Two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection

Abstract

The invention provides a two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection, and belongs to the technical field of medical information intelligent processing. The technical problems that a traditional method is insufficient in global search capability and low in selection efficiency in processing high-dimensional data feature selection, and indeterminacy and fuzziness exist in complex schizophrenia disease feature selection are solved. According to the technical scheme, the method comprises the following steps: S1, obtaining and preprocessing schizophrenia disease data; s2, constructing a fuzzy neighborhood decision space and fuzzy neighborhood mutual information according to the data; s3, a schizophrenia disease feature selection module of fuzzy neighborhood differential evolution; s4, a schizophrenia disease feature selection module of a fuzzy neighborhood decision tree; and S5, carrying out feature selection fusion on the two stages. The method has the beneficial effects that the uncertainty in the data can be effectively processed, and the global search capability and the selection efficiency in a high-dimensional space are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent processing of medical information, and in particular to a two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection. Background Art

[0002] With the growth of the global population and the increasing stress of daily life, schizophrenia has become a major mental illness that has attracted much attention. The disease can cause serious damage to an individual's cognitive functions, such as attention, learning ability, memory, and language. Early diagnosis is essential for the effective management and treatment of schizophrenia. However, accurate diagnosis faces great challenges because its clinical manifestations often resemble symptoms of normal aging or other mental illnesses. Patients with schizophrenia can choose a range of effective health care options, including drug treatment, psychoeducation, family intervention, cognitive behavioral therapy, and psychosocial rehabilitation (such as life skills training). However, current traditional diagnostic methods mainly rely on clinical mental status examinations and patients' subjective reports, which have obvious limitations, including strong subjectivity, qualitative analysis, high uncertainty, and high misdiagnosis rates.

[0003] At present, resting state functional magnetic resonance imaging (fMRI) is gradually being applied to solve schizophrenia. This technology measures the blood oxygen level dependent signals of different brain regions and calculates the time series correlation to evaluate the functional connection strength between brain regions, and then classifies and selects the features of schizophrenia by using traditional machine learning algorithms (such as K nearest neighbor, random forest and support vector machine). However, these methods have some technical problems: the dimension of fMRI data is usually very high, and each scanning time point contains information of thousands of voxels (volume pixels). This combination of high dimension and low sample size will cause the machine learning model to be easily overfitted. High-dimensional data also increases the complexity of calculation and storage requirements, which brings challenges to data processing and analysis. In order to solve these problems, the present invention provides a two-stage fuzzy neighborhood feature selection method for schizophrenia detection. Compared with the traditional method, this method depicts the fuzzy boundary and search space set to deal with the common fuzziness and imprecision in actual schizophrenia data. By using the fuzzy neighborhood differential evolution module and the fuzzy neighborhood decision tree module, the influence of each feature on the decision is evaluated according to the fuzzy neighborhood mutual information of each feature, thereby effectively suppressing irrelevant features. This method not only improves the coverage of the search, but also ensures that the features with information content for predicting the target variable can be quickly and accurately identified during the feature selection process. Therefore, the present invention provides a more efficient, accurate, stable and reliable method for detecting schizophrenia, which can be applied to clinical diagnosis and disease prevention in doctor-patient groups. Summary of the invention

[0004] The purpose of the present invention is to provide a two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection, aiming to improve the accuracy and generalization ability of mental classification disease detection. This method shows broad prospects in clinical medical applications. The fuzzy neighborhood differential evolution module increases the tolerance to uncertainty and noise data through fuzzy logic, making the feature selection process more robust. The accuracy of feature evaluation is further enhanced by combining the fuzzy neighborhood decision tree. Because mutual information can quantify the amount of mutual information between features and target variables, the application of fuzzy neighborhood allows the algorithm to more accurately evaluate this correlation in the presence of data ambiguity, ultimately providing an effective solution for high-dimensional mental classification disease data feature selection.

