A Memetic Method for Segmentation of Gene Chip Images

By introducing the Memetic method into the k-means algorithm, using new cross operators and population update strategies, the problem that traditional algorithms are prone to fall into local optimal solutions is solved, and the accuracy and efficiency of gene chip image segmentation are improved.

CN114596438BActive Publication Date: 2024-11-12NANJING UNIV OF POSTS & TELECOMM
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210237176.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-10
Publication Date
2024-11-12
Estimated Expiration
2042-03-10

AI Technical Summary

Technical Problem

Traditional k-means algorithms are prone to fall into local optimal solutions when processing large-scale gene chip image segmentation, and it is difficult to find global optimal solutions.

Method used

The Memetic method is adopted, combined with the k-means algorithm as local search, and a new cross operator and diversity-based population update strategy are used to enhance the global search ability and avoid local optimal solutions.

Benefits of technology

It improves the accuracy and efficiency of gene chip image segmentation, overcomes the problem of local optimal solutions, and can better handle large-scale gene chip image segmentation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114596438B_ABST
    Figure CN114596438B_ABST
Patent Text Reader

Abstract

A Memetic method for solving gene chip image segmentation, which mainly consists of four parts: population initialization, i.e., random generation p initial solutions and thus form an initial population; use the k-means algorithm as a local optimization method; a crossover operator for generating offspring solutions; and a population update method based on population diversity. By using a new crossover operator and a population update strategy based on diversity, this method enhances the global search capability and overcomes the disadvantage of using the traditional k-means algorithm that is prone to falling into the local optimal solution. This greatly improves the accuracy and efficiency of the gene chip image segmentation results and is conducive to processing large-scale gene chip image segmentation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention is applicable to the field of image segmentation and provides a Memetic method for solving gene chip image segmentation. Background Art

[0002] With the gradual implementation of the human genome (sequencing) project and the rapid development of molecular biology related disciplines, more and more animal, plant and microbial genome sequences have been determined, and gene sequence data is growing at an unprecedented rate. However, how to study the functions of so many genes in the life process has become a common topic for life science workers around the world. Gene chip (also known as DNA chip, biochip) technology is a product that complies with the requirements of this scientific development. This technology refers to the process of fixing a large number of probe molecules (usually with a dot density of more than 400 per square centimeter) on a support and hybridizing them with labeled sample molecules, and obtaining the number and sequence information of sample molecules by detecting the hybridization signal intensity of each probe molecule. By processing and analyzing the gene chip hybridization detection image, a large amount of genetic information in biological cells or tissues can be analyzed.

[0003] Image segmentation is an essential step in gene chip applications. The process of gene chip image segmentation is to determine the target signal (target) and the background signal, that is, to identify the target signal in the background, in order to locate each target and obtain various relevant information corresponding to each target for further analysis and research.

[0004] The above image segmentation problem can be converted into a least squares clustering problem. As the most widely used k-means clustering algorithm in cluster analysis applications, it can be used to solve the image segmentation processing of gene chips. However, since there are a large number of local optimal solutions when processing large-scale gene chip image segmentation, it is difficult for simple local search algorithms (such as k-means) to find the global optimal solution, so it is necessary to design a Memetic method for solving gene chip image segmentation to better find the global optimal solution of the objective function. Summary of the invention

[0005] The present invention aims to solve the disadvantage of using the traditional k-means algorithm to segment large-scale gene chip images, which is prone to fall into the local optimal solution and enhance the global search ability of the algorithm. A Memetic method for solving gene chip image segmentation is proposed. The algorithm uses k-means as a local search and a new crossover operator and a population update strategy based on diversity as a global search. The method is not prone to fall into the local optimal solution and enhances the diversity of the solution and the global search ability, which is conducive to the segmentation of large-scale gene chip images.

