CN109753997B - Automatic accurate robust segmentation method for liver tumor in CT image - Google Patents
Automatic accurate robust segmentation method for liver tumor in CT image Download PDFInfo
- Publication number
- CN109753997B CN109753997B CN201811553883.3A CN201811553883A CN109753997B CN 109753997 B CN109753997 B CN 109753997B CN 201811553883 A CN201811553883 A CN 201811553883A CN 109753997 B CN109753997 B CN 109753997B
- Authority
- CN
- China
- Prior art keywords
- liver
- image
- pixel
- segmentation
- gray
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Landscapes
- Apparatus For Radiation Diagnosis (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an automatic accurate robust segmentation method for liver tumor in CT image, comprising: (1) Preprocessing a CT image and extracting a liver region in the CT image; (2) Performing multi-level iterative segmentation on a liver region by using an image superpixel segmentation method based on LI-SLIC (Living Subtraction-Left theory), dividing the region with consistent gray level and texture in the liver into the same superpixel, and acquiring a boundary between normal liver parenchyma and liver tumor; (3) According to the local gray level and the texture features of the image, performing normal parenchyma/liver tumor secondary classification on each pixel point of the liver region; (4) And (3) classifying the super pixels generated in the step (2) according to the classification result of the pixel points in the liver region to obtain a final liver tumor segmentation result. The invention can effectively solve the segmentation difficulty caused by CT imaging noise, fuzzy liver tumor boundary, complex structure, various gray levels and the like in the CT image, and improve the efficiency and the precision of the computer-aided diagnosis of the liver diseases.
Description
Technical Field
The invention relates to the technical field of image processing and pattern recognition, in particular to an automatic accurate robust segmentation method for liver tumors in a CT image.
Background
Computed Tomography (CT) has the characteristics of small human body trauma, high image resolution, capability of visually and accurately reflecting the liver and lesion areas of a patient and the like, and is widely applied to clinical diagnosis of liver diseases. The segmentation of liver tumors in CT images is an important premise for liver tumor load analysis, and information such as the shape, position, size, distribution, activity degree, metastasis condition and the like of the liver tumors can be rapidly and accurately acquired by using segmentation results, so that the method plays a vital role in early diagnosis, surgical treatment and radiotherapy of liver diseases.
Due to the complex anatomical structure of liver organs, large difference between different individuals, and the influence of noise, deviation, contrast agent and the like during imaging, the obtained liver CT image usually has complexity and diversity, and the liver tumor tissue in the image usually has the characteristics of fuzzy boundary, various forms, uneven gray scale and the like. The manual segmentation of the tumor region in the CT image is time-consuming and labor-consuming, the subjectivity is strong, and the segmentation result depends heavily on the experience and skill of doctors. At present, the realization of automatic segmentation of liver tumors by computers has become a hot spot of continuous research and exploration of numerous scholars at home and abroad. In order to reduce the complexity and difficulty of automatic liver tumor segmentation and improve the segmentation accuracy, most of the current automatic segmentation methods preprocess a CT image before liver tumor segmentation to obtain a liver region of interest therein. The methods disclosed in the documents "A pathological local region-based spark pattern composition for liver segmentation in CT scans" (pattern recognition, pp.88-106,2016 ") and" Medical image segmentation by combining patterns and oriented active imaging models "(IEEE transaction on image processing, pp.2035-2046, 2012) are both automatically effective for segmenting liver regions in abdominal CT images. On the basis of liver segmentation, the existing liver tumor automatic segmentation methods are mainly divided into two categories, namely unsupervised and supervised. The unsupervised segmentation method is a method for segmenting information which can be directly obtained from an image by using gray scale, gradient or texture and the like, and mainly comprises threshold value, clustering, region growing, an active contour model, image segmentation and the like. The methods only use image bottom layer data for segmentation, do not combine high-level prior knowledge, and are generally difficult to adapt to the complexity and diversity of the liver CT image. The supervised method mainly refers to a method for segmenting by combining image feature prior with machine learning, and although the method can effectively distinguish tumor tissues from normal liver parenchyma by adding training samples and solve the problems of various liver tumor forms, uneven gray levels and the like, the accurate liver tumor boundary cannot be obtained.
Disclosure of Invention
The invention fully considers the defects of the prior art, and aims to provide an automatic and accurate robust segmentation method for liver tumors in a CT image, thereby providing technical support and decision service for computer-aided diagnosis and treatment of liver diseases.