[0005] In order to achieve the above-mentioned invention object, the present invention adopts the following technical scheme: a two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection, comprising the following steps:

[0006] S1: Acquisition and preprocessing of schizophrenia disease data: First, a 3T General Electric MRI scanner was used to acquire imaging data through resting-state functional magnetic resonance imaging (rs-fMRI). In the data preprocessing stage, SPM8 software was used to stabilize, correct, and filter the acquired data. The collected images were segmented into multiple regions using the AAL module, and average time series data were extracted from each region to obtain feature information. After data processing, a dataset containing multiple samples and features was constructed;

[0007] S2: Constructing fuzzy neighborhood decision space and fuzzy neighborhood mutual information: Construct a fuzzy neighborhood decision space FNDS in the feature space<U,C∪D,V,f> ,U={x1,x2,x3,...,x N} is the domain, C and D are the conditional attribute set and decision attribute set respectively, V is the value range of all attributes, and f is the related information function. This space not only considers the distance between data samples, but also considers the fuzzy relationship between them, thus more accurately reflecting the heterogeneity of schizophrenia disease sample information. Fuzzy probability distribution is used to calculate the fuzzy neighborhood mutual information FNMI δ (S1; S2), where S1 and S2 are subsets of attributes in F. Fuzzy neighborhood mutual information not only considers the direct correlation between features, but also the fuzzy relationship between features;

[0008] S3: Design the first stage: schizophrenia disease feature selection module based on fuzzy neighborhood differential evolution, improve fuzzy neighborhood and define a new dynamic neighborhood radius calculation method x iis the midpoint of the fuzzy boundary. In the early stage of the algorithm, the neighborhood range is wider to enhance the search ability, while in the later stage, the neighborhood range is narrowed to refine the boundary area. Then, a new mutation strategy v for the offspring g is proposed by combining the fuzzy boundary and the mutation factor. i (g), dynamically adjust the weight coefficient to adapt to different data dimensions and spaces. Finally, set fuzzy rules to dynamically adjust the crossover factor CR to increase the robustness and effect of the algorithm, so as to better adapt to the population diversity and algorithm operation process. The final crossover factor is obtained through defuzzification to ensure the effectiveness of mutation and crossover operations;

[0009] S4: Design the second stage: the schizophrenia feature selection module of the fuzzy neighborhood decision tree, which divides the data set by selecting attributes with high consistency with the class label to generate a simpler and more generalized decision tree. During the construction process, first check the number of samples and the number of categories of the current node. If there is only one sample or category in the node, stop growing. Otherwise, start node division: calculate the fuzzy neighborhood mutual information when Ai=cj for each possible value cj of each feature Ai. Among all attributes and all their split points, select the attribute with the largest FNMI value and its corresponding split point as the optimal attribute A and the optimal split point c*. If the fuzzy neighborhood complementary mutual information FNMI(A) of attribute A to the data set is less than the threshold condition, the decision tree stops growing; otherwise, the sample of the node is split into two child nodes according to the optimal attribute and the optimal split point. According to the comparison between FNMI(A) and the average value, if the former is lower than the average value, the decision tree is pruned and optimized through iteration. Otherwise, the pruning optimization process ends to obtain the final optimized fuzzy neighborhood mutual information decision tree FNMIDT;

[0010] S5: The two stages are integrated to form a complete feature selection process: In the first stage, a subset of schizophrenia disease features is obtained, and the model-based feature selection process is entered in the second stage. In the second stage, feature selection is continuously optimized through modeling feedback, and verification is used to ensure the rationality of feature selection. The final schizophrenia detection model is constructed using the selected feature set to classify and predict clinical data.

[0011] As a further optimization scheme of the two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection provided by the present invention, the specific steps of step S2 are as follows:

[0012] Step S2.1: Construct a FNDS= according to the data information<U,C∪D,V,f> Fuzzy neighborhood decision system, U = {x1,x2,x3,...,x N} is the domain, C and D are the condition attribute set and decision attribute set respectively, V is the value range of all attributes, f is the related information function, B is the attribute subset in the domain U,

[0013] Step S2.2: Construct fuzzy similarity relationship and fuzzy similarity matrix, in the fuzzy neighborhood decision system FNDS =<U,C∪D,V,f> , R a is the fuzzy similarity relationship derived from feature a, and is defined as

[0014]

[0015] for Fuzzy similarity matrix[x] a (y) = R a (x,y), then the fuzzy similarity matrix of x on U with respect to B is defined as

[0016]

[0017] Among them, R B is a fuzzy similarity relationship with a value range between [0,1]. The closer the value is to 1, the more similar the samples are. The fuzzy similarity matrix [x] a (y) is an n×n matrix, where n is the number of samples;

[0018] Step S2.3: Construct parameterized fuzzy neighborhood information granularity, where the information granularity can be adjusted by parameters. This method is usually used to deal with uncertainty and ambiguity, and better capture and utilize information in data analysis and decision-making. Fuzzy neighborhood decision system FNDS = <U,C∪D,V,f>, for The granularity of parameterized fuzzy neighborhood information is constructed as follows:

[0019]

