An alzheimer's disease detection method and device based on an improved arithmetic optimization algorithm, an electronic device, and a storage medium

Abstract

The application provides an Alzheimer's disease detection method and device based on an improved arithmetic optimization algorithm, an electronic device and a storage medium. The method comprises the following steps: S1, acquiring a multi-modal image of a target object, wherein the multi-modal image comprises an sMRI image, an A beta-PET image and a tau-PET image; S2, performing feature extraction on the multi-modal image to obtain multi-modal features; S3, acquiring a target feature subset constructed in advance based on the improved arithmetic optimization algorithm, performing feature selection on the multi-modal features based on the target feature subset, and obtaining a feature selection result; and S4, classifying the feature selection result by using a fuzzy support vector machine classifier to obtain an Alzheimer's disease detection result. The improved arithmetic optimization algorithm improves the global search efficiency and exploration-development balance ability of the original arithmetic optimization algorithm through a vertical-horizontal crossing strategy and a whale falling strategy.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical image analysis technology, specifically to an Alzheimer's disease detection method, device, electronic device, and storage medium based on an improved arithmetic optimization algorithm. Background Technology

[0002] Alzheimer's disease (AD) is the most common type of dementia among the elderly, and its incidence continues to rise with population aging. The core pathological mechanisms of AD involve abnormal deposition of β-amyloid (Aβ) forming extracellular plaques, hyperphosphorylation of tau protein forming intracellular neurofibrillary tangles, and the subsequent loss of neurons and brain atrophy. These pathological changes can begin years or even decades before the onset of clinical symptoms, a stage known as the preclinical phase, providing a crucial window for early intervention.

[0003] Current diagnostic methods for Alzheimer's disease include cognitive function assessment, cerebrospinal fluid / blood biomarker testing, neuroimaging examinations, and genetic risk assessment. Among these, neuroimaging technology has become the most commonly used early identification method due to its maturity and clinical operability. Structural magnetic resonance imaging (sMRI) can clearly display brain anatomy and accurately assess the degree of atrophy; positron emission tomography (PET) is highly sensitive to metabolic and molecular changes and can detect Aβ and tau pathology before the appearance of clinical symptoms of cognitive decline. However, single-modality imaging has the problems of incomplete information and insufficient early sensitivity: sMRI has limited identification ability in the stage of mild cognitive impairment (MCI), while PET, although sensitive, is limited by equipment and tracers.

[0004] In recent years, machine learning and deep learning have developed in parallel in the assisted diagnosis of Alzheimer's disease, but their performance is highly dependent on the sample size and data distribution. In scenarios with small samples and high-dimensional medical images, traditional support vector machines (SVMs) are susceptible to noise interference and get stuck in local optima; while deep learning models have great potential, they face challenges such as unstable training, poor generalization ability, and insufficient interpretability. Multimodal fusion (sMRI+Aβ-PET+tau-PET) can provide complementary information, but due to high-dimensional feature redundancy, modal heterogeneity, and noise problems, traditional feature selection methods (such as principal component analysis and manual screening) are inefficient and have poor interpretability. Arithmetic optimization algorithms (AOA) can assist in feature selection, but they suffer from premature convergence and an imbalance between exploration and utilization capabilities.

[0005] Therefore, there is an urgent need for an efficient, robust, and clinically interpretable multimodal integrated framework that can simultaneously achieve high-dimensional feature optimization and noise robust classification, while mining complementary biomarkers between modalities to reflect the natural laws of asynchronous progression of Alzheimer's pathology (early Aβ, late tau, and the entire atrophy process), so as to significantly improve the accuracy of early diagnosis, generalization ability, and clinical application value. Summary of the Invention

[0006] In view of this, to overcome the problems of low accuracy in single-modal diagnosis, low efficiency in multimodal high-dimensional feature selection, model sensitivity to noise, and insufficient interpretability, this application proposes an Alzheimer's disease detection method, device, electronic device, and storage medium based on an improved arithmetic optimization algorithm. Specifically, this application is implemented through the following technical solution:

[0007] According to a first aspect of the embodiments of this specification, an Alzheimer's disease detection method based on an improved arithmetic optimization algorithm is provided, the method comprising the following steps:

[0008] Step S1: Acquire multimodal images of the target object, including sMRI images, Aβ-PET images, and tau-PET images;

[0009] Step S2: Extract features from the multimodal image to obtain multimodal features;

[0010] Step S3: Obtain a target feature subset pre-constructed based on an improved arithmetic optimization algorithm, and perform feature selection on the multimodal features based on the target feature subset to obtain the feature selection result;

[0011] Step S4: Use a fuzzy support vector machine classifier to classify the feature selection results to obtain the Alzheimer's disease detection results;

[0012] The process of constructing the target feature subset based on the improved arithmetic optimization algorithm is as follows:

[0013] Step A: Construct a sample set, which includes multiple sets of multimodal samples and their corresponding category labels. The sample set is divided into a training set and a test set.

[0014] Step B: Initialize the population matrix of the improved arithmetic optimization algorithm. Each row vector of the population matrix represents an individual, and each individual corresponds to a subset of candidate features. The dimension of the row vector is equal to the total number of features of the multimodal samples, and the value of each dimension is a continuous search variable. This continuous search variable, after being masked, indicates the corresponding feature selection state of the training samples.

[0015] Step C involves iteratively updating each dimension value in the population matrix through the following sub-steps until the maximum number of iterations is reached or the fitness value converges:

[0016] C1, using an exploratory update strategy or an exploitative update strategy with an arithmetic optimization algorithm to update the current values ​​of each dimension, and obtain the first update result of the population matrix;

[0017] C2 employs a cross-sectional strategy to perform horizontal and vertical cross-sectional calculations on each individual in the first update result of the population matrix. The horizontal cross-sectional strategy weights and fuses the dimension values ​​of two different individuals in the same dimension, then perturbs the differences to generate a first temporary candidate feature subset for each individual. If the fitness value corresponding to the first temporary candidate feature subset of each individual is less than the fitness value of the original candidate feature subset before the horizontal cross-sectional strategy, then the candidate feature subset of that individual in the first update result is updated to a first temporary candidate feature subset, resulting in the second update result of the population matrix. The vertical cross-sectional strategy randomly weights and fuses the dimension values ​​of the same individual in different dimensions to generate a second temporary candidate feature subset for that individual. If the fitness value corresponding to the second temporary candidate feature subset of that individual is less than the fitness value of the candidate feature subset in the second update result, then the candidate feature subset of that individual in the second update result is updated to a second temporary candidate feature subset, resulting in the cross-sectional update result of the population matrix.