The invention is realized by the following scheme:
an automatic accurate robust segmentation method for liver tumor in CT image includes the following steps:
(1) Preprocessing the CT image by adopting a sparse shape combination or an image segmentation algorithm to obtain a liver region in the image;
(2) In order to effectively obtain a weak boundary between liver tumor and normal liver parenchyma, an image super-pixel segmentation method Based on LI-SLIC (Simple Linear Iterative Clustering on Local Information) is proposed to carry out multi-level Iterative segmentation on a liver region, the region with consistent gray level and texture in the liver is divided into the same super-pixel, and the super-pixel segmentation result is marked as S i (i =1, 2.. ·, n), wherein n is the number of superpixels;
(3) Training a normal liver parenchyma/liver tumor secondary classification classifier by using local gray features and texture features of the CT image, classifying each pixel point of the liver region obtained in the step (1) by using the trained classifier, recording a classification result as F, taking the value of F (p) as 1 if the pixel point p is classified as the normal liver parenchyma, and taking the value of F (p) as-1 if the pixel point p is classified as the tumor tissue;
(4) Classifying the superpixels generated in the step (2) according to the classification result of the pixel points in the liver region to obtain the final accurate segmentation result of the liver tumor, wherein the method comprises the following steps: for each super pixel S i The weighted sum of all the pixel classification results it contains is calculated:
wherein the weight w p And the normalization factor M are respectively defined as follows:
wherein d is p Is a pixel point p and a super pixel S i Of the center of mass, d max Is a super pixel S i The maximum Euclidean distance between all the pixel points and the centroid of the pixel points, the closer the pixel point p is to the centroid of the super pixel, the weight w p The larger the value, the greater the contribution of the pixel point to the super-pixel classification. If the weighted sum λ is calculated from the above formula i If less than 0, the corresponding super pixel S is set i Marked as tumor area, otherwise, as normal liver parenchyma.
The automatic accurate robust segmentation method for the liver tumor in the CT image is further characterized in that in the step (2), the image superpixel segmentation method based on the LI-SLIC specifically comprises the following steps:
the method comprises the following steps of (I) dividing an original image into continuous non-overlapping image sub-blocks with side length of h, wherein the value of h is as follows:
wherein N is the total number of image pixels, N 1 Setting the number of the super pixels to be generated as a natural number greater than 0;
(II) determining an initial clustering center C according to the average gray scale and the average space information of all pixels in the image sub-blocks k ：
C k ＝[l k ,x k ,y k ]
Wherein k =1,2, Λ, n 1 ，l k Is the mean value of the gray levels, x, of the pixels in the kth image sub-block k And y k The mean value of the pixel row coordinate and the pixel column coordinate in the kth image sub-block is obtained;
(III) calculating by coordinates (x) k ,y k ) Each pixel point p and cluster center C in subimage corresponding to rectangular box with center and 2h +1 as side length k Distance measure D (p, C) k )：
Wherein d is c (p,C k ) And d s (p,C k ) Respectively representing a pixel point p and a cluster center C k The parameter m is a normal number, preferably a normal number of 10 to 30, for controlling the gray-scale distance and spatial distance versus distance measure D (p, C) k ) The larger the value of m, the larger the influence of the spatial distance on it, and vice versa;
(IV) use of the distance metric D (p, C) k ) Iterative clustering is carried out on image pixels, and a clustering center C is continuously updated k Obtaining the superpixel segmentation result until the Euclidean distance between the newly updated cluster center and the previous cluster center is smaller than a preset threshold value epsilon, wherein epsilon is preferably 10 -4 A normal number of 3.
In the step (III), in order to remove the influence of image noise on the super-pixel segmentation and enhance the segmentation robustness of the texture image, the invention introduces neighborhood information to calculate the gray scale distance d c (p,C k )：
Wherein L (p) represents a neighborhood pixel set with pixel p as the center and the size of (2a + 1) × (2a + 1), a is a natural number greater than or equal to 0, preferably a natural number of 1-10, and L q Is the gray value of pixel q, l k Is the mean value of the gray levels of the pixels in the kth image sub-block, the weight w pq Satisfy the requirement of
It is defined as follows:
wherein Z is a normalization factor:
γ is the pixel gray scale standard deviation of the neighborhood set of pixels L (p).
In the step (III), the spatial distance d s (p,C k ) The calculation is as follows:
wherein x is p And y p Respectively the abscissa and ordinate, x, of pixel point p k And y k The average of the pixel row coordinates and the column coordinates in the kth image sub-block, respectively.