[0020] Among them, δ represents the fuzzy neighborhood radius, 0≤δ≤1, through the fuzzy similarity relationship R B and the neighborhood radius δ is determined The fuzzy neighborhood particles control the size of the fuzzy neighborhood, thus affecting the coarseness and fineness of the information granularity;

[0021] Step S2.4: Using the fuzzy probability distribution, calculate the fuzzy neighborhood mutual information. FNDS = <U,C∪D,V,f> is a fuzzy neighborhood decision system, U = {x1,x2,x3,...,x N} is the domain of discourse, this set of samples has numerical or discrete features F, where S1 and S2 are subsets of attributes in F, and sample x1 can be expressed in the feature subspace S1∪S2 as Then the fuzzy neighborhood mutual information is expressed as:

[0022]

[0023] Then the fuzzy neighborhood conditional mutual information is derived. This group of samples has a numerical or discrete feature F, where S is a subset of the attributes in F, and D is the decision attribute. Then the fuzzy neighborhood conditional mutual information of S for D is expressed as follows:

[0024]

[0025] Among them, fuzzy neighborhood mutual information FNMI δ (S1; S2) reflects the information relationship between attribute subsets S1 and S2, N represents the variable parameters, and the fuzzy neighborhood conditional mutual information FNCMI δ (D; S) reflects the relationship between the attribute subset S and the decision attribute D. The larger the fuzzy neighborhood conditional mutual information between D and S, the stronger the correlation between S and D.

[0026] As a further optimization scheme of the two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection provided by the present invention, the specific steps of step S3 are as follows:

[0027] Step S3.1: The fuzzy neighborhood differential evolution module uses the fuzzy boundary to select the uncertainty of data and form an uncertain data search band, where the membership function is defined as:

[0028]

[0029] Among them, x i is the midpoint of the fuzzy boundary, x j For all individuals, find the Euclidean distance between them, and d is the radius of the fuzzy boundary. Different from the original fuzzy neighborhood, the improved fuzzy neighborhood adopts a new calculation method for the neighborhood radius δ, and its dynamic changes are as follows:

[0030]

[0031]

[0032] Among them, f is a constant parameter controlling the fuzzy boundary, F is the number of individuals in the fuzzy boundary, pol and Maxpol are the current population and the maximum population respectively, and x F is the point closest to the center x in the fuzzy boundary i The Fth individual of

[0033] Step S3.2: Select a larger neighborhood range in the early stage to improve the search ability, and select a smaller neighborhood range in the later stage to find the optimal edge area, select individuals according to the membership degree, and form the final neighborhood. The method is as follows:

[0034]

[0035] in By x i The final neighborhood boundary determined, rand j is a random value in the jth dimension [0,1], NP is the total number, and then a new mutation strategy is proposed by combining the fuzzy boundary with the mutation factor, as shown below:

[0036]

[0037] Among them, W FN is the weight coefficient of the fuzzy neighborhood, F' is the new variation factor combined with the fuzzy neighborhood weight, x FNB (g) is the center point individual within the fuzzy boundary, and To obtain two more individuals within the fuzzy boundary, the weight coefficient W FN The range is [0,1], so when W FN As F' decreases as it approaches 0, FN When it is close to 1, F' will increase, and finally the target individual v i (g);

[0038] Step S3.3: During the crossover process, the crossover factor CR is dynamically adjusted according to multiple input variables, and fuzzy rules are set to adjust CR to adapt it to the diversity of the population. Defuzzification is used to make the dynamic adjustment of CR adapt to the operation process of the algorithm, thereby increasing the robustness and effect of the algorithm. The formulas of the defuzzified CR value and the crossover operation are as follows:

[0039]

[0040]

[0041] in, is the membership degree of the i-th individual in the fuzzy boundary, is the fuzzy output, where barycentric defuzzification is necessary, D is the total number of dimensions, rand(0,1) is a random number uniformly generated in the range (0,1), and j rand [1,D] ensures that at least one dimension comes from the mutation vector to ensure that the test individual and the target individual are not exactly the same, and finally the test individual u is obtained. i,j (g+1);

[0042] Step S3.4: Construct a fitness function. The introduced candidate features provide information related to the category, but also introduce redundant information between the features. An advanced fitness function is designed to evaluate the adaptability of the fuzzy boundary. The function is expressed as follows:

[0043]

[0044] Where X represents the subset of selected features, n is the length of the selected feature subset, λ represents the variable coefficient, and H(x) represents the calculation of the mutual information of x; FNCMI δ (a i ; C) is used to evaluate the amount of information each feature in the selected subset can provide about the category in the presence of other features;