[0006] A Memetic method for solving gene chip image segmentation includes the following steps:

[0007] Step 1, input the gray value vectors of all pixels in the gene chip image after preprocessing and grid positioning, randomly generate p initial solutions and construct an initial population P from them;

[0008] Step 2: Use the k-means algorithm to perform local optimization on each individual in population P, and then record the best solution in population P, denoted as S b ;

[0009] Step 3, randomly select two parent solutions from the population P, and use a new crossover operator to generate a new offspring solution S0;

[0010] Step 4, use k-means algorithm to perform local search on S0;

[0011] Step 5: Update the population P using the diversity-based population update strategy and S0. Then compare S0 with the best solution S found so far. b Make a comparison to determine whether to update S b ;

[0012] Step 6, determine whether the best solution is improved. If there is an improvement, execute NoImprove←0 (NoImprove refers to the number of consecutive no improvements, initialized to 0), and then jump to step 3; if there is no improvement, increase NoImprove by 1, and then go to step 7.

[0013] Step 7: Determine whether NoImprove≤L max (L max Refers to the search depth) is true, if true, jump to step 3; if not, the algorithm stops and outputs the optimal solution.

[0014] Furthermore, in step 1, the initial population is constructed as follows:

[0015] Step 1.1: From the given data set X = {x1, x2, ..., x n}(n refers to the total number of nodes in the data set X) randomly selects K (K refers to the total number of classes) points as the initial center, denoted as y q (q=1,2,...,K), and then assign the unassigned points to the cluster C where the center point closest to it is located in Euclidean distance. q (q=1,2,...,K) until all points in X have been assigned, and this solution is recorded as S1;

[0016] Step 1.2: Execute the initial solution generation procedure p times continuously to obtain p initial solutions, thus forming an initial population P = (S1, S2, ..., S p ).

[0017] Furthermore, in step 2, the steps of using the k-means algorithm to perform local optimization on each individual in the population P are as follows:

[0018] Step 2.1: Update the center of each class in the solution to the centroid of all points in that class, expressed as:

[0019]

[0020] Among them, |C q | represents the number of nodes of the qth class.

[0021] Step 2.2: If the center no longer changes, exit the loop; otherwise, assign each point in the data set to the center closest to it to form a new cluster, and then jump to step 2.1.

[0022] Finally, according to the objective function value, record the best solution in the current population P, denoted as S b Here, the objective function value f(s) refers to the sum of the squares of the Euclidean distances from each node in each class to its respective center point, which can be expressed as:

[0023]

[0024] in, Refers to each node x in each class i To the center point y of each class q The square of the Euclidean distance.

[0025] Furthermore, in step 3, the main steps of using a new crossover operator to generate a new offspring solution S0 are as follows:

[0026] Step 3.1: Randomly select two solutions from the population P as parent solutions and record them as P1 and P2. Definition and are the various types of P1 and P2 respectively, and are the various center points of P1 and P2 respectively.

[0027] Step 3.2: For the center points of the two solutions P1 and P2 and Here, we use a greedy matching criterion to match the class center point sets of the two solutions. The greedy matching method has K iterations. At each step, if the Euclidean distance between the two class centers is the smallest, they are paired with each other until all classes are paired.

[0028] Step 3.3: Cross the two parent solutions P1 and P2 to generate a descendant solution S0. First, take the intersection of the K paired classes respectively. Get a new K classes Next, use formula (1) to calculate each class The center point Then, the remaining points after taking the intersection during pairing are assigned to the class of the center point with the closest Euclidean distance to it, until all the remaining points have been assigned; finally, hybridization generates a new offspring solution S0.

[0029] Furthermore, in step 5, we define a solution S i (i=1,2,...,p) Scoring function for population P:

[0030]

[0031] Among them, α is a parameter, f(S i ) is S i (i=1,2,...,p), D(S i ,P) represents an S i The distance between (i=1,2,...,p) and population P, which is defined as S i The minimum distance between the solution and other solutions in the population P is expressed as:

[0032] D(S i ,P)=min{d ij |S j ∈P,j≠i,j=1,2,...,p} (4)

[0033] Among them, d ij Indicates S i With S j The distance between them is defined as S i With S j The sum of the Euclidean distances between the corresponding cluster centers in is expressed as:

[0034]

[0035] in, and Respectively refers to S i and S jThe corresponding C q center.

[0036] Scoring function Score(S i ,P) is smaller, indicating that S i The better the quality.