The automatic accurate robust segmentation method for liver tumors in CT images is further characterized in that in the step (2), a multi-level iterative segmentation method is applied to a liver region by using an image superpixel segmentation method based on LI-SLIC, and specifically comprises the following steps:
performing coarse superpixel segmentation on the liver region by adopting an image superpixel segmentation method based on LI-SLIC (Lei-SulIc), wherein the number n of superpixels to be generated 1 Preferably a natural number of 50 to 500;
(II) calculating each superpixel P i The gray standard deviation sigma of all the pixels i If σ is i Greater than a predetermined threshold value sigma (sigma is a normal number greater than 0, preferably 1 to 35), for the superpixel P i Sub-image f corresponding to the minimum circumscribed rectangle of (1) i Performing LI-SLIC superpixel segmentation, wherein the number n of superpixels to be generated 1 Preferably a natural number of 2 to 10;
(III) dividing the subimage f i Middle P i The super pixel segmentation result of the region is given to P i Realization of P i Performing one-time iterative segmentation;
(IV) repeating steps (II) and (III) until the standard deviation sigma of the gray levels of all superpixels in the image is sigma i The values are all less than or equal to a threshold value sigma, and sigma is preferably a normal number between 1 and 35;
(V) to eliminate the redundancy of segmentation caused by multi-level iterative segmentation, neighboring superpixels are usedCarrying out super-pixel combination on the gray features, and specifically comprising the following steps: for each super pixel P i Calculate P i Adjacent to it, a super pixel P j Minimum gray difference mu of i And obtaining when mu i Minimum corresponding neighboring super pixel P opt ：
Wherein the content of the first and second substances,
and
are respectively a super pixel P i And P j Mean value of gray level of (NP) i Is P i Adjacent superpixel sets of (2). Mu.f i If the value is less than the preset threshold value mu, the super pixel P is considered i Adjacent to it, a super pixel P opt Are very close and should belong to the same target area, they are merged, with the parameter mu preferably being a normal number between 5 and 30.
The automatic accurate robust segmentation method for liver tumor in CT image is characterized in that in the step (3), a normal liver parenchyma/liver tumor two-classification classifier is trained, and the method for classifying each pixel point of the liver region by using the trained classifier specifically comprises the following steps:
training a normal liver parenchyma/liver tumor two-classification classifier, wherein the method comprises the following steps: respectively selecting a sufficient number of sub-images with the size of (2b + 1) × (2b + 1) (b is a natural number larger than 0, preferably a natural number of 3-30) and containing normal liver parenchyma and liver tumor regions from the abdominal CT image as training images, extracting gray features and texture features of the training images, wherein the gray features comprise average gray, standard deviation and entropy, the texture features comprise Local Binary Pattern (LBP) features and multi-scale Gabor features, inputting the extracted features into a support vector machine, and training a classifier for classifying the normal liver parenchyma/liver tumor II;
(II) for the liver region to be detected, selecting a sub-image f with the size of (2b + 1) × (2b + 1) with each pixel point p as the center p Extracting the subimage f by the method in the step (I) p Inputting the extracted features into a trained normal parenchyma of the liver/tumor tissue two-classification classifier for classification, and assigning the classification result to a sub-image f p The center pixel point p.
Drawings
Fig. 1 (a) to 1 (c) are three original CT images, and fig. 1 (d) to 1 (f) are result diagrams of liver segmentation performed by the embodiment of the present invention;
fig. 2 is an example of a result of LI-SLIC-based image superpixel multi-level iterative segmentation, where fig. 2 (a) to 2 (c) are respectively result diagrams of performing superpixel multi-level iterative segmentation on the images shown in fig. 1 (a) to 1 (c) by using an embodiment of the present invention;
fig. 3 illustrates LBP coding, wherein fig. 3 (a) illustrates selection of pixels in a circular neighborhood, fig. 3 (b) illustrates original gray scale of an image, and fig. 3 (c) illustrates a binary pattern;
FIG. 4 is a LBP weight template, in which FIG. 4 (a) to FIG. 4 (h) are 8 weight templates generated by rotation respectively
Fig. 5 shows schematic diagrams of Gabor filters in different directions, wherein fig. 5 (a) to 5 (f) are graphs of Gabor filters in 6 directions of 0 °, 30 °, 60 °, 90 °, 120 ° and 150 °, respectively;
fig. 6 shows an example of classification results of pixel points in a liver region of a CT image, where fig. 6 (a) to 6 (c) are respectively result diagrams of classification of pixel points in the liver region shown in fig. 1 (d) to 1 (f) by using the embodiment of the present invention;
fig. 7 (a) to 7 (c) are diagrams illustrating liver tumor segmentation results of the images shown in fig. 1 (a) to 1 (c) according to the embodiment of the present invention.