[0045] Step S3.5: During the selection process, individuals are selected from the experimental population u according to the greedy algorithm criterion to form the next generation population x i (g+1), the specific selection method is expressed as follows:

[0046]

[0047] Among them, fit(x) represents the fitness of individual x. This whole process will be repeated in each generation until the shutdown condition is met, that is, the maximum number of iterations or fitness standard is reached. Through this selection mechanism, the module explores the solution space and strikes a balance between global search and local search, so as to find the optimal subset of data.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. In the fuzzy neighborhood differential evolution stage, the present invention uses fuzzy neighborhood to dynamically adjust the mutation factor and crossover probability, increasing the algorithm's ability to adapt to complex search spaces. The fuzzy neighborhood decision tree module is combined to further screen and optimize features to evaluate the importance of features, and then the performance of the decision tree is improved through pruning and optimization processes, thereby improving the quality of the final feature subset. The two-stage iterative process ensures continuous optimization before the stop condition is met, which can improve the accuracy of feature selection for schizophrenia disease data.

[0050] 2. The present invention also solves the problems of insufficient global search capability and low selection efficiency in high-dimensional data feature selection, and proposes a two-stage integration method based on fuzzy neighborhood differential evolution and fuzzy neighborhood decision tree modules. The present invention aims to improve the performance of traditional differential evolution and decision trees in feature selection performance, and optimize the initial input processing of high-dimensional data. By introducing the fuzzy neighborhood strategy, the present invention effectively improves the generalization ability of the model, especially showing a more efficient application value in the feature selection of schizophrenia-related data. This strategy not only enables the model to adapt to diverse data features more flexibly, but also shows significant advantages in processing redundant and noisy data. Through the collaborative work of the two-stage integration method, the system can quickly locate key features with less computational overhead. In addition, by optimizing input processing and feature screening, the present invention further promotes breakthrough progress in a big data environment, providing strong support for clinical data analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention but do not constitute a limitation of the present invention.

[0052] Figure 1 The flowchart of the two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection of the present invention.

[0053] Figure 2 This is an overall framework diagram of the two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection of the present invention.

[0054] Figure 3 This is a sample illustration of the two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection of the present invention.

[0055] Figure 4 It is a comparative convergence diagram of the two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection of the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. Of course, the specific embodiments described here are only used to explain the present invention and are not used to limit the present invention.

[0057] Example 1

[0058] See also Figures 1 to 3The technical solution provided in this embodiment is a two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection. Taking the first group of samples {x0, x1, x2, ..., x8} in the data set as an example, from loading data to obtaining feature selection results, the following steps are included:

[0059] S1: Acquisition and preprocessing of schizophrenia disease data: First, a 3T General Electric MRI scanner was used to acquire imaging data through resting-state functional magnetic resonance imaging (rs-fMRI). In the data preprocessing stage, SPM8 software was used to stabilize, correct, and filter the acquired data. The collected images were segmented into multiple regions using the AAL module, and average time series data were extracted from each region to obtain feature information. After data processing, a dataset containing multiple samples and features was constructed;

[0060] S2: Constructing fuzzy neighborhood decision space and fuzzy neighborhood mutual information: Construct a fuzzy neighborhood decision space FNDS in the feature space<U,C∪D,V,f> ,U={x1,x2,x3,...,x N} is the domain, C and D are the conditional attribute set and decision attribute set respectively, V is the value range of all attributes, and f is the related information function. This space not only considers the distance between data samples, but also considers the fuzzy relationship between them, thus more accurately reflecting the heterogeneity of schizophrenia disease sample information. Fuzzy probability distribution is used to calculate the fuzzy neighborhood mutual information FNMI δ (S1; S2), where S1 and S2 are subsets of attributes in F. Fuzzy neighborhood mutual information not only considers the direct correlation between features, but also the fuzzy relationship between features;

[0061] S3: Design the first stage: schizophrenia disease feature selection module based on fuzzy neighborhood differential evolution, improve fuzzy neighborhood and define a new dynamic neighborhood radius calculation method FNB xi , x i is the midpoint of the fuzzy boundary. In the early stage of the algorithm, the neighborhood range is wider to enhance the search ability, while in the later stage, the neighborhood range is narrowed to refine the boundary area. Then, a new mutation strategy v for the offspring g is proposed by combining the fuzzy boundary and the mutation factor. i (g), dynamically adjust the weight coefficient to adapt to different data dimensions and spaces. Finally, set fuzzy rules to dynamically adjust the crossover factor CR to increase the robustness and effect of the algorithm, so as to better adapt to the population diversity and algorithm operation process. The final crossover factor is obtained through defuzzification to ensure the effectiveness of mutation and crossover operations;