[0018] C3, when the whale fall triggering condition is met in the current iteration round, the current individual is perturbed and updated based on the current individual in the cross-up update result, other randomly selected individuals, and the decay step size corresponding to the current round, to obtain a new candidate feature subset after perturbed update. If the fitness value of the new candidate feature subset is less than the fitness value before perturbed update, the candidate feature subset of the individual in the cross-up update result is updated to the new candidate feature subset, and the population matrix after this iteration is obtained.

[0019] Step D: After the iteration is completed, the individual with the smallest fitness value is determined from the final population matrix, and the candidate feature subset of this individual is taken as the target feature subset.

[0020] According to a second aspect of the embodiments of this specification, an Alzheimer's disease detection device based on an improved arithmetic optimization algorithm is provided, the device comprising:

[0021] An image acquisition unit is used to acquire multimodal images of a target object, including sMRI images, Aβ-PET images, and tau-PET images;

[0022] The feature extraction unit is used to extract features from the multimodal image to obtain multimodal features;

[0023] The feature selection unit is used to obtain a target feature subset pre-constructed based on an improved arithmetic optimization algorithm, and to perform feature selection on the multimodal features based on the target feature subset to obtain the feature selection result;

[0024] The classification and detection unit is used to classify the feature selection results using a fuzzy support vector machine classifier to obtain Alzheimer's disease detection results;

[0025] The process of constructing the target feature subset based on the improved arithmetic optimization algorithm is as follows:

[0026] Step A: Construct a sample set, which includes multiple sets of multimodal samples and their corresponding category labels. The sample set is divided into a training set and a test set.

[0027] Step B: Initialize the population matrix of the improved arithmetic optimization algorithm. Each row vector of the population matrix represents an individual, and each individual corresponds to a subset of candidate features. The dimension of the row vector is equal to the total number of features of the multimodal samples, and the value of each dimension is a continuous search variable. This continuous search variable, after being masked, indicates the corresponding feature selection state of the training samples.

[0028] Step C involves iteratively updating each dimension value in the population matrix through the following sub-steps until the maximum number of iterations is reached or the fitness value converges:

[0029] C1, using an exploratory update strategy or an exploitative update strategy with an arithmetic optimization algorithm to update the current values ​​of each dimension, and obtain the first update result of the population matrix;

[0030] C2 employs a cross-sectional strategy to perform horizontal and vertical cross-sectional calculations on each individual in the first update result of the population matrix. The horizontal cross-sectional strategy weights and fuses the dimension values ​​of two different individuals in the same dimension, then perturbs the differences to generate a first temporary candidate feature subset for each individual. If the fitness value corresponding to the first temporary candidate feature subset of each individual is less than the fitness value of the original candidate feature subset before the horizontal cross-sectional strategy, then the candidate feature subset of that individual in the first update result is updated to a first temporary candidate feature subset, resulting in the second update result of the population matrix. The vertical cross-sectional strategy randomly weights and fuses the dimension values ​​of the same individual in different dimensions to generate a second temporary candidate feature subset for that individual. If the fitness value corresponding to the second temporary candidate feature subset of that individual is less than the fitness value of the candidate feature subset in the second update result, then the candidate feature subset of that individual in the second update result is updated to a second temporary candidate feature subset, resulting in the cross-sectional update result of the population matrix.

[0031] C3, when the whale fall triggering condition is met in the current iteration round, the current individual is perturbed and updated based on the current individual in the cross-up update result, other randomly selected individuals, and the decay step size corresponding to the current round, to obtain a new candidate feature subset after perturbed update. If the fitness value of the new candidate feature subset is less than the fitness value before perturbed update, the candidate feature subset of the individual in the cross-up update result is updated to the new candidate feature subset, and the population matrix after this iteration is obtained.

[0032] Step D: After the iteration is completed, the individual with the smallest fitness value is determined from the final population matrix, and the candidate feature subset of this individual is taken as the target feature subset.

[0033] According to a third aspect of the embodiments of this specification, an electronic device is provided, comprising: a processor; and a computer-readable storage medium storing computer program instructions that, when executed by the processor, cause the processor to perform the method as described in the first aspect.

[0034] According to a fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided having a computer program stored thereon, the computer program being executed by a processor as described in the first aspect.

[0035] The embodiments of this application have at least the following technical effects:

[0036] First, the embodiments of this application propose an improved arithmetic optimization algorithm, which enhances the global search efficiency and exploration-exploitation balance of the original arithmetic optimization algorithm through cross-sectional and cross-sectional strategies and whale fall strategies, providing an efficient tool for selecting high-dimensional medical features in small samples;

[0037] Second, the embodiments of this application deeply integrate the improved arithmetic optimization algorithm with the fuzzy support vector machine (FSVM) classifier to achieve the integration of feature selection and classification. Based on fuzzy membership, FSVM can effectively reduce the influence of medical image noise and outliers, and enhance the robustness and generalization ability of the classifier.

[0038] Third, in the detection of Alzheimer's disease, the embodiments of this application verify modal complementarity based on multimodal images. For example, Aβ-PET images dominate the detection of early mild cognitive impairment and normal cognition, tau-PET images dominate late neurodegenerative changes, and sMRI images provide evidence of stable structures at all stages, which is consistent with the biological law of asynchronous progression of Alzheimer's pathology. The extracted multimodal features, such as bilateral hippocampus, amygdala, parahippocampal gyrus, and posterior cingulate cortex, are highly consistent with the classic pathological model, significantly improving clinical interpretability and the potential for biomarker discovery.