Detailed Description
Example 1
An automatic accurate robust segmentation method for liver tumor in CT image comprises the following steps:
(1) Preprocessing an original CT image by adopting a sparse shape combination to obtain a liver region; FIG. 1 (a) to FIG. 1 (c) are three original CT images, and FIG. 1 (d) to FIG. 1 (f) are the results of pre-processing by the method of this embodiment;
(2) Performing multi-level iterative segmentation on a liver region by using an image superpixel segmentation method based on LI-SLIC (Living Supper-Sulframe-segmented edge segmentation), dividing a region with consistent gray level and texture in the liver into the same superpixel, acquiring a boundary between liver tumor and normal liver parenchyma, and recording a superpixel segmentation result as S i (i =1, 2.. ·, n), wherein n is the number of superpixels;
in the step (2), the image superpixel segmentation method based on the LI-SLIC specifically comprises the following steps:
dividing an original image into continuous non-overlapping image sub-blocks with the side length of h, wherein the value of h is as follows:
wherein N is the total number of image pixels, N 1 Setting the number of the super pixels to be generated as a natural number greater than 0;
(II) determining an initial clustering center C according to the average gray scale and average spatial information of all pixels in the image sub-blocks k ：
C k ＝[l k ,x k ,y k ]
Wherein k =1,2, Λ, n 1 ，l k Is the mean value of the gray levels, x, of the pixels in the kth image sub-block k And y k The mean value of the pixel row coordinate and the pixel column coordinate in the kth image sub-block is obtained;
(III) calculation with coordinates (x) k ,y k ) Each pixel point p and cluster center C in the subimage corresponding to the rectangular frame with the center and the side length of 2h +1 k Of (2) isMeasure D (p, C) k )：
Wherein, d c (p,C k ) And d s (p,C k ) Respectively representing pixel points p and cluster centers C k The parameter m is a normal number, preferably a normal number of 10 to 30, for controlling the gray-scale distance and spatial distance versus distance measure D (p, C) k ) The larger the value of m, the larger the spatial distance has an influence on it, and vice versa, and m =20 is preferred in the present embodiment. In order to remove the influence of image noise on superpixel segmentation and enhance the segmentation robustness of a texture image, neighborhood information is introduced to calculate the gray scale distance d c (p,C k )：
Where L (p) represents a neighborhood pixel set centered on pixel p and having a size of (2a + 1) × (2a + 1), and a is a natural number greater than or equal to 0, preferably a natural number between 1 and 10, and in this embodiment, a =4,l is preferred q Is the gray value of pixel q, weight w pq Satisfy the requirement of
It is defined as follows:
wherein Z is a normalization factor:
γ is the pixel gray scale standard deviation of the neighborhood set of pixels L (p). Furthermore, the spatial distance d s (p,C k ) The calculation is as follows:
wherein x is p And y p Respectively the abscissa and ordinate, x, of pixel point p k And y k The average of the pixel row coordinates and the column coordinates in the k-th image sub-block, respectively.
(IV) iterative clustering is carried out on the image pixels by adopting the distance measurement D, and the clustering center C is continuously updated k Until the euclidean distance between the newly updated cluster center and the previous cluster center is smaller than a preset threshold epsilon, a superpixel segmentation result can be obtained, and epsilon =1 is preferably selected in the embodiment;
in the step (2), performing multi-level iterative segmentation on the liver region by using an image superpixel segmentation method based on LI-SLIC, specifically comprising:
performing superpixel rough segmentation on an image by adopting an image superpixel segmentation method based on LI-SLIC (Living Supper-edge segmentation and Supper segmentation for the image), wherein the number n of superpixels to be generated 1 A natural number of 50 to 500 is preferable, and n is preferable in this embodiment 1 ＝200；
(II) computationally generating each superpixel P i The gray standard deviation sigma of all the pixels included i If σ is i If the value is larger than a preset threshold value sigma (sigma is preferably a normal number between 1 and 35, and sigma =14 is preferred in the embodiment), the image superpixel segmentation method based on the LI-SLIC is adopted to segment the superpixel P i Sub-image f corresponding to the minimum circumscribed rectangle of i Performing superpixel segmentation, wherein the number n of superpixels to be generated 1 A natural number of 2 to 10 is preferable, and n is preferable in this embodiment 1 ＝6；
(III) dividing the subimage f i Middle P i The super pixel segmentation result of the region is given to P i Realization of P i Performing one-time iterative segmentation;
(IV) repeating steps (II) and (III) until the standard deviation sigma of the gray levels of all superpixels in the image i Are all less than or equal to a threshold value sigma;
(V) in order to eliminate the segmentation redundancy caused by multi-level iterative segmentation, the gray feature between adjacent superpixels is utilized to carry out superpixelThe pixel merging specifically comprises: for each super pixel P i Calculate P i Adjacent to it, a super pixel P j Minimum gray difference mu of i And obtaining when mu i Minimum corresponding neighboring super pixel P opt ：
Wherein the content of the first and second substances,
and
are respectively a super pixel P i And P j Mean value of gray level of (NP) i Is P i Adjacent superpixel sets of (2). Mu.s of i If the value is less than the preset threshold value mu, the super pixel P is considered i Adjacent to it, a super pixel P opt The gray levels of (a) are very close and belong to the same target region, so that the gray levels are combined, wherein the parameter mu is preferably a normal number of 5-30, and the parameter mu =11 in the embodiment;