[0062] S4: The second stage of the design: The schizophrenia disease feature selection module of the fuzzy neighborhood decision tree divides the data set by selecting attributes with high consistency with the class label to generate a more concise and more generalization-capable decision tree. During the construction process, first check the number of samples and the number of classes at the current node. If there is only one sample or class in the node, stop growing. Otherwise, start node division: Calculate the fuzzy neighborhood mutual information when Ai = cj for each possible value cj of each feature Ai. Among all attributes and all split points, select the attribute with the largest FNMI value and its corresponding split point as the optimal attribute A and the optimal split point c*. If the fuzzy neighborhood complementary mutual information FNMI(A) of attribute A for the data set is less than the threshold condition, the decision tree stops growing; otherwise, split the samples at this node into two child nodes according to the optimal attribute and the optimal split point. According to the comparison between FNMI(A) and the average value, if the former is lower than the average value, iteratively prune and optimize the decision tree, otherwise, end the pruning optimization process to obtain the finally optimized fuzzy neighborhood mutual information decision tree FNMIDT;

[0063] S5: The two stages are fused to form a complete feature selection process: Obtain the preliminary screened schizophrenia disease feature subset {x1, x3, x6, x7} in the first stage and enter the model-based feature selection process in the second stage to get the result {x1, x6}. Use the selected feature set to construct the final schizophrenia detection model for classification and prediction of clinical data.

[0064] Specifically, the specific steps of step S2 are as follows:

[0065] Step S2.1: Construct a fuzzy neighborhood decision system FNDS = <U, C ∪ D, V, f>, where U = {x1, x2, x3,..., x N} is the universe of discourse among them, C and D are the conditional attribute set and the decision attribute set respectively, V is the value range on all attributes, f is the relevant information function, and B is the attribute subset in the universe of discourse U.

[0066] Step S2.2: Construct the fuzzy similarity relation and the fuzzy similarity matrix. In the fuzzy neighborhood decision system, R a is the fuzzy similarity relation derived from feature a. For the fuzzy similarity matrix [x] a (y) = R a (x, y), then the fuzzy similarity matrix of x on U with respect to B is

[0067]

[0068] Among them, according to the samples {x0, x1, x2,..., x8}, the values of the similarity matrix are obtained as

[0069] Step S2.3: Construct parameterized fuzzy neighborhood information granularity, where the information granularity can be adjusted by parameters. Fuzzy Neighborhood Decision System FNDS =<U,C∪D,V,f> ,for Construct parameterized fuzzy neighborhood information granularity δ B (x), where we can get δ B The value of )x) is 0.5, which can be used to control the size of the fuzzy neighborhood, thereby affecting the coarseness of the information granularity;

[0070] Step S2.4: Calculate the fuzzy neighborhood mutual information using the fuzzy probability distribution. This set of samples has numerical or discrete features {x0, x1, x2, ..., x8}, where x i is a subset of the attributes in F. FNMI can be calculated according to formula (4): δ The value of (x1; x2) is 0.375, and the other fuzzy neighborhood mutual information is calculated later.

[0071] Then the fuzzy neighborhood conditional mutual information is derived. This set of samples has numerical or discrete features {x0,x1,x2,...,x8}, where x i For a subset of the attributes, FNCMI can be calculated according to formula (5): δ The value of (D; x1) is 0.782, and other fuzzy neighborhood conditional mutual information is calculated subsequently.

[0072] Among them, fuzzy neighborhood mutual information reflects the information relationship between attribute subsets. And fuzzy neighborhood conditional mutual information FNCMI δ (D; x i ) reflects the attribute subset x i The relationship between D and S is the fuzzy neighborhood conditional mutual information, which indicates a stronger correlation.