[0037] The population update rule is: first insert the offspring solution S0 into the population P, that is, P'=P∪{S0}; then calculate each S in the population P' i Score(S i ,P), and record the solution with the worst quality (i.e. the largest scoring function value), denoted as S worst Then replace the worst solution S in the population P' with S0 worst , that is, P = P ∪ {S0} \ {S worst}.

[0038] Finally, compare the next solution S0 with the best solution S found so far b For comparison, if f(S0)<f(S b ), then use S0 to update S b .

[0039] The beneficial effects achieved by the present invention are as follows: by utilizing a new crossover operator and a population update strategy based on diversity, the present method enhances the global search capability and overcomes the shortcoming of the traditional k-means algorithm being prone to falling into a local optimal solution, which greatly improves the accuracy and efficiency of the gene chip image segmentation results and is conducive to processing large-scale gene chip image segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is an image of a gene chip image in an embodiment of the present invention after preprocessing and grid positioning.

[0041] Figure 2 It is a general flow chart of the Memetic method combined with the k-means method in an embodiment of the present invention.

[0042] Figure 3 It is a schematic diagram of matching the center points of two parent solutions in an embodiment of the present invention.

[0043] Figure 4 Schematic diagram of a crossover operator in an embodiment of the present invention.

[0044] Figure 5 It is a schematic diagram of the population update process in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings.

[0046] like Figure 1 As shown in the figure, in the target areas of multiple gene chip images after preprocessing and grid positioning, each target area consists of a pixel and its background area. The circle in each small square represents a pixel, and a pixel can be represented as a point. Then the point vector value corresponds to the pixel gray value vector, that is, the horizontal coordinate and vertical coordinate of the point vector correspond to the horizontal coordinate and vertical coordinate of the pixel gray value vector. Therefore, the image segmentation process is transformed into the clustering of point vectors, thereby dividing all pixels into two categories: target and background. The quality of the clustering result is determined by the objective function To determine (K is generally 2), the optimal image segmentation result is solved by continuously searching for the minimum value of the objective function.

[0047] The present invention proposes a Memetic method for solving gene chip image segmentation, such as Figure 2 As shown, the following steps are included:

[0048] Step 1: Input the gray value vectors of all pixels in the gene chip image after preprocessing and grid positioning, randomly generate 10 initial solutions and construct an initial population P from them.

[0049] In step 1, the method for constructing the initial population is as follows:

[0050] Step 1.1: From the given data set X = {x1, x2, ..., x n} (n refers to the total number of nodes in the data set X, that is, the number of pixels) randomly select K (K refers to the total number of classes, K is generally 2) points as the initial center, recorded as y q (q=1,2,...,K), and then assign the unassigned points to the cluster C where the center point closest to it is located in Euclidean distance. q (q=1,2,...,K) until all points in X have been assigned, and this solution is recorded as S1.

[0051] Step 1.2: Execute the initial solution generation procedure 10 times in succession (i.e., execute step 1.1 10 times respectively) to obtain 10 initial solutions, thus forming an initial population P = (S1, S2, ..., S 10 ).

[0052] Step 2: Use the k-means algorithm to perform local optimization on each individual in population P, and then record the best solution in population P, denoted as S b .

[0053] In step 2, the steps of using the k-means algorithm to perform local optimization on each individual in the population P are as follows:

[0054] Step 2.1: Update the center of each class in the initial solution to the centroid of all points in that class, expressed as:

[0055]

[0056] Among them, |C q | represents the number of nodes of the qth class.

[0057] Step 2.2: If the center no longer changes, exit the loop; otherwise, assign each point in the data set to the center closest to it to form a new cluster, and then jump to step 2.1.

[0058] Finally, according to the objective function value, record the best solution in the current population P, denoted as S b Here, the objective function value f(s) refers to the sum of the squares of the Euclidean distances from each node in each class to its respective center point, which can be expressed as:

[0059]

[0060] in, Refers to each node x in each class i To the center point y of each class q The square of the Euclidean distance.

[0061] Step 3, randomly select two parent solutions from the population P and use a new crossover operator to generate a new offspring solution S0.