fig. 2 (a) -2 (c) are results of LI-SLIC-based image superpixel multi-level iterative segmentation performed on the CT images shown in fig. 1 (a) -1 (c) by using the method of this embodiment, and it can be seen that the generated superpixels can effectively fit weak target boundaries in the images, such as liver tumor boundaries, blood vessel boundaries, soft tissue boundaries, and the like, and pixel points included in each superpixel belong to the same target;
(3) Training a classifier for classifying normal liver parenchyma \ liver tumor two by using local gray features and texture features of the CT image, classifying each pixel point of the liver region obtained in the step (1) by using the trained classifier, recording a classification result as F, and taking the value of F (p) as 1 if the pixel point p is classified into normal liver parenchyma and-1 if the pixel point p is classified into tumor tissue; the method specifically comprises the following steps:
(I) respectively selecting 500 sub-images with the size of (2b + 1) × (2b + 1) containing normal liver parenchyma and liver tumor regions from the abdominal CT image as training images, wherein b is a natural number greater than 0, preferably a natural number of 3-30, and preferably b =10 in the embodiment;
and (II) extracting the gray features and the texture features of the training image, wherein the gray features comprise image gray mean values, standard deviations and entropies, and the texture features comprise rotation-invariant LBP features and multi-scale Gabor features.
The specific method comprises the following steps:
respectively extracting 3 gray features of the image gray mean value avg, the standard deviation std and the entropy etc by adopting the following formulas:
wherein f represents a training image, I p Is the gray value of pixel p, N f Is the total number of pixels, H, of the image f g Represents the probability of the occurrence of a gray level G in the image f, G being the number of image gray levels, G =256 for an 8-bit gray image;
(ii) extracting rotation invariant LBP features of the training image. The method comprises the following steps: 1) Performing LBP coding on each pixel in the image by using a circular neighborhood, which specifically comprises the following steps: firstly, selecting a circular neighborhood pixel as shown in fig. 3 (a), wherein a pixel shown by a black rectangle in the figure is a current pixel, taking the pixel as a center of a circle, r is a radius, and 45 ° is an interval, selecting a circular neighborhood pixel as shown by a gray rectangle in the figure, wherein r is a natural number greater than 0, preferably, r =5 in the embodiment, then comparing the gray levels of the neighborhood pixel and the current center pixel, if the gray level of the neighborhood pixel is greater than that of the center pixel, taking the corresponding binary pattern as 1, otherwise, taking the value as 0, wherein fig. 3 (b) is an image gray level example, fig. 3 (c) is an obtained binary pattern, and finally, in order to ensure that the obtained LBP code has rotational invariance, performing weighted summation on the binary pattern and 8 weight patterns obtained by rotation as shown in fig. 4, taking the minimum value thereof as the LBP code value, wherein the LBP code value totally contains 36 different code values; 2) Carrying out probability statistics on the LBP coding value of the image pixel to obtain a 36-dimensional rotation invariant LBP feature;
(iii) extracting multi-scale Gabor features of the training image. The method comprises the following steps: performing Gabor filtering on the training image by using filters with different scales and directions, in this embodiment, filtering the image by using 12 Gabor filters with sizes of 21 × 21 and 41 × 41 in 6 directions as shown in fig. 5 (a) to 5 (f), and then extracting a mean value and a variance of the filtered image, so as to obtain 24 feature values in total;
(III) inputting the extracted 3+36+24 + 63 dimensional features into a support vector machine, and training a classifier for normal liver parenchyma/liver tumor two classification;
(IV) for the liver region to be detected, selecting a subimage f with the size of (2b + 1) × (2b + 1) by taking each pixel point p as the center p Extracting the subimage f by adopting the method in the step (II) p Inputting the extracted features into a trained normal parenchyma of liver/tumor tissue two-classification classifier for classification, and assigning classification results to sub-images f p The classification result is recorded as F, if the pixel point p is classified as normal liver parenchyma, the value of F (p) is 1, and if the pixel point p is classified as tumor tissue, the value of F (p) is-1;
fig. 6 (a) to 6 (c) show the results of classifying the liver regions shown in fig. 1 (d) to 1 (f) according to the present embodiment, wherein the white region represents normal liver parenchyma, the black region represents liver tumor, and the gray region is irrelevant background;
(4) Classifying the superpixels generated in the step (2) according to the classification result F of the pixel points in the liver region to obtain the final accurate segmentation result of the liver tumor, wherein the method comprises the following steps: for each super pixel S i The weighted sum of all the pixel classification results it contains is calculated:
wherein the weight w p And the normalization factor M are defined as follows:
wherein d is p Is a pixel point p and a super pixel S i Of the centroid, d max Is a super pixel S i The maximum Euclidean distance between all the pixel points and the centroid of the pixel points, the closer the pixel point p is to the center of mass of the super pixel, the weight w p The larger the value, the greater the contribution of the pixel point to the super-pixel classification. If the weighted sum λ is calculated from the above formula i If the value is less than 0, the corresponding super pixel S is set i Marked as a tumor region, otherwise, as normal liver parenchyma.