[0073] As the two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection provided by this embodiment, the specific steps of step S3 are as follows:

[0074] Step S3.1: The fuzzy neighborhood differential evolution module uses the fuzzy boundary to select the uncertainty of data and form an uncertain data search band. Since the membership function is defined as Among them, x i The value of is 0.25. The improved fuzzy neighborhood adopts a new calculation method of the neighborhood radius δ. According to formulas (7) and (8), its dynamic change results are obtained. f is a constant parameter that controls the fuzzy boundary. Here, the fixed value is 0.75. F is 3. pol and Maxpol are 4 and 9 respectively. Fis the point closest to the center x in the fuzzy boundary i The Fth individual of

[0075] Step S3.2: Select a larger neighborhood range in the early stage to improve the search ability, and select a smaller neighborhood range in the later stage to find the optimal edge area. In this way, the final fuzzy boundary is obtained according to formula (9): After that, the fuzzy boundary is combined with the variation factor, where W FN The value of is 0.352, the value of F' is 0.2, and the value of x FNB (g) is the center point individual within the fuzzy boundary, and To obtain the other two individuals within the fuzzy boundary. Because the weight coefficient W FN The range is [0,1], so when W FN As F' decreases as it approaches 0, FN When it is close to 1, F' will increase. This combination can adjust the mutation factor more flexibly to adapt to different data dimensions and spaces.

[0076] Step S3.3: During the crossover process, the crossover factor CR may be dynamically adjusted according to multiple input variables. Therefore, fuzzy rules are set to adjust CR to adapt to the diversity of the population. According to formula (11), is the membership degree of the i-th individual in the fuzzy boundary, is the fuzzy output, and the final CR value is 0.8. D is the total number of dimensions, rand(0,1) is a random number uniformly generated in the range (0,1), and j rand [1,D] ensures that at least one dimension comes from the mutation vector to ensure that the test individual and the target individual are not exactly the same;

[0077] Step S3.4: construct fitness function Fitness(X), wherein the values ​​of each fitness function are calculated respectively: Fitness(x1)=0.8712, Fitness(x2)=0.3451, Fitness(x3)=0.7532, Fitness(x4)=0.7644, Fitness(x5)=0.8521, Fitness(x6)=0.2896, Fitness(x7)=0.6731; FNCMI δ (a i ; C) is used to evaluate the amount of information each feature in the selected subset can provide about the category in the presence of other features;

[0078] Step S3.5: During the selection process, individuals are selected from the experimental population u according to the greedy algorithm criterion to form the next generation population. The specific selection method is expressed as follows:

[0079]

[0080] Among them, according to the value of the fitness function obtained in step S3.4, a comparative search is performed to obtain the optimal subset of data {x1, x3, x6, x7}.

[0081] Example 2

[0082] See also Figure 4 The technical solution provided in this embodiment is a two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection, which is compared with the convergence and classification accuracy of the traditional method on a set of schizophrenia disease data sets, including the following steps:

[0083] Convergence analysis is performed here to compare the performance of the two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection with traditional algorithms (Raw, DEFS, MDEFS, FCDE, MOFS-BDE and ABCoDT) on the dataset. The maximum number of iterations is set to 1000 and the population size is set to 50 to evaluate and compare the convergence speed and fitness value of the method. Figure 4 The convergence graphs of various algorithms on the schizophrenia dataset are shown. As can be seen from the graph, the two-stage fuzzy neighborhood feature selection method can obtain the best fitness after more than 200 iterations, which shows that the method has a fast convergence speed and high efficiency.

[0084] After that, the classification accuracy was compared. The experimental results shown in the table below show the classification accuracy (i.e., mean and standard deviation) of different algorithms on the schizophrenia dataset for KNN, SVM, and RF classifiers. In this analysis, the algorithms compared include Raw, DEFS, MDEFS, FCDE, MOFS-BDE, ABCoDT, and the two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection. In the table, the data corresponding to the highest classification accuracy are shown in bold.

[0085] Table 1 Classification accuracy of different algorithms on KNN, SVM and RF classifiers (i.e., mean and standard deviation)

[0086]

[0087] As can be seen from Table 1, the two-stage fuzzy neighborhood feature selection method always shows the highest and second highest classification accuracy on the schizophrenia disease dataset compared with other traditional methods. In addition, the two-stage fuzzy neighborhood feature selection method for schizophrenia disease also has the lowest standard deviation, indicating that its performance is stable and reliable. These results all prove that the present invention shows a more efficient application value in feature selection of schizophrenia-related data.