[0062] In step 3, the main steps of using a new crossover operator to generate a new offspring solution S0 are as follows:

[0063] Step 3.1: Randomly select two solutions from the population P as parent solutions and record them as P1 and P2. Definition and are the various types of P1 and P2 respectively, and are the various center points of P1 and P2 respectively.

[0064] Step 3.2: For the center points of the two solutions P1 and P2 and For matching. Figure 3 As shown in Figure 1, a greedy matching criterion is used to match the cluster center point sets of the two solutions. The greedy matching method has K iterations. At each step, if the Euclidean distance between the two cluster center points is the smallest, they are paired with each other until all clusters are paired.

[0065] Step 3.3: Cross the two parent solutions P1 and P2 to generate a descendant solution S0. First, take the intersection of the K paired classes respectively. Get a new K classes Next, use formula (1) to calculate each class The center point Then, the remaining points after taking the intersection during pairing are assigned to the class of the center point with the closest Euclidean distance to it, until all the remaining points have been assigned; finally, hybridization generates a new offspring solution S0. The hybridization process is as follows Figure 4 shown.

[0066] Step 4: Use the k-means algorithm to perform a local search on S0. The k-means algorithm local search process is shown in step 2.

[0067] Step 5: Update the population P using the diversity-based population update strategy and S0. Then compare S0 with the best solution S found so far. b Make a comparison to determine whether to update S b .

[0068] In step 5, a solution S is defined i (i=1,2,...,10) Scoring function for population P:

[0069]

[0070] Among them, α is a parameter, the value is α = 0.08n, f(S i ) is S i (i=1,2,...,10), D(S i ,P) represents an S i The distance between (i=1,2,...,10) and population P, which is defined as S i The minimum distance between the solution and other solutions in the population P is expressed as:

[0071] D(S i ,P)=min{d ij |S j ∈P,j≠i,j=1,2,...,10} (4)

[0072] Among them, d ij Indicates S i With S j The distance between them is defined as S i With S j The sum of the Euclidean distances between the corresponding cluster centers in is expressed as:

[0073]

[0074] in, and Respectively refers to S i and S j The corresponding C q center.

[0075] Scoring function Score(S i ,P) is smaller, indicating that S i The better the quality.

[0076] like Figure 5 As shown in the figure, the population update rule is: first insert the offspring solution S0 into the population P, that is, P'=P∪{S0}; then calculate each S in the population P' i Score(S i ,P), and record the solution with the worst quality (i.e. the largest scoring function value), denoted as S worst Then replace the worst solution S in the population P' with S0 worst , that is, P = P ∪ {S0} \ {S worst}.

[0077] Finally, compare the next solution S0 with the best solution S found so far b For comparison, if f(S0)<f(S b ), then use S0 to update S b .

[0078] Step 6, determine whether the best solution in step 5 has been updated. If there is an improvement, execute NoImprove←0 (NoImprove refers to the number of consecutive no improvements, initialized to 0), and then jump to step 3; if there is no improvement, increase NoImprove by 1, and then go to step 7.

[0079] Step 7: Determine whether NoImprove≤L max (L max Refers to the search depth, L max If yes, then jump to step 3; if no, then the process stops and the optimal solution, i.e., the optimal image segmentation result, is output.

[0080] The above description is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiment. Any equivalent modifications or changes made by ordinary technicians in this field based on the contents disclosed by the present invention should be included in the protection scope recorded in the claims.

Claims

1. A Memetic method for solving gene chip image segmentation, characterized by: The method comprises the following steps: Step 1, input the gray value vectors of all pixels in the gene chip image after preprocessing and grid positioning, randomly generate p initial solutions and construct an initial population P from them; Step 2: Use the k-means algorithm to perform local optimization on each individual in population P, and then record the best solution in population P, denoted as S b ; Step 3, randomly select two parent solutions from the population P and use the crossover operator to generate a new offspring solution S0; Step 4, use k-means algorithm to perform local search on S0; Step 5: Update the population P using the diversity-based population update strategy and S0; then compare S0 with the best solution S found so far. b Make a comparison to determine whether to update S b ; Step 6, determine whether the best solution is improved. If it is improved, execute NoImprove←0, where NoImprove refers to the number of consecutive times that the solution has not been improved, initialized to 0, and then jump to step 3; If there is no improvement, increase NoImprove by 1 and go to step 7; Step 7: Determine whether NoImprove≤L max Is it established? L max Refers to the search depth. If true, jump to step 3; If not, the process stops and the optimal solution, that is, the optimal image segmentation result, is output.