[0039] Fourth, the classification accuracy of this application's embodiment, based on 154 multimodal data from the ADNI database, reached 96.8% under trimodal fusion, which is superior to single-modal, traditional AOA-SVM, and other mainstream detection methods. At the same time, it provides visual interpretation through heatmaps and Top 10 feature analysis, demonstrating good reproducibility and clinical translation prospects. Attached Figure Description

[0040] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Some specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings in an exemplary and non-limiting manner. The same reference numerals in the drawings indicate the same or similar parts or components. Those skilled in the art should understand that these drawings are not necessarily drawn to scale. In the drawings:

[0041] Figure 1 This is a flowchart illustrating an Alzheimer's disease detection method based on an improved arithmetic optimization algorithm, as shown in an exemplary embodiment of this application.

[0042] Figure 2 This is a schematic diagram illustrating a detailed detection process for Alzheimer's disease, as shown in an exemplary embodiment of this application.

[0043] Figure 3 This is a flowchart illustrating an improved arithmetic optimization algorithm according to an exemplary embodiment of this application;

[0044] Figure 4 This is an exemplary embodiment of the present application illustrating a heatmap of the importance of a characteristic brain region;

[0045] Figure 5 This is a schematic diagram illustrating a performance comparison between an improved arithmetic optimization algorithm (EAOA-WFCS) and a mainstream metaheuristic algorithm, as shown in an exemplary embodiment of this application.

[0046] Figure 6 This is a block diagram illustrating an electronic device according to an exemplary embodiment of this application;

[0047] Figure 7 This is a block diagram of an Alzheimer's disease detection device based on an improved arithmetic optimization algorithm, as illustrated in an exemplary embodiment of this application. Detailed Implementation

[0048] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0049] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0050] To overcome the problems of low accuracy in single-modal diagnosis, low efficiency in multimodal high-dimensional feature selection, model sensitivity to noise, and insufficient interpretability, this application provides a fuzzy support vector machine diagnostic framework (EAOA-WFCS-FSVM) driven by a feature selection optimization algorithm that integrates whale fall and cross-sectional strategies. This method addresses the challenge of high-dimensional heterogeneous data fusion in the neuroimaging diagnosis of Alzheimer's disease by significantly improving upon traditional arithmetic optimization algorithms (AOA) and support vector machines (SVM). Through improvements to the arithmetic optimization algorithm and the fuzzy membership mechanism of the support vector machine, it achieves complementary fusion of multimodal biomarkers (sMRI, Aβ-PET, tau-PET), automatic extraction of sensitive features, and robust classification, thereby significantly improving the accuracy of early Alzheimer's disease diagnosis, model generalization ability, and clinical interpretability.

[0051] The embodiments described in this specification will now be described in detail.

[0052] Figure 1 This is a flowchart illustrating an exemplary embodiment of an Alzheimer's disease detection method based on an improved arithmetic optimization algorithm, as shown in this application. Figure 1 As shown, the Alzheimer's disease detection method includes the following steps:

[0053] Step S1: Obtain multimodal images of the target object, including sMRI images, Aβ-PET images, and tau-PET images.

[0054] The target subjects were patients undergoing testing. Head scans were performed using MRI equipment conforming to clinical standards. Structural magnetic resonance imaging (sMRI) employed T1-weighted three-dimensional magnetization to prepare fast gradient echo sequences, acquiring high-resolution structural images of the entire brain. Aβ-Positron Emission Tomography (Aβ-PET) and tau positron emission tomography (tau-PET) were dynamically acquired using specific tracers to obtain molecular images reflecting amyloid deposition and tau tangles distribution.

[0055] The acquired raw multimodal images underwent standardized preprocessing, including format conversion and AC-PC correction. For example, the raw images were uniformly converted to NIFTI format. DICOM format medical images were processed using MRICroGL software, and HRRT format images were processed using DeBabeler software. After format conversion, all images underwent anterior commissure (AC)-posterior commissure (PC) correction to eliminate the influence of head tilt on subsequent analysis.

[0056] Then, the sMRI images were preprocessed using the CAT12 toolbox in the MATLAB platform. For example, image quality was checked first to remove images with severe motion artifacts or imaging noise; then, the image space was normalized to the Montreal Neurological Institute (MNI) standard space; finally, tissue segmentation was performed to divide the brain tissue into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF).

[0057] Finally, for Aβ-PET and tau-PET images, the Coregister operation in the SPM12 toolbox is first used to register the PET image with the sMRI image of the target object to ensure spatial alignment of the two modalities; then, the deformation field parameters generated during the spatial normalization process of the sMRI image are used to normalize the PET image to the MNI space.

[0058] After completing the above preprocessing, standardized sMRI, Aβ-PET, and tau-PET images that can be used for subsequent feature extraction are obtained.

[0059] Step S2: Extract features from the multimodal image to obtain multimodal features.

[0060] The multimodal features include gray matter volume features based on sMRI images, normalized uptake ratio features based on Aβ-PET images, and normalized uptake ratio features based on tau-PET images.

[0061] The extraction process of gray matter volume features based on sMRI images is as follows: Automated Anatomical Labeling 3 (AAL3) brain atlas is used as a brain region segmentation template. The AAL3 template contains 166 regions of interest (ROIs), covering the main gray matter structures of the entire brain. Based on the sMRI images in step S1, an sMRI gray matter probability map can be obtained. The sMRI gray matter probability map and the AAL3 template are overlaid in MNI space, and the average gray matter volume within each ROI is calculated as the structural feature of the ROI. Considering that the cerebellum and vermis are usually used as reference areas in PET modality SUVR calculations, and that their pathological changes are weakly correlated with the core features of Alzheimer's disease, this embodiment excludes 26 ROIs belonging to the cerebellum and vermis. Therefore, 140-dimensional gray matter volume features are extracted from each sMRI image.

