and H _q is the qth classifier learned in the combined space based on texture features and context features, Indicates the combination of the pixel point x texture feature in the image V _u to be segmented and the context feature obtained based on the q-1th classification mapping. 8. The CT image liver segmentation method based on the automatic context model according to claim 7, characterized in that, the S106 is specifically: for the image V _u to be segmented, convert it into an undirected graph G=(V, E ), the node set V={v ₁ ,v ₂ ...v _N }∪{l ₀ ,l ₁ }, v _i represents the graph node corresponding to the pixel with index i, N is the total number of nodes, l ₀ and l ₁ represent the end nodes of the non-liver area and the liver area respectively; the edge set E is composed of E _T-link and E _N-link , where E _T-link is the two end nodes l ₀ , l ₁ and the pixel node v For the edge set between _i , the weights of the edges are: $\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; q</span>}}\\ {\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; q</span>}}\end{array}\right.<span\; class="notranslate">\; ,</span>$ in, is the edge weight between end node l ₁ and pixel v _i , and its value is the probability value that the pixel with index i belongs to the liver area obtained after q iterations of the context model same, is the edge weight between end node l ₀ and pixel v _i , and its value is the probability value that the pixel with index i does not belong to the liver area, which represents the prior information of the pixel in the image, and E _N-link represents the relative The connection relationship between adjacent pixels, its weight It is determined by the gray value of the pixel in the image to be segmented. At the same time, for the pixel with a probability value of 1 belonging to the liver area, it is marked as the seed point of the liver area; the pixel with a probability value of 0 is marked as a non-liver area the seed point, Based on the above graphical model, using the marked seed points, a new objective function with prior constraints is established to minimize it: ${\mathrm{<span\; class="notranslate">\; E.</span>}}_{\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; p</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; l</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; =</span>\underset{{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; E.</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&gamma;</span>}<span\; class="notranslate">\; (</span>\underset{{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; V</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\underset{{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; V</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$ in, is the objective function with prior constraints, e _ij represents the edge connecting the pixel point with index i and the pixel point with index i, the first term of the above formula Represents the original random walk objective function, the second item is the prior constraint item based on the context model, γ is the adjustment parameter, is the probability that the pixel with index i in the image belongs to category s, s={0,1}, which respectively represent the liver category and non-liver category, and the above formula can be expressed as a matrix, which can be obtained: ${\mathrm{<span\; class="notranslate">\; E.</span>}}_{\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; p</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; l</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; T</span>}}{\mathrm{<span\; class="notranslate">\; Lx</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&gamma;</span>}<span\; class="notranslate">\; \&lsqb;</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; T</span>}}{\mathrm{<span\; class="notranslate">\; \&Lambda;</span>}}^{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}}{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; T</span>}}{\mathrm{<span\; class="notranslate">\; \&Lambda;</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; ,</span>$ Among them, x ^s is the probability that each pixel in the image node set belongs to different categories, matrix L is the Laplacian matrix of the image to be segmented, and Λ ^s is the value of the i-th row on the diagonal Diagonal matrix of , in order to solve the above formula, all vertices in the node set V of the undirected graph are divided into two subsets, the seed node set V _M (marked point set) and the unmarked point set V _U , and the above formula is decomposed And take the differential with respect to x _U , we can get: $<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&gamma;\&Lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}}^{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&gamma;\&Lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}}^{\mathrm{<span\; class="notranslate">\; S</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Bx</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; ,</span>$ Among them, L _U is the Laplacian matrix of the unlabeled point set, is the probability value of unmarked points belonging to category s, The value of the i-th row on the diagonal of the unmarked point set is the diagonal matrix, The value of the i-th row on the diagonal of the unmarked point set is The diagonal matrix of Mathematical derivation, B is a matrix, Indicates the probability value of the marked pixel point reaching the seed point of category s for the first time, that is, if the pixel point with index i itself is the seed point of category s, then otherwise, Based on the above-mentioned symmetric positive definite linear equations containing |V _U | unknowns, the probability values from non-marked points to two types of seed points are solved, and the maximum transition probability As a criterion to judge the category label(i) of the pixel with index i, namely: $\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; b</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; \&Greater\; Equal;</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}\\ \mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}\end{array}\right.<span\; class="notranslate">\; ,</span>$ The above formula indicates that when the probability of reaching the liver seed point of the non-marking point is greater than or equal to the probability of reaching the non-liver seed point, the non-marking point belongs to the liver region, thereby realizing the final segmentation of the liver region in the image. 9. The CT image liver segmentation method based on the automatic context model according to claim 1, characterized in that, said S107 is specifically: the algorithm proposed is to realize the liver segmentation of the three-dimensional CT image layer by layer on the two-dimensional slice of the image , the boundary discontinuity or even boundary overlap will appear on the obtained liver boundary. The present invention realizes the interpolation and completion of the discontinuous area of the liver boundary, so as to obtain a smooth and continuous liver surface. The specific calculation process is as follows: Input the results of liver segmentation, use the Scan-conversion algorithm to decompose the segmentation results into octrees, and decompose the segmentation results into finer subspaces; During the octree decomposition process, when the intersection points of all decomposition lines and the original model are at the leaves of the octree, stop the decomposition; mark boundaries with intersections as "intersecting edges"; Randomly select a vertex P from the original model, and mark it as "0"; extend it along the boundary of the octree, when it passes through the "intersecting edge", the label will change to "1", and so on, every time After one "intersecting edge", the label changes once until the entire octree traversal ends; The vertices containing only "0" and "1" are accurately reconstructed using the Dual Contouring algorithm to obtain a model after hole filling. 10. A CT image liver segmentation system based on an automatic context model, comprising: A reading module, configured to read the training image set and the image to be segmented, wherein the training image in the training image set and the image to be segmented are CT images of the liver; The first extraction module is used to extract the texture features of each pixel in the training image and the image to be segmented; The classification module is used to use a classifier to classify the features of each pixel of the image to be segmented to obtain an initial liver probability map; The second extraction module is used to extract the context features of each pixel in the training image and the image to be segmented; The iterative module is used to combine the context feature with the texture feature, learn again to obtain a new classifier, obtain the liver probability map again, extract the context feature of each pixel in the training image and the image to be segmented again, repeat the above algorithm, and learn A series of classifiers until convergence, to obtain the probability that the pixel belongs to the liver area, and then obtain the liver probability map; The segmentation module is used to use the liver probability map as prior information and as prior constraints, add it to the objective function of random walk, obtain a random walk model based on context constraints, and realize liver segmentation; The filling module is used to realize the liver segmentation of the three-dimensional CT image layer by layer on the two-dimensional slice of the image to be segmented, and realize the interpolation and completion of the discontinuous area of the liver boundary, so as to obtain a smooth and continuous liver surface.