[0088] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection, characterized in that: The following steps are involved: S1: Acquisition and preprocessing of schizophrenia disease data: First, a 3T General Electric MRI scanner was used to acquire imaging data through resting-state functional magnetic resonance imaging (rs-fMRI). In the data preprocessing stage, SPM8 software was used to stabilize, correct and filter the acquired data. The collected images were segmented into multiple regions using the AAL module, and the average time series data were extracted from each region to obtain feature information. After data processing, a dataset containing multiple samples and features was constructed. S2: Constructing fuzzy neighborhood decision space and fuzzy neighborhood mutual information: Construct a fuzzy neighborhood decision space FNDS in the feature space<U,C∪D,V,f> ,U={x1,x2,x3,...,x N } is the domain, C and D are the conditional attribute set and decision attribute set respectively, V is the value range of all attributes, and f is the related information function. This space not only considers the distance between data samples, but also considers the fuzzy relationship between them, thus more accurately reflecting the heterogeneity of schizophrenia disease sample information. Fuzzy probability distribution is used to calculate the fuzzy neighborhood mutual information FNMI δ (S1; S2), where S1 and S2 are subsets of attributes in F. Fuzzy neighborhood mutual information not only considers the direct correlation between features, but also the fuzzy relationship between features; S3: Design the first stage: schizophrenia disease feature selection module based on fuzzy neighborhood differential evolution, improve fuzzy neighborhood and define a new dynamic neighborhood radius calculation method FNB xi , x i is the midpoint of the fuzzy boundary. In the early stage of the algorithm, the neighborhood range is wider to enhance the search ability, while in the later stage, the neighborhood range is narrowed to refine the boundary area. Then, a new mutation strategy v for the offspring g is proposed by combining the fuzzy boundary and the mutation factor. i (g), dynamically adjust the weight coefficient to adapt to different data dimensions and spaces. Finally, set fuzzy rules to dynamically adjust the crossover factor CR to increase the robustness and effect of the algorithm, so as to better adapt to the population diversity and algorithm operation process. The final crossover factor is obtained through defuzzification to ensure the effectiveness of mutation and crossover operations; S4: Design the second stage: the schizophrenia disease feature selection module of the fuzzy neighborhood decision tree. The data set is divided by selecting attributes with high consistency with the class label to generate a decision tree. During the construction process, the number of samples and the number of categories of the current node are first checked. If there is only one sample or category in the node, the growth is stopped, otherwise, the node division is started: for each possible value cj of each feature Ai, the fuzzy neighborhood mutual information when Ai=cj is calculated. Among all attributes and all split points, the attribute with the largest FNMI value and its corresponding split point are selected as the optimal attribute A and the optimal split point c*. If the fuzzy neighborhood complementary mutual information FNMI(A) of attribute A to the data set is less than the threshold condition, the decision tree stops growing; otherwise, the samples of the node are split into two child nodes according to the optimal attribute and the optimal split point. According to the comparison between FNMI(A) and the average value, if the former is lower than the average value, the decision tree is pruned and optimized through iteration, otherwise, the pruning optimization process is terminated to obtain the final optimized fuzzy neighborhood mutual information decision tree FNMIDT; S5: The two stages are integrated to form a complete feature selection process: in the first stage, a subset of schizophrenia disease features is obtained for preliminary screening, and the model-based feature selection process is entered in the second stage. In the second stage, feature selection is continuously optimized through modeling feedback, and verification is used to ensure the rationality of feature selection. The selected feature set is used to build the final schizophrenia detection model for classification and prediction of clinical data.

2. The two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection according to claim 1, characterized in that: The specific steps of step S2 are as follows: Step S2.1: Construct a FNDS= according to the data information<U,C∪D,V,f> Fuzzy neighborhood decision system, U={x1,x2,x3,...,x N } is the domain, C and D are the condition attribute set and decision attribute set respectively, V is the value range of all attributes, f is the related information function, B is the attribute subset in the domain U, Step S2.2: Construct fuzzy similarity relationship and fuzzy similarity matrix, in the fuzzy neighborhood decision system FNDS =<U,C∪D,V,f> , R a is the fuzzy similarity relationship derived from feature a, and is defined as for Fuzzy similarity matrix[x] a (y) = R a (x,y), then the fuzzy similarity matrix of x on U with respect to B is defined as Among them, R B is a fuzzy similarity relationship with a value range between [0,1]. The closer the value is to 1, the more similar the samples are. The fuzzy similarity matrix [x] a (y) is an n×n matrix, where n is the number of samples; Step S2.3: Construct parameterized fuzzy neighborhood information granularity, where the information granularity is adjusted by parameters. This method is usually used to deal with uncertainty and ambiguity, and better capture and utilize information in data analysis and decision-making. Fuzzy neighborhood decision system FNDS =<U,C∪D,V,f> ,for The granularity of parameterized fuzzy neighborhood information is constructed as follows: Among them, δ represents the fuzzy neighborhood radius, 0≤δ≤1, through the fuzzy similarity relationship R B and the neighborhood radius δ The fuzzy neighborhood particles control the size of the fuzzy neighborhood, thus affecting the coarseness and fineness of the information granularity; Step S2.4: Using the fuzzy probability distribution, calculate the fuzzy neighborhood mutual information, FNDS =<U,C∪D,V,f> is a fuzzy neighborhood decision system, U={x1,x2,x3,...,x N } is the domain of discourse, this set of samples has numerical or discrete features F, where S1 and S2 are subsets of attributes in F, and sample x1 is expressed in the feature subspace S1∪S2 as Then the fuzzy neighborhood mutual information is expressed as: Then the fuzzy neighborhood conditional mutual information is derived. This group of samples has a numerical or discrete feature F, where S is a subset of the attributes in F, and D is the decision attribute. Then the fuzzy neighborhood conditional mutual information of S for D is expressed as follows: Among them, fuzzy neighborhood mutual information FNMI δ (S1; S2) reflects the information relationship between attribute subsets S1 and S2, N represents the variable parameters, and the fuzzy neighborhood conditional mutual information FNCMI δ (D; S) reflects the relationship between the attribute subset S and the decision attribute D. The larger the fuzzy neighborhood conditional mutual information between D and S, the stronger the correlation between S and D.