Fig. 7 (a) to 7 (c) show the results of segmenting the CT images shown in fig. 1 (a) to 1 (c) according to the present embodiment, and it can be seen that a plurality of liver tumor regions with various shapes and fuzzy boundaries are effectively segmented completely.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. An automatic accurate robust segmentation method for liver tumor in CT image is characterized by comprising the following steps:
(1) Preprocessing a CT image to obtain a liver region in the image;
(2) Performing multi-level iterative segmentation on a liver region by using an image superpixel segmentation method based on LI-SLIC (Living Subtraction-Left theory), dividing the region with consistent gray level and texture in the liver into the same superpixel, acquiring the boundary between liver tumor and normal liver parenchyma, and marking the superpixel segmentation result as S i I =1,2,. N, where n is the number of super pixels;
(3) Training a normal liver parenchyma/liver tumor secondary classification classifier by using local gray features and texture features of the CT image, classifying each pixel point of the liver region obtained in the step (1) by using the trained classifier, recording a classification result as F, taking the value of F (p) as 1 if the pixel point p is classified as the normal liver parenchyma, and taking the value of F (p) as-1 if the pixel point p is classified as the tumor tissue;
(4) Classifying the super-pixels generated in the step (2) according to the classification result F of the pixel points in the liver region to obtain a final accurate segmentation result of the liver tumor, wherein the method comprises the following steps: for each super pixel S i The weighted sum of all the pixel classification results it contains is calculated:
wherein the weight w p And the normalization factor M are defined as follows:
wherein d is p Is a pixel point p and a super pixel S i Of the center of mass, d max Is a super pixel S i The maximum Euclidean distance between all the pixel points and the centroid of the pixel points, the closer the pixel point p is to the center of mass of the super pixel, the weight w p The larger the value is,the greater the contribution of the pixel point to the super-pixel classification; if the weighted sum λ is calculated from the above formula i If less than 0, the corresponding super pixel S is set i Marked as tumor area, otherwise, as normal liver parenchyma.
2. A method for automatic accurate robust segmentation of liver lesions in CT images as claimed in claim 1 wherein: in the step (1), the preprocessing method is sparse shape combination or graph cut algorithm.
3. The method of claim 1, wherein the method comprises the following steps: in the step (2), the image superpixel segmentation method based on the LI-SLIC specifically comprises the following steps:
the method comprises the following steps of (I) dividing an original image into continuous non-overlapping image sub-blocks with side length of h, wherein the value of h is as follows:
wherein N is the total number of image pixels, N 1 Setting the number of the super pixels to be generated as a natural number greater than 0;
(II) determining an initial clustering center C according to the average gray scale and average spatial information of all pixels in the image sub-blocks k ：
C k ＝[l k ,x k ,y k ]
Wherein k =1,2, \ 8230;, n 1 ，l k Is the mean value of the gray levels, x, of the pixels in the kth image sub-block k And y k The mean value of the pixel row coordinate and the pixel column coordinate in the kth image sub-block is obtained;
(III) calculation with coordinates (x) k ,y k ) Each pixel point p and cluster center C in the subimage corresponding to the rectangular frame with the center and the side length of 2h +1 k Distance measure D (p, C) k )：
Wherein d is c (p,C k ) Representing pixel p and cluster center C k The gray scale distance of (a) is calculated from the local neighborhood information of the pixel point p, d s (p,C k ) Representing a pixel point p and a cluster center C k M is a normal number, for controlling the gray-scale distance and the spatial distance versus distance measure D (p, C) k ) The larger the value of m, the larger the influence of the spatial distance on it, and vice versa;
(IV) using a distance metric D (p, C) k ) Performing iterative clustering on image pixels, and continuously updating a clustering center C k And obtaining a superpixel segmentation result until the Euclidean distance between the newly updated cluster center and the previous cluster center is smaller than a preset threshold epsilon.