[0062] The extraction process of standardized uptake value ratio features based on Aβ-PET images is as follows: For the Aβ-PET images in step S1, the same AAL3 brain atlas as the sMRI images is used (26 ROIs excluding the cerebellum and vermis), and the standardized uptake value ratio (SUVR) within each ROI is calculated. The formula for calculating SUVR is:

[0063] (1)

[0064] in, The number of voxels within the target ROI. For the first ROI Standardized intake values ​​of individual factors; The number of voxels within the entire cerebellar ROI. For the entire cerebellum Individual factors value.

[0065] In this embodiment, the entire cerebellum is used as the reference area to calculate SUVR, and 140-dimensional normalized uptake ratio features are extracted from each Aβ-PET image.

[0066] The extraction process of the normalized uptake ratio feature based on tau-PET images is the same as that of Aβ-PET images. For the tau-PET images in step S1, the SUVR value of each ROI is calculated based on the AAL3 brain atlas (excluding 26 ROIs in the cerebellum and vermis) to obtain 140-dimensional normalized uptake ratio features that characterize the pathological features of tau protein.

[0067] The 140-dimensional gray matter volume features extracted from sMRI images, the 140-dimensional normalized uptake ratio features extracted from Aβ-PET images, and the 140-dimensional normalized uptake ratio features extracted from tau-PET images are concatenated to obtain the final multimodal features, with a total dimension of 140 × 3 = 420. These 420-dimensional multimodal features serve as input to step S3 for subsequent feature selection based on a subset of the target features.

[0068] Step S3: Obtain a target feature subset pre-constructed based on an improved arithmetic optimization algorithm, and perform feature selection on the multimodal features based on the target feature subset to obtain the feature selection result.

[0069] Step S4: Use a fuzzy support vector machine classifier to classify the feature selection results to obtain the Alzheimer's disease detection results.

[0070] The Alzheimer's disease test results include the following three categories:

[0071] Cognitively Normal (CN): This indicates that the target subject did not show obvious cognitive impairment and was considered a normal control.

[0072] Mild cognitive impairment (MCI): This indicates that the target individual has a decline in cognitive function, but has not yet met the diagnostic criteria for dementia. It belongs to the early or prodromal stage of Alzheimer's disease.

[0073] Alzheimer's disease: This indicates that the target individual has met the clinical diagnostic criteria for Alzheimer's disease and is in the late stage of the disease.

[0074] In practical applications, the fuzzy support vector machine classifier in step S4 can output any of the three categories mentioned above, or it can output a binary classification result between any two categories according to clinical needs, such as distinguishing between Alzheimer's disease and normal cognition, Alzheimer's disease and mild cognitive impairment, or mild cognitive impairment and normal cognition.

[0075] The process of constructing the target feature subset based on the improved arithmetic optimization algorithm in this embodiment is as follows:

[0076] Step A: Construct a sample set, which includes multiple sets of multimodal samples and their corresponding category labels. The sample set is divided into a training set and a test set.

[0077] Step B: Initialize the population matrix of the improved arithmetic optimization algorithm. Each row vector of the population matrix represents an individual, and each individual corresponds to a subset of candidate features. The dimension of the row vector is equal to the total number of features of the multimodal samples, and the value of each dimension is a continuous search variable. This continuous search variable, after being masked, indicates the corresponding feature selection state of the training samples.

[0078] Step C involves iteratively updating each dimension value in the population matrix through the following sub-steps until the maximum number of iterations is reached or the fitness value converges:

[0079] C1, using an exploratory update strategy or an exploitative update strategy with an arithmetic optimization algorithm to update the current values ​​of each dimension, and obtain the first update result of the population matrix;

[0080] C2 employs a cross-sectional strategy to perform horizontal and vertical cross-sectional calculations on each individual in the first update result of the population matrix. The horizontal cross-sectional strategy weights and fuses the dimension values ​​of two different individuals in the same dimension, then perturbs the differences to generate a first temporary candidate feature subset for each individual. If the fitness value corresponding to the first temporary candidate feature subset of each individual is less than the fitness value of the original candidate feature subset before the horizontal cross-sectional strategy, then the candidate feature subset of that individual in the first update result is updated to a first temporary candidate feature subset, resulting in the second update result of the population matrix. The vertical cross-sectional strategy randomly weights and fuses the dimension values ​​of the same individual in different dimensions to generate a second temporary candidate feature subset for that individual. If the fitness value corresponding to the second temporary candidate feature subset of that individual is less than the fitness value of the candidate feature subset in the second update result, then the candidate feature subset of that individual in the second update result is updated to a second temporary candidate feature subset, resulting in the cross-sectional update result of the population matrix.

[0081] C3, when the whale fall triggering condition is met in the current iteration round, the current individual is perturbed and updated based on the current individual in the cross-up update result, other randomly selected individuals, and the decay step size corresponding to the current round, to obtain a new candidate feature subset after perturbed update. If the fitness value of the new candidate feature subset is less than the fitness value before perturbed update, the candidate feature subset of the individual in the cross-up update result is updated to the new candidate feature subset, and the population matrix after this iteration is obtained.

[0082] Step D: After the iteration is completed, the individual with the smallest fitness value is determined from the final population matrix, and the candidate feature subset of this individual is taken as the target feature subset.

[0083] The fitness value of each candidate feature subset is the cost function value, used to evaluate the overall quality of the candidate feature subset. In this embodiment, the fitness value is calculated based on the classification error rate of the candidate feature subset on the fuzzy support vector machine classifier, the number of selected features, and the stability of the classifier's cross-validation.

[0084] Next, combined Figure 2 Figure 3 This application provides a detailed description of the Alzheimer's disease detection process in its embodiments.

[0085] like Figure 2 As shown in this embodiment, a fuzzy support vector machine diagnostic framework driven by a feature selection optimization algorithm that integrates whale fall and cross-sectional strategies is proposed. This method fuses sMRI images, Aβ-PET images, and tau-PET images. Through the synergy of an improved arithmetic optimization algorithm and fuzzy support vector machines, it achieves simultaneous optimization selection of high-dimensional features and noise-robust classification, thereby significantly improving the diagnostic accuracy, model generalization ability, and clinical interpretability of Alzheimer's disease.