3. The two-stage fuzzy neighborhood feature selection method for schizophrenia disease detection according to claim 1, characterized in that: The specific steps of step S3 are as follows: Step S3.1: The fuzzy neighborhood differential evolution module uses the fuzzy boundary to select the uncertainty of data and form an uncertain data search band, where the membership function is defined as: Among them, x i is the midpoint of the fuzzy boundary, x j For all individuals, find the Euclidean distance between them, d is the radius of the fuzzy boundary. Different from the original fuzzy neighborhood, the improved fuzzy neighborhood adopts a new calculation method of the neighborhood radius δ, and its dynamic changes are as follows: Among them, f is a constant parameter controlling the fuzzy boundary, F is the number of individuals in the fuzzy boundary, pol and Maxpol are the current population and the maximum population respectively, and x F is the point closest to the center x in the fuzzy boundary i The Fth individual of Step S3.2: Select a larger neighborhood range in the early stage to improve the search ability, and select a smaller neighborhood range in the later stage to find the optimal edge area, select individuals according to the membership degree, and form the final neighborhood. The method is as follows: in By x i The final neighborhood boundary determined, rand j is a random value in the jth dimension [0,1], NP is the total number, and then a new mutation strategy is proposed by combining the fuzzy boundary with the mutation factor, as shown below: Among them, W FN is the weight coefficient of the fuzzy neighborhood, F' is the new variation factor combined with the fuzzy neighborhood weight, x FNB (g) is the center point individual within the fuzzy boundary, and To obtain two more individuals within the fuzzy boundary, the weight coefficient W FN The range is [0,1], so when W FN As F' decreases as it approaches 0, FN When it is close to 1, F' will increase, and finally the target individual v i (g); Step S3.3: During the crossover process, the crossover factor CR is dynamically adjusted according to multiple input variables, and fuzzy rules are set to adjust CR to adapt it to the diversity of the population. Defuzzification is used to make the dynamic adjustment of CR adapt to the operation process of the algorithm, thereby increasing the robustness and effect of the algorithm. The formulas of the defuzzified CR value and the crossover operation are as follows: in, is the membership degree of the i-th individual in the fuzzy boundary, is the fuzzy output, where barycentric defuzzification is necessary, D is the total number of dimensions, rand(0,1) is a random number uniformly generated in the range (0,1), and j rand [1,D] ensures that at least one dimension comes from the mutation vector to ensure that the test individual and the target individual are not exactly the same, and finally the test individual u is obtained. i,j (g+1); Step S3.4: Construct a fitness function. The introduced candidate features provide information related to the category, but also introduce redundant information between the features. An advanced fitness function is designed to evaluate the adaptability of the fuzzy boundary. The function is expressed as follows: Where X represents the subset of selected features, n is the length of the selected feature subset, λ represents the variable coefficient, and H(x) represents the calculation of the mutual information of x; FNCMI δ (a i ; C) is used to evaluate the amount of information each feature in the selected subset can provide about the category in the presence of other features; Step S3.5: During the selection process, individuals are selected from the experimental population u according to the greedy algorithm criterion to form the next generation population x i (g+1), the specific selection method is expressed as follows: Among them, fit(x) represents the fitness of individual x. This whole process will be repeated in each generation until the shutdown condition is met, that is, the maximum number of iterations or fitness standard is reached. Through this selection mechanism, the module explores the solution space and strikes a balance between global search and local search, so as to find the optimal subset of data.

Images