4. A method of automatic accurate robust segmentation of liver lesions in CT images as claimed in claim 3 wherein: in the step (III), the gray scale distance d c (p,C k ) Calculated by the following formula:
wherein L (p) represents a neighborhood pixel set centered on pixel p and having a size of (2a + 1) × (2a + 1), a being a natural number greater than or equal to 0, L q Is the gray value of pixel q, l k Is the gray level average value of the pixels in the kth image sub-block, and the weight w pq Satisfy the requirements of
It is defined as follows:
wherein Z is a normalization factor:
γ is the pixel gray scale standard deviation of the neighborhood set of pixels L (p).
5. A method of automatic accurate robust segmentation of liver tumors in CT images as claimed in claim 3 wherein: in said step (III), the spatial distance d s (p,C k ) Calculated by the following formula:
wherein x is p And y p Respectively the abscissa and ordinate, x, of pixel point p k And y k The average of the pixel row coordinates and the column coordinates in the k-th image sub-block, respectively.
6. A method for automatic accurate robust segmentation of liver lesions in CT images as claimed in claim 4 wherein: m is a normal number of 10-30, and epsilon is 10 -4 -3, said a being a natural number from 1 to 10.
7. The method of claim 1, wherein the method comprises the following steps: in the step (2), a multi-level iterative segmentation method is performed on the liver region by using an image superpixel segmentation method based on LI-SLIC, and the method specifically comprises the following steps:
performing superpixel coarse segmentation on a liver region by adopting an image superpixel segmentation method based on LI-SLIC (Living Supper-Sulframe), wherein the number n of superpixels to be generated 1 Is a natural number of 50 to 500;
(II) computationally generating each superpixel P i The gray standard deviation sigma of all the pixels included i If σ is i Greater than a predetermined thresholdσ, then for the super pixel P i Sub-image f corresponding to the minimum circumscribed rectangle of i Performing LI-SLIC superpixel segmentation, wherein the number n of superpixels to be generated 1 Is a natural number of 2 to 10;
(III) dividing the subimage f i Middle P i The super pixel segmentation result of the region is given to P i Realization of P i Performing one-time iterative segmentation;
(IV) repeating steps (II) and (III) until the standard deviation sigma of the gray levels of all superpixels in the image i Are all less than or equal to a threshold value sigma;
(V) in order to eliminate the segmentation redundancy caused by multi-level iterative segmentation, the gray features between adjacent superpixels are utilized to perform superpixel combination, and the method specifically comprises the following steps: for each super pixel P i Calculate P i Adjacent to it, a super pixel P j Minimum gray difference mu of i And obtaining when mu i Minimum corresponding neighboring super pixel P opt ：
Wherein the content of the first and second substances,
and
are respectively a super pixel P i And P j Mean value of gray level of (NP) i Is P i A set of contiguous superpixels; mu.f i If the value is less than the preset threshold value mu, the super pixel P is considered i Adjacent to it, a super pixel P opt Should belong to the same target area, and then be combined.
8. The method of claim 7, wherein the method comprises the following steps: the σ is preferably a normal number of 1 to 35, and the μ is preferably a normal number of 5 to 30.
9. A method for automatic accurate robust segmentation of liver lesions in CT images as claimed in claim 1 wherein: in the step (3), a classifier for normal liver parenchyma \ liver tumor secondary classification is trained, and a method for classifying each pixel point of a liver region by using the trained classifier specifically comprises the following steps:
(I) selecting from the CT images a sufficient number of sub-images of size (2b + 1) × (2b + 1) containing normal liver parenchyma and liver tumor regions, respectively, where b is a natural number greater than 0, as training images;
(II) extracting gray features and texture features of the training image, wherein the gray features comprise an image gray mean value, a standard deviation and entropy, and the texture features comprise a rotation invariant LBP feature and a multi-scale Gabor feature;
(III) inputting the extracted features into a support vector machine, and training a normal liver parenchyma/liver tumor two-classification classifier;
(IV) for the liver region to be detected, selecting a sub-image f with the size of (2b + 1) × (2b + 1) with each pixel point p as the center p Extracting the subimage f by adopting the method in the step (II) p Inputting the extracted features into a trained normal parenchyma of liver/tumor tissue two-classification classifier for classification, and assigning classification results to sub-images f p The center pixel point p.