[0086] In this embodiment, a sample set is pre-constructed to form a subset of target features based on an improved arithmetic optimization algorithm.

[0087] In this embodiment, multimodal neuroimaging data of participants were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. The multimodal neuroimaging data included sMRI sample images, Aβ-PET sample images, and tau-PET sample images. This embodiment used multimodal images from 154 participants, including three groups: cognitively normal, mild cognitive impairment, and Alzheimer's disease. Sample size and demographic information for each group are detailed in Table 1.

[0088] Table 1: Patient Information

[0089]

[0090] Then, the sample images are preprocessed and feature extracted.

[0091] After obtaining the original multimodal sample images, for the sMRI sample images, MATLAB's CAT12 was used to complete the image quality check, spatial normalization to MNI standard space, tissue segmentation and other processes. Based on the AAL3 brain atlas, gray matter volume features were extracted. After excluding 26 ROIs belonging to the cerebellum and vermis, 140-dimensional gray matter volume features were obtained, which formed the feature vector of the sMRI modality.

[0092] It is worth noting that the AAL3 template used in this embodiment is an enhanced brain region map, which adds several fine regions closely related to cognition, emotion and reward systems on the basis of AAL and AAL2, supporting accurate analysis of AD-related brain regions (such as hippocampus, amygdala, posterior cingulate cortex, etc.).

[0093] For Aβ-PET and tau-PET sample images, the Coregister operation of SPM12 was first used to register with the corresponding sMRI. Then, the deformation field generated by sMRI was normalized to MNI space using Normalise. Finally, the normalized uptake ratio features of each ROI were extracted. Similarly, 26 ROIs of the cerebellum were excluded. 140-dimensional features were obtained for each modality, for a total of 420-dimensional multimodal feature vectors.

[0094] Thus, the sample set is constructed through the above steps. After the sample set is constructed, it is divided into multiple training and test sets according to a preset ratio, so that the mean accuracy and accuracy standard deviation can be calculated based on 5-fold cross-validation.

[0095] This embodiment employs a fuzzy support vector machine (SVM) classifier. Traditional SVM classifiers assume all samples are equally important, but medical image data often contains noise, outliers, or labeling uncertainties. Therefore, fuzzy membership is introduced. , The membership degree is calculated based on the distance of the sample to its class center: typical samples that are closer to the class center are assigned a higher membership degree (close to 1), while noisy or outlier samples are assigned a lower membership degree (close to 0), thereby reducing their impact on the decision surface.

[0096] A fuzzy support vector machine classifier is constructed using a Gaussian kernel function and prepared for subsequent optimization.

[0097] The improved arithmetic optimization algorithm in this embodiment is as follows: Figure 3 As shown, the following population matrix is ​​first constructed:

[0098] (2)

[0099] in, Indicates the number of individuals in a population, for example, ; This represents the total number of candidate features for the current modality combination. , Indicates the first The individual in the first Continuous search values ​​across feature dimensions.

[0100] In this embodiment, a Gaussian distribution is used for population initialization instead of a uniform random distribution.

[0101] (3)

[0102]

[0103] in, Let be the standard deviation of the Gaussian distribution. Let be the variance of the Gaussian distribution; and The first Upper and lower bounds of a dimensional variable, for example, , ; Indicates the first step in the initialization process. The individual in the first Initial continuous search values ​​on each feature dimension.

[0104] This initialization method ensures that the initial solutions are mainly distributed in the middle of the search space, improving early search efficiency.

[0105] In the In the next iteration, step C1 is executed first: update the individual position according to the arithmetic optimization algorithm.

[0106] Mathematical optimization acceleration factor Used to control the switching between the exploration and development phases, mathematical optimization probability parameters. Used to adjust the search step size, the two are as follows:

[0107] (4)

[0108] in, This represents the current iteration number. This represents the maximum number of iterations. These are the minimum and maximum values ​​of the acceleration function, used to control the speed at which the algorithm transitions from global exploration to local exploitation. A sensitive parameter for controlling search accuracy, used to adjust the degree to which the search step size decays with the number of iterations.

[0109] For example, , , In other implementations, adjustments can be made based on the data size and convergence requirements.

[0110] when At that time, perform multiplication or division to explore and update:

[0111] (5)

[0112] when During this time, perform addition or subtraction to update the process:

[0113] (6)

[0114] in, Indicates the first In the iteration, the individual with the minimum fitness value in the population is at the . The value of dimension, To prevent division by zero of extremely small constants, For control parameters, Belongs to Random numbers.

[0115] Subsequently, step C2 is performed: the updated population is further optimized based on the cross-cutting strategy.

[0116] Horizontal crossing in two different individuals and Between:

[0117] (7)

[0118] in, and The first temporary dimension value for the two individuals; , .

[0119] This horizontal crossover operation enhances population diversity and local exploitation by fusing feature dimension information from two individuals and introducing differential perturbations. If the fitness value of the first temporary dimension is better than that of the parent, the parent individual is replaced.

[0120] Vertically intersecting in the same individual Different dimensions and Between:

[0121] (8)

[0122] in, The second temporary dimension value generated for the vertical intersection. , , .

[0123] The vertical crossover operation further enhances the algorithm's global exploration capability. After the vertical crossover, the selection is still based on the fitness value.

[0124] Next, proceed to step C3: introduce a whale fall strategy for perturbation updates.

[0125] The whale fall update formula is:

[0126] (9)

[0127] in, An individual randomly selected from the current population. , = , It is the step length of the whale fall process.

[0128] Among them, whale fall probability Defined as:

[0129] (10)

[0130] When the whale fall trigger condition is met At this point, the aforementioned perturbation is applied to individuals to increase population diversity and avoid getting trapped in local optima. The whale fall strategy ensures that even at the end of the iteration, the algorithm has a strong ability to escape local optima. Here... It is a random number. , .

[0131] Following a whale fall disturbance, competitive selection also occurs based on fitness values.

[0132] By repeatedly executing the iterative process from C1 to C3 until the maximum number of iterations is reached or the fitness value converges, the final population matrix is ​​obtained.

[0133] After obtaining the final population matrix, the individual with the smallest fitness value is determined from the final population matrix. The candidate feature subset of this individual is used as the target feature subset. Based on the target feature subset, feature selection is performed on the multimodal feature vector to obtain the feature selection result.

[0134] In some embodiments, the fitness value of each candidate feature subset is obtained through the following steps:

[0135] Based on a preset mapping threshold, feature mask mapping is performed on the values ​​of each dimension in the candidate feature subset to obtain a candidate feature selection mask sequence;

[0136] Based on the candidate feature selection mask sequence, the selected features in each test sample in the test set are determined, and the feature selection result corresponding to each test sample is obtained.

[0137] The fuzzy support vector machine classifier is used to classify the feature selection results corresponding to each test sample to obtain the sample detection results corresponding to each test sample.

[0138] Based on the sample detection results and their corresponding category labels for all test samples, the average accuracy and the standard deviation of accuracy are obtained.

[0139] The fitness value of each candidate feature subset is calculated based on the average accuracy and the standard deviation of accuracy.

[0140] Specifically, the individuals generated in each iteration Since the upper and lower bounds of the search variables are set to 1 and 0 in this embodiment, the continuous search variables are converted into binary candidate feature selection masks by using a threshold of 0.5.

[0141] (11)

[0142] in, This indicates that the first eigenvector in the multimodal feature vector is retained. Each dimension value This indicates the removal of the first eigenvector from the multimodal eigenvectors. Each dimension value.

[0143] The number of features selected by the current individual is .

[0144] A subset of candidate features is obtained by selecting a mask sequence based on the candidate features. The feature subset is then input into a fuzzy support vector machine classifier, and 5-fold cross-validation is performed on the test set to obtain the average accuracy. Accuracy standard deviation .

[0145] To achieve a balance between classification performance, feature count, and model stability, the improved arithmetic optimization algorithm uses the following improved fitness function:

[0146] (12)

[0147] in, For fitness value, The preset weighting coefficients, The average accuracy rate. The standard deviation of the accuracy is given.

[0148] The improved arithmetic optimization algorithm in this embodiment updates the population matrix with the goal of minimizing the fitness value, and outputs the individual with the smallest fitness value as the optimal feature subset after the iteration.

[0149] In addition, experimental tests and result analysis were conducted in this embodiment.

[0150] Based on 154 multimodal data points from the ADNI dataset, the experimental results shown in Tables 2 and 3 were obtained in this experiment.

[0151] Table 2: AD prediction results, including accuracy, precision, recall, and F1 score.

[0152]

[0153] As shown in Table 2, the trimodal features, after being optimized by the improved arithmetic algorithm, achieved a classification accuracy of 96.8% in Alzheimer's disease and normal cognitive tasks, which is significantly better than the single-modal method. Specifically, sMRI achieved approximately 92.6%, Aβ-PET approximately 93.1%, and tau-PET approximately 93.6%.

[0154] Table 3: Statistical results of the top ten features with the most selection times for each modality during the feature selection process.

[0155]

[0156] Combined with Table 3 and Figure 4 As shown, feature selection heatmaps and Top 10 ROI analysis obtained through multiple runs revealed that frequently selected brain regions included the left hippocampus (lHIP), right parahippocampal gyrus (rPHG), bilateral amygdalae (lAMYG, rAMYG), and posterior cingulate cortex (rPCC), which highly correlated with classic Alzheimer's disease pathology. Modal contribution analysis confirmed that Aβ-PET was superior in detecting early mild cognitive impairment, tau-PET dominated in neurodegenerative changes in late Alzheimer's disease, and sMRI provided stable structural evidence across all stages, reflecting the natural asynchronous progression of Alzheimer's pathology. The improved arithmetic optimization algorithm provided in this embodiment performed excellently in multiple metrics, including accuracy, precision, recall, and F1-score, exhibiting stronger robustness and clinical interpretability, especially in small-sample, high-dimensional scenarios.

[0157] like Figure 5 The convergence curve diagram shown in this embodiment demonstrates that the EAOA-WFCS algorithm in this embodiment, compared with the IEEE CEC2017 benchmark function, shows that its optimization performance is significantly better than classic metaheuristic algorithms such as the original AOA and the improved AOA optimization algorithm.

[0158] Figure 6 This is a schematic diagram of an electronic device illustrated in this specification according to an exemplary embodiment. Please refer to... Figure 6 At the hardware level, the device includes a processor 610, an internal bus 620, a network interface 630, memory 640, a hardware acceleration device 650, and non-volatile memory 660, and may also include other hardware required for its functions. One or more embodiments of this application can be implemented in software, for example, the processor 610 reads the corresponding computer program from the non-volatile memory 660 into the memory 640 and then runs it. Of course, in addition to software implementation, one or more embodiments of this application do not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. That is to say, the execution subject of the above processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0159] Figure 7 This is a structural block diagram of an Alzheimer's disease detection device based on an improved arithmetic optimization algorithm, as illustrated in an exemplary embodiment of this application. The Alzheimer's disease detection device can be applied to, for example... Figure 6 The electronic device shown implements the technical solution of this application. The Alzheimer's disease detection device includes: an image acquisition unit 710, a feature extraction unit 720, a feature selection unit 730, and a classification detection unit 740, wherein:

[0160] The image acquisition unit 710 is used to acquire multimodal images of the target object, including sMRI images, Aβ-PET images, and tau-PET images;

[0161] The feature extraction unit 720 is used to extract features from the multimodal image to obtain multimodal features;

[0162] The feature selection unit 730 is used to obtain a target feature subset pre-constructed based on an improved arithmetic optimization algorithm, and to perform feature selection on the multimodal features based on the target feature subset to obtain a feature selection result;

[0163] The classification and detection unit 740 is used to classify the feature selection results using a fuzzy support vector machine classifier to obtain Alzheimer's disease detection results;

[0164] The process of constructing the target feature subset based on the improved arithmetic optimization algorithm is as follows:

[0165] Step A: Construct a sample set, which includes multiple sets of multimodal samples and their corresponding category labels. The sample set is divided into a training set and a test set.

[0166] Step B: Initialize the population matrix of the improved arithmetic optimization algorithm. Each row vector of the population matrix represents an individual, and each individual corresponds to a subset of candidate features. The dimension of the row vector is equal to the total number of features of the multimodal samples, and the value of each dimension is a continuous search variable. This continuous search variable, after being masked, indicates the corresponding feature selection state of the training samples.

[0167] Step C involves iteratively updating each dimension value in the population matrix through the following sub-steps until the maximum number of iterations is reached or the fitness value converges:

[0168] C1, using an exploratory update strategy or an exploitative update strategy with an arithmetic optimization algorithm to update the current values ​​of each dimension, and obtain the first update result of the population matrix;

[0169] C2 employs a cross-sectional strategy to perform horizontal and vertical cross-sectional calculations on each individual in the first update result of the population matrix. The horizontal cross-sectional strategy weights and fuses the dimension values ​​of two different individuals in the same dimension, then perturbs the differences to generate a first temporary candidate feature subset for each individual. If the fitness value corresponding to the first temporary candidate feature subset of each individual is less than the fitness value of the original candidate feature subset before the horizontal cross-sectional strategy, then the candidate feature subset of that individual in the first update result is updated to a first temporary candidate feature subset, resulting in the second update result of the population matrix. The vertical cross-sectional strategy randomly weights and fuses the dimension values ​​of the same individual in different dimensions to generate a second temporary candidate feature subset for that individual. If the fitness value corresponding to the second temporary candidate feature subset of that individual is less than the fitness value of the candidate feature subset in the second update result, then the candidate feature subset of that individual in the second update result is updated to a second temporary candidate feature subset, resulting in the cross-sectional update result of the population matrix.

[0170] C3, when the whale fall triggering condition is met in the current iteration round, the current individual is perturbed and updated based on the current individual in the cross-up update result, other randomly selected individuals, and the decay step size corresponding to the current round, to obtain a new candidate feature subset after perturbed update. If the fitness value of the new candidate feature subset is less than the fitness value before perturbed update, the candidate feature subset of the individual in the cross-up update result is updated to the new candidate feature subset, and the population matrix after this iteration is obtained.

[0171] Step D: After the iteration is completed, the individual with the smallest fitness value is determined from the final population matrix, and the candidate feature subset of this individual is taken as the target feature subset.

[0172] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0173] Accordingly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any of the above embodiments.

[0174] Accordingly, embodiments of this application also provide a computer program product configured to perform the methods described in any of the above embodiments.

[0175] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, which can take the form of a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email sending and receiving device, game console, tablet computer, wearable device, or any combination of these devices.

[0176] In a typical configuration, a computer includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0177] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0178] Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random-access memory (SRAM), dynamic random-access memory (DRAM), other types of random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage, quantum memory, graphene-based storage media or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0179] While this specification contains numerous specific implementation details, these should not be construed as limiting the scope of any invention or the scope of the claims, but rather are primarily intended to describe features of specific embodiments of a particular invention. Certain features described in the various embodiments herein may also be implemented in combination in a single embodiment. Conversely, various features described in a single embodiment may also be implemented separately in various embodiments or in any suitable sub-combination. Furthermore, while features may function in certain combinations as described above and even initially claimed in this way, one or more features from a claimed combination may be removed from that combination in some cases, and a claimed combination may refer to a sub-combination or a variation thereof.

[0180] Similarly, although the operations are depicted in a specific order in the accompanying drawings, this should not be construed as requiring these operations to be performed in the specific order shown or sequentially, or requiring all illustrated operations to be performed to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Furthermore, the separation of various system modules and components in the above embodiments should not be construed as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.

[0181] Thus, specific embodiments of the subject matter have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims may be performed in a different order and still achieve the desired result. Furthermore, the processes depicted in the drawings are not necessarily shown in a specific order or sequence to achieve the desired result. In some implementations, multitasking and parallel processing may be advantageous.

[0182] It should be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0183] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. An Alzheimer's disease detection method based on an improved arithmetic optimization algorithm, characterized in that, Includes the following steps: Step S1: Acquire multimodal images of the target object, including sMRI images, Aβ-PET images, and tau-PET images; Step S2: Extract features from the multimodal image to obtain multimodal features; Step S3: Obtain a target feature subset pre-constructed based on an improved arithmetic optimization algorithm, and perform feature selection on the multimodal features based on the target feature subset to obtain the feature selection result; Step S4: Use a fuzzy support vector machine classifier to classify the feature selection results to obtain the Alzheimer's disease detection results; The process of constructing the target feature subset based on the improved arithmetic optimization algorithm is as follows: Step A: Construct a sample set, which includes multiple sets of multimodal samples and their corresponding category labels. The sample set is divided into a training set and a test set. Step B: Initialize the population matrix of the improved arithmetic optimization algorithm. Each row vector of the population matrix represents an individual, and each individual corresponds to a subset of candidate features. The dimension of the row vector is equal to the total number of features of the multimodal samples, and the value of each dimension is a continuous search variable. This continuous search variable, after being masked, indicates the corresponding feature selection state of the training samples. Step C involves iteratively updating each dimension value in the population matrix through the following sub-steps until the maximum number of iterations is reached or the fitness value converges: C1, using an exploratory update strategy or an exploitative update strategy with an arithmetic optimization algorithm to update the current values ​​of each dimension, and obtain the first update result of the population matrix; C2 employs a cross-sectional strategy to perform horizontal and vertical cross-sectional calculations on each individual in the first update result of the population matrix. The horizontal cross-sectional strategy weights and fuses the dimension values ​​of two different individuals in the same dimension, then perturbs the differences to generate a first temporary candidate feature subset for each individual. If the fitness value corresponding to the first temporary candidate feature subset of each individual is less than the fitness value of the original candidate feature subset before the horizontal cross-sectional strategy, then the candidate feature subset of that individual in the first update result is updated to a first temporary candidate feature subset, resulting in the second update result of the population matrix. The vertical cross-sectional strategy randomly weights and fuses the dimension values ​​of the same individual in different dimensions to generate a second temporary candidate feature subset for that individual. If the fitness value corresponding to the second temporary candidate feature subset of that individual is less than the fitness value of the candidate feature subset in the second update result, then the candidate feature subset of that individual in the second update result is updated to a second temporary candidate feature subset, resulting in the cross-sectional update result of the population matrix. C3, when the whale fall triggering condition is met in the current iteration round, the current individual is perturbed and updated based on the current individual in the cross-up update result, other randomly selected individuals, and the decay step size corresponding to the current round, to obtain a new candidate feature subset after perturbed update. If the fitness value of the new candidate feature subset is less than the fitness value before perturbed update, the candidate feature subset of the individual in the cross-up update result is updated to the new candidate feature subset, and the population matrix after this iteration is obtained. Step D: After the iteration is completed, the individual with the smallest fitness value is determined from the final population matrix, and the candidate feature subset of this individual is taken as the target feature subset.

2. The method according to claim 1, characterized in that, The two different individuals in C2 , The horizontal crossover results are as follows: ； ； in, for Random numbers within a range Within the range Random numbers, For two individuals Dimension value, The first temporary dimension value for the two individuals.

3. The method according to claim 1, characterized in that, Individuals in C2 The results of the vertical intersections are as follows: ； in, For individuals The second temporary dimension value, For individuals The Dimension and First Dimensional values ​​under a dimension for A random number within a given range.

4. The method according to claim 1, characterized in that, In C3, when the probability of a whale fall... Not greater than the random reference value When the current iteration round satisfies the whale fall trigger condition, it is determined that: ， ； For the current iteration round, The maximum number of iterations, for A random number within a given range.

5. The method according to claim 1, characterized in that, The fitness value of each candidate feature subset is obtained through the following steps: Based on a preset mapping threshold, feature mask mapping is performed on the values ​​of each dimension in the candidate feature subset to obtain a candidate feature selection mask sequence; Based on the candidate feature selection mask sequence, the selected features in each test sample in the test set are determined, and the feature selection result corresponding to each test sample is obtained. The fuzzy support vector machine classifier is used to classify the feature selection results corresponding to each test sample to obtain the sample detection results corresponding to each test sample. Based on the sample detection results and their corresponding category labels for all test samples, the average accuracy and the standard deviation of accuracy are obtained. The fitness value of each candidate feature subset is calculated based on the average accuracy and the standard deviation of accuracy.

6. The method according to claim 5, characterized in that, The expression for the fitness value is: ； in, For fitness value, The preset weighting coefficients, The average accuracy rate. The standard deviation of the accuracy is given.

7. The method according to claim 1, characterized in that, The multimodal features in step S2 include gray matter volume features based on sMRI images, normalized uptake ratio features based on Aβ-PET images, and normalized uptake ratio features based on tau-PET images.

8. An Alzheimer's disease detection device based on an improved arithmetic optimization algorithm, characterized in that, The device includes: An image acquisition unit is used to acquire multimodal images of a target object, including sMRI images, Aβ-PET images, and tau-PET images; The feature extraction unit is used to extract features from the multimodal image to obtain multimodal features; The feature selection unit is used to obtain a target feature subset pre-constructed based on an improved arithmetic optimization algorithm, and to perform feature selection on the multimodal features based on the target feature subset to obtain the feature selection result; The classification and detection unit is used to classify the feature selection results using a fuzzy support vector machine classifier to obtain Alzheimer's disease detection results; The process of constructing the target feature subset based on the improved arithmetic optimization algorithm is as follows: Step A: Construct a sample set, which includes multiple sets of multimodal samples and their corresponding category labels. The sample set is divided into a training set and a test set. Step B: Initialize the population matrix of the improved arithmetic optimization algorithm. Each row vector of the population matrix represents an individual, and each individual corresponds to a subset of candidate features. The dimension of the row vector is equal to the total number of features of the multimodal samples, and the value of each dimension is a continuous search variable. This continuous search variable, after being masked, indicates the corresponding feature selection state of the training samples. Step C involves iteratively updating each dimension value in the population matrix through the following sub-steps until the maximum number of iterations is reached or the fitness value converges: C1, using an exploratory update strategy or an exploitative update strategy with an arithmetic optimization algorithm to update the current values ​​of each dimension, and obtain the first update result of the population matrix; C2 employs a cross-sectional strategy to perform horizontal and vertical cross-sectional calculations on each individual in the first update result of the population matrix. The horizontal cross-sectional strategy weights and fuses the dimension values ​​of two different individuals in the same dimension, then perturbs the differences to generate a first temporary candidate feature subset for each individual. If the fitness value corresponding to the first temporary candidate feature subset of each individual is less than the fitness value of the original candidate feature subset before the horizontal cross-sectional strategy, then the candidate feature subset of that individual in the first update result is updated to a first temporary candidate feature subset, resulting in the second update result of the population matrix. The vertical cross-sectional strategy randomly weights and fuses the dimension values ​​of the same individual in different dimensions to generate a second temporary candidate feature subset for that individual. If the fitness value corresponding to the second temporary candidate feature subset of that individual is less than the fitness value of the candidate feature subset in the second update result, then the candidate feature subset of that individual in the second update result is updated to a second temporary candidate feature subset, resulting in the cross-sectional update result of the population matrix. C3, when the whale fall triggering condition is met in the current iteration round, the current individual is perturbed and updated based on the current individual in the cross-up update result, other randomly selected individuals, and the decay step size corresponding to the current round, to obtain a new candidate feature subset after perturbed update. If the fitness value of the new candidate feature subset is less than the fitness value before perturbed update, the candidate feature subset of the individual in the cross-up update result is updated to the new candidate feature subset, and the population matrix after this iteration is obtained. Step D: After the iteration is completed, the individual with the smallest fitness value is determined from the final population matrix, and the candidate feature subset of this individual is taken as the target feature subset.

9. An electronic device, characterized in that, include: processor; A computer-readable storage medium storing computer program instructions that, when executed by the processor, cause the processor to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor according to any one of claims 1 to 7.

Images