2. The Memetic method for solving gene chip image segmentation according to claim 1, characterized in that: The construction method of the initial population in step 1 is as follows: Step 1.1: From the given data set X = {x1, x2, ..., x n } randomly select K points as the initial center, denoted as y q , q = 1, 2, ..., K, n refers to the total number of nodes in the data set X, K refers to the total number of classes, and then the unassigned points are assigned to the class C where the center point with the closest Euclidean distance is located. q , q=1,2,...,K, until all points in X have been assigned, and this solution is recorded as S1; Step 1.2: Execute the initial solution generation procedure p times continuously to obtain p initial solutions, thus forming an initial population P = (S1, S2, ..., S p ).

3. The Memetic method for solving gene chip image segmentation according to claim 1, characterized in that: The steps of using the k-means algorithm to perform local optimization on each individual in the population P in step 2 are as follows: Step 2.1: Update the center of each class in the solution to the centroid of all points in that class, expressed as: Among them, |C q | represents the number of nodes in the qth class; Step 2.2: If the center no longer changes, exit the loop; otherwise, assign each point in the data set to the center closest to it to form a new cluster, and then jump to step 2.1; Step 2.3: Record the best solution in the current population P according to the objective function value, denoted as S b ; The objective function value f(s) refers to the sum of the squares of the Euclidean distances from each node in each class to its respective center point, which can be expressed as: in, Refers to each node x in each class i To the center point y of each class q The square of the Euclidean distance.

4. The Memetic method for solving gene chip image segmentation according to claim 1, characterized in that: The main steps of using the crossover operator in step 3 to generate a new offspring solution S0 are as follows: Step 3.1: Randomly select two solutions from the population P as parent solutions and record them as P1 and P2; define and are the various types of P1 and P2 respectively, and are the center points of the clusters P1 and P2, i=1,2,...,K, j=1,2,...,K; Step 3.2: For the center points of the two solutions P1 and P2 and Matching is performed; the class center point sets of the two solutions are matched pairwise using the greedy matching criterion; the greedy matching criterion has K iteration steps, and at each step, if the Euclidean distance between the two class centers is the smallest, they are paired with each other until all classes are paired; Step 3.3: Cross the two parent solutions P1 and P2 to generate a descendant solution S0; first, take the intersection of the K paired groups respectively Get a new K classes i=1,2,...,K; Then, use formula (1) to calculate the The center point i=1,2,...,K; then, assign the remaining points after taking the intersection during pairing to the class where the center point with the closest Euclidean distance is located, until all the remaining points have been assigned; finally, hybridization generates a new offspring solution S0.

5. The Memetic method for solving gene chip image segmentation according to claim 1, characterized in that: In step 5, a solution S is defined i For the scoring function of population P, i=1,2,...,p: Among them, α is a parameter, f(S i ) is S i The objective function value, D(S i ,P) represents S i The distance from population P is defined as S i The minimum distance between the solution and other solutions in the population P is expressed as: D(S i ,P)=min{d ij |S j ∈P,j≠i,j=1,2,...,p} (4) Among them, d ij Indicates S i With S j The distance between them is defined as S i With S j The sum of the Euclidean distances between the corresponding cluster centers in is expressed as: in, and Respectively refers to S i and S j The corresponding C q the center of Scoring function Score(S i ,P) is smaller, indicating that S i The better the quality; The population update rule is to first insert the offspring solution S0 into the population P, that is, P'=P∪{S0}; then calculate each S in the population P' i Score(S i ,P), and record the solution with the worst quality, i.e. the solution with the largest scoring function value, denoted as S worst ; Then replace the worst solution S in the population P' with S0 worst , that is, P = P ∪ {S0} \ {S worst }; Finally, compare the next solution S0 with the best solution S found so far b For comparison, if f(S0)<f(S b ), then use S0 to update S b .