10. The method of claim 9, wherein the method comprises the following steps: and b is a natural number of 3 to 30.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811553883.3A CN109753997B (en) | 2018-12-19 | 2018-12-19 | Automatic accurate robust segmentation method for liver tumor in CT image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811553883.3A CN109753997B (en) | 2018-12-19 | 2018-12-19 | Automatic accurate robust segmentation method for liver tumor in CT image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109753997A CN109753997A (en) | 2019-05-14 |
| CN109753997B true CN109753997B (en) | 2022-11-22 |
Family
ID=66402800
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811553883.3A Active CN109753997B (en) | 2018-12-19 | 2018-12-19 | Automatic accurate robust segmentation method for liver tumor in CT image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109753997B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110503649B (en) * | 2019-07-05 | 2022-09-30 | 陕西科技大学 | Liver segmentation method based on spatial multi-scale U-net and superpixel correction |
| CN110647939B (en) * | 2019-09-24 | 2022-05-24 | 广州大学 | Semi-supervised intelligent classification method and device, storage medium and terminal equipment |
| CN110706794B (en) * | 2019-09-26 | 2022-10-25 | 中国科学院深圳先进技术研究院 | Medical image processing system and medical image processing method |
| CN111476292B (en) * | 2020-04-03 | 2021-02-19 | 北京全景德康医学影像诊断中心有限公司 | Small sample element learning training method for medical image classification processing artificial intelligence |
| CN115148330B (en) * | 2022-05-24 | 2023-07-25 | 中国医学科学院北京协和医院 | POP treatment scheme forming method and system |
| CN116188786B (en) * | 2023-05-04 | 2023-08-01 | 潍坊医学院附属医院 | Image segmentation system for hepatic duct and biliary tract calculus |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105957063A (en) * | 2016-04-22 | 2016-09-21 | 北京理工大学 | CT image liver segmentation method and system based on multi-scale weighting similarity measure |
| CN108596887A (en) * | 2018-04-17 | 2018-09-28 | 湖南科技大学 | A kind of abdominal CT sequence image liver neoplasm automatic division method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107977952A (en) * | 2016-10-21 | 2018-05-01 | 冯原 | Medical image cutting method and device |
-
2018
- 2018-12-19 CN CN201811553883.3A patent/CN109753997B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105957063A (en) * | 2016-04-22 | 2016-09-21 | 北京理工大学 | CT image liver segmentation method and system based on multi-scale weighting similarity measure |
| CN108596887A (en) * | 2018-04-17 | 2018-09-28 | 湖南科技大学 | A kind of abdominal CT sequence image liver neoplasm automatic division method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109753997A (en) | 2019-05-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109753997B (en) | Automatic accurate robust segmentation method for liver tumor in CT image | |
| Wu et al. | Automatic liver segmentation on volumetric CT images using supervoxel-based graph cuts | |
| CN108596887B (en) | Automatic segmentation method for liver tumor region image in abdominal CT sequence image | |
| CN108364294B (en) | Multi-organ segmentation method for abdominal CT image based on superpixels | |
| CN109934235B (en) | Unsupervised abdominal CT sequence image multi-organ simultaneous automatic segmentation method | |
| Kumar et al. | Automatic liver and lesion segmentation: a primary step in diagnosis of liver diseases | |
| Jeyavathana et al. | A survey: analysis on pre-processing and segmentation techniques for medical images | |
| Pulagam et al. | Automated lung segmentation from HRCT scans with diffuse parenchymal lung diseases | |
| Chen et al. | Pathological lung segmentation in chest CT images based on improved random walker | |
| Memon et al. | Segmentation of lungs from CT scan images for early diagnosis of lung cancer | |
| Casiraghi et al. | Automatic abdominal organ segmentation from CT images | |
| Fan et al. | Lung nodule detection based on 3D convolutional neural networks | |
| Susomboon et al. | A hybrid approach for liver segmentation | |
| Zidan et al. | Level set-based CT liver image segmentation with watershed and artificial neural networks | |
| Sagar et al. | Color channel based segmentation of skin lesion from clinical images for the detection of melanoma | |
| Shelke et al. | Automated segmentation and detection of brain tumor from MRI | |
| Jiang et al. | An adaptive region growing based on neutrosophic set in ultrasound domain for image segmentation | |
| Fazilov et al. | Improvement of Image Enhancement Technique for Mammography Images | |
| Borzooie et al. | Modified density-based data clustering for interactive liver segmentation | |
| Sasikala et al. | Soft Computing based Brain Tumor Categorization with Machine Learning Techniques | |
| Alirr et al. | Automatic liver segmentation from ct scans using intensity analysis and level-set active contours | |
| Abdellatif et al. | K2. Automatic pectoral muscle boundary detection in mammograms using eigenvectors segmentation | |
| Vanmore et al. | Survey on automatic liver segmentation techniques from abdominal CT images | |
| Song et al. | Liver segmentation based on SKFCM and improved GrowCut for CT images | |
| Dawod et al. | Adaptive Slices in Brain Haemorrhage Segmentation Based on the SLIC Algorithm. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |