Image genome-based Alzheimer's disease scale score prediction device
By combining imaging genomic data and multi-strategy improved crown porcupine optimization algorithm, the problem of predicting progress in the early stages of Alzheimer's disease is solved, and more accurate scale score prediction and more comprehensive disease understanding are achieved, supporting the implementation of precision medicine.
Patent Information
- Application Number
- CN202510066882.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-16
- Publication Date
- 2025-06-10
AI Technical Summary
The prior art is difficult to accurately identify and predict patient progress in the early stages of Alzheimer's disease, making it difficult to grasp the treatment timing.
Using a prediction device based on the image genome, the data acquirer, data preprocessor, MCPO module and prediction module are connected to the magnetic resonance imaging pictures, single nucleotide polymorphism data and psychological assessment scores, and feature selection and prediction are used for multi-strategy improved crown porcupine optimization algorithm (MCPO) and support vector regression (SVR).
Improved the prediction accuracy of Alzheimer's Scale scores, providing a more comprehensive understanding of the pathogenesis and clinical manifestations of Alzheimer's disease, thereby supporting the implementation of precision medicine.
Smart Images
Figure CN120126798A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of brain imaging genomics, and relates to an Alzheimer's disease scale score prediction device based on imaging genome. Background Art
[0002] Alzheimer's disease (AD) is an irreversible neurodegenerative disease that accounts for approximately 70% of all dementia cases. MCI is considered an intermediate stage of AD, with 10-15% of MCI progressing to AD each year. Unfortunately, the cause and mechanism of AD are still not fully understood, and there is no cure. The clinical manifestations and pathological characteristics of AD patients are highly heterogeneous, and this heterogeneity makes it difficult for researchers to capture the common characteristics of patients. Therefore, accurately identifying MCI and AD patients remains a major challenge.
[0003] AD is an irreversible neurodegenerative disease that worsens with age. Clinical manifestations include a series of mental and cognitive disorders, including memory loss and behavioral changes, which affect people's ability to live a normal life. Cognitive impairment can be regarded as the early stage of Alzheimer's disease. Nearly 10%-15% of patients with cognitive impairment are converted to Alzheimer's disease patients every year. So far, no effective treatment has been found. Therefore, it is particularly important to identify and intervene in the cognitive impairment stage as early as possible.
[0004] Changes in AD biomarkers can be detected through different neuroimaging methods. However, the symptoms in the early stages of AD are often not obvious, and the changes in imaging and biomarkers are also small, making it difficult to distinguish from normal aging. Genetic data can provide early risk assessment of the disease and is becoming increasingly important in AD research. Studies have found that apolipoprotein E (APOE) is one of the strongest genetic factors, and carriers of its ε4 allele are more likely to have amyloid deposition. Imaging genomics combines the two types of data to provide information at different levels of the disease, thereby improving the accuracy of disease classification. Summary of the invention
[0005] In order to solve the defects in the prior art, the present invention provides an Alzheimer's disease scale score prediction device based on image genome, and the technical solution is:
[0006] It includes a data acquisition device, a data preprocessor, a MCPO module and a prediction module connected in sequence, wherein:
[0007] The data acquired by the data acquirer include magnetic resonance imaging images, single nucleotide polymorphism data and psychological test scores;
[0008] The data preprocessor normalizes, strips and segments the magnetic resonance imaging images, performs standard quality control, estimates missing values and digitally encodes the single nucleotide polymorphism data;
[0009] The MCPO module integrates multiple data sources and updates the results;
[0010] The prediction module is combined with the MCPO module to obtain the optimal features.
[0011] Preferably, the data preprocessor processes the magnetic resonance imaging images using CAT12.
[0012] Preferably, the data preprocessor uses PLINK to perform standard quality control on the single nucleotide polymorphism data.
[0013] Preferably, the MCPO module comprises an initial submodule, a dynamic reverse submodule, a cycle reduction submodule and an improved defense submodule which are connected in sequence.
[0014] Preferably, the initial submodule is initialized based on the good point set strategy, specifically:
[0015] Assume G s is the unit cube in s-dimensional Euclidean space, r∈G s ,
[0016]
[0017] Its deviation satisfies
[0018]
[0019] Where C(r,ε)n -1+ε is a constant that is only related to r and ε, P n (k) represents the set of optimal points, where r is the optimal point; represents the decimal part, and n represents the number of points;
[0020] make Mapping it to the search space is expressed as,
[0021]
[0022] where x i (j) is the jth dimension of the i-th crested porcupine, is its advantage, j and lb j denotes the upper and lower bounds of the j-th dimension.
[0023] Preferably, the dynamic reverse submodule is specifically represented by a mathematical model as follows:
[0024] Xd =X+c 1 ×(c 2 ×(ub+lb-X)-X) (4)
[0025] Where X d is the dynamic inverse solution, c 1 and c 2 is a random number between 0 and 1, and ub and lb are its upper and lower limits.
[0026] Preferably, the cycle reduction submodule is specifically represented by a mathematical model as follows:
[0027]
[0028] Where N is the population size, N min is the minimum number of individuals in the newly generated population, so that the population size cannot be less than N min , T c Used to determine the number of loops, T is the maximum number of function evaluations.
[0029] Preferably, the improved defense submodule includes four defense strategies, specifically:
[0030] The inverse cumulative distribution function of the Cauchy distribution improves the first defense strategy: The probability density function of the Cauchy distribution is
[0031]
[0032] where x 0 is the location parameter that defines the peak position of the distribution, is the scale parameter of the half width at half the maximum value, and the random variable X follows the Cauchy distribution as X~C(γ,x 0 ),γ=1,x 0 = 0 is called the standard Cauchy distribution, and its probability density function and corresponding cumulative distribution function are shown in formulas (7) and (8).
[0033]
[0034] In the case where the cumulative distribution function of the Cauchy distribution is calculated as an inverse function, the inverse cumulative distribution function of the Cauchy distribution is defined as,
[0035]
[0036] Where p = randn(1, dim), the first defense strategy is expressed as formula (10),
[0037]
[0038] where τ 1 is a random number based on normal distribution, τ2 is a random number in the interval [0,1], is the optimal solution, is the position of the i-th individual at the t-th iteration;
[0039] Second defense strategy: Crested porcupines threaten predators by making noise. When a predator approaches, the crested porcupine's voice will become louder, represented by:
[0040]
[0041] where r1 and r2 are random integers between [1,N], τ 3 is a random number between [0,1], Represents the vector generated between the current individual and the randomly selected individual; binary vector Includes randomly generated 0s and 1s. When a cell in this vector includes 0, the predator stands still and does not move toward the crested porcupine. When a cell contains 1, it means that the predator may move closer to or away from the crested porcupine.
[0042] Third defense strategy: Entering the third area, the crested porcupine secretes a foul odor that spreads in the area around it to prevent predators from approaching it, represented by:
[0043]
[0044] The vector is used To simulate the three possible scenarios in this strategy, r3 is a parameter between [1, N] used to control the search direction, γ t and represent the defense factor and odor diffusion factor, respectively, as shown in formulas (13)-(15),
[0045]
[0046] represents the objective function value of the i-th individual at the t-th iteration, ∈ is a small number to avoid division by zero, is a vector containing randomly generated values between 0 and 1, and rand is a random variable between 0 and 1;
[0047] Tangential flight strategy improves the fourth defense strategy: The calculation formula of the tangential flight strategy is as follows,
[0048]
[0049] Where v = randn(1, dim), applying the tangent branch operator to the fourth defense strategy is expressed as:
[0050]
[0051] Where α is the convergence speed factor, τ 4 and τ 5 is a random value between [0,1], is the average force affecting the ith individual, m i is the mass of the ith individual, is a vector containing generated random values between 0 and 1.
[0052] Preferably, the prediction module combines feature selection and prediction, and calculates scores for the image gene features after the data preprocessor through the MCPO module.
[0053] Preferably, the prediction module models the problem as maximizing modal correlation and minimizing error, and the fitness function is constructed as:
[0054] fitness=MSE(Z,Z ′ )+ω(-U′X′YV+‖U‖ 1 +‖V‖ 1 ) (20)
[0055] Among them, X and Y are image and gene features, U and V are their corresponding weight coefficients, Z and Z' are the actual scores and SVR predicted scores of the subjects, respectively. 1 is the l1 norm, ω is the regularization coefficient, and through formula (20), MCPO iteratively optimizes the weights of the features and finally selects the optimal features to input into SVR for prediction.
[0056] Compared with the prior art, the present invention includes at least the following beneficial effects: the Crested Porcupine Optimizer (CPO) algorithm is improved and combined with Support Vector Regression (SVR) to predict the patient's neurological scale score, making full use of the cutting-edge nature of gene features to improve the reliability of prediction. The experimental results verified that the multi-strategy improved Crested Porcupine Optimizer (MCPO) has excellent performance on the CEC2017 benchmark. In addition, by integrating multiple data sources, we have a more comprehensive understanding of the pathogenesis and clinical manifestations of AD. In addition, the MCPO algorithm is combined with canonical correlation analysis (CCA) for feature selection and regression, which further enhances the ability to evaluate AD neurological scale scores. This method provides more reliable support for precision medicine and highlights the potential of intelligent optimization algorithms in the medical field. BRIEF DESCRIPTION OF THE DRAWINGS
[0057] Figure 1 It is a schematic diagram of the structure of an Alzheimer's disease scale score prediction device based on image genome according to an embodiment of the present invention;
[0058] Figure 2 A convergence curve diagram (dim=30) of a benchmark function for evaluating an Alzheimer's disease scale score prediction device based on image genome according to an embodiment of the present invention;
[0059] Figure 3 It is a convergence curve diagram (dim=100) of the benchmark function for evaluating the Alzheimer's disease scale score prediction device based on image genome according to an embodiment of the present invention. DETAILED DESCRIPTION
[0060] 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. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0061] On the contrary, the present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention as defined by the claims. Further, in order to make the public have a better understanding of the present invention, some specific details are described in detail in the detailed description of the present invention below. Those skilled in the art can fully understand the present invention without the description of these details.
[0062] See also Figure 1 , is a schematic diagram of the structure of an Alzheimer's disease scale score prediction device based on image genome according to an embodiment of the present invention, comprising a data acquisition device 10, a data preprocessor 20, an MCPO module 30 and a prediction module 40 connected in sequence, wherein:
[0063] The data acquired by the data acquisition device 10 include magnetic resonance imaging images, single nucleotide polymorphism data and psychological test scores; in a specific embodiment, the patient's magnetic resonance imaging (MRI) and single nucleotide polymorphism (SNP), as well as subjective psychological test scores are acquired from the public database of the Alzheimer's Disease Neuroimaging Project (ADNI). The imaging genome information of a total of 695 subjects was retrieved, including 101 AD patients, 373 MCI patients and 221 healthy controls (CN).
[0064] The data preprocessor 20 normalizes, strips and segments the magnetic resonance imaging images, performs standard quality control, estimates missing values and digitally encodes the single nucleotide polymorphism data;
[0065] The MCPO module 30 integrates various data sources and updates the results;
[0066] The prediction module 40 is combined with the MCPO module 30 to obtain the optimal features.
[0067] The data preprocessor 20 processes the magnetic resonance imaging images using CAT12, including normalization, skull stripping, segmentation and other steps, and extracts the gray matter volumes of 160 regions for classification.
[0068] Data preprocessor 20 performed standard quality control requirements on the single nucleotide polymorphism data using PLINK. Subsequently, the SNP missing values were estimated using MaCH software, and the SNPs were recoded into discrete numbers (0 / 1 / 2) to represent the number of minor alleles. Finally, loci from the top 40 high-risk genes were selected for inclusion in the study.
[0069] The MCPO module 30 includes an initial submodule, a dynamic reverse submodule, a cycle reduction submodule and an improved defense submodule which are connected in sequence.
[0070] The initial submodule is initialized based on the good point set strategy, specifically:
[0071] Assume G s is the unit cube in s-dimensional Euclidean space, r∈G s ,
[0072]
[0073] Its deviation satisfies
[0074]
[0075] Where C(r,ε)n -1+ε is a constant that is only related to r and ε, P n (k) represents the set of optimal points, where r is the optimal point; represents the decimal part, and n represents the number of points;
[0076] make Mapping it to the search space is expressed as,
[0077]
[0078] where x i (j) is the jth dimension of the i-th crested porcupine, is its advantage, j and lb j denotes the upper and lower bounds of the j-th dimension.
[0079] The dynamic reverse submodule is used to further increase the population diversity and the number of operating individuals, and expand the algorithm search area. The specific mathematical model is as follows:
[0080] X d=X+c 1 ×(c 2 ×(ub+lb-X)-X) (4)
[0081] Where X d is the dynamic inverse solution, c 1 and c 2 is a random number between 0 and 1, and ub and lb are its upper and lower limits.
[0082] The loop reduction submodule is specifically expressed in a mathematical model as follows:
[0083]
[0084] Where N is the population size, N min is the minimum number of individuals in the newly generated population, so that the population size cannot be less than N min , T c Used to determine the number of loops, T is the maximum number of function evaluations.
[0085] The improved defense submodule includes four defense strategies, specifically:
[0086] The inverse cumulative distribution function of the Cauchy distribution improves the first defense strategy: The probability density function of the Cauchy distribution is
[0087]
[0088] where x 0 is the location parameter that defines the peak position of the distribution, is the scale parameter of the half width at half the maximum value, and the random variable X follows the Cauchy distribution as X~C(γ,x 0 ),γ=1,x 0 = 0 is called the standard Cauchy distribution, and its probability density function and corresponding cumulative distribution function are shown in formulas (7) and (8).
[0089]
[0090] In the case where the cumulative distribution function of the Cauchy distribution is calculated as an inverse function, the inverse cumulative distribution function of the Cauchy distribution is defined as,
[0091]
[0092] Where p = randn(1, dim), the first defense strategy is expressed as formula (10),
[0093]
[0094] where τ 1 is a random number based on normal distribution, τ 2 is a random number in the interval [0,1], is the optimal solution, is the position of the i-th individual at the t-th iteration;
[0095] Second defense strategy: Crested porcupines threaten predators by making noise. When a predator approaches, the crested porcupine's voice will become louder, represented by:
[0096]
[0097] where r1 and r2 are random integers between [1,N], τ 3 is a random number between [0,1], Represents the vector generated between the current individual and the randomly selected individual; binary vector Includes randomly generated 0s and 1s. When a cell in this vector includes 0, the predator stands still and does not move toward the crested porcupine. When a cell contains 1, it means that the predator may move closer to or away from the crested porcupine.
[0098] Third defense strategy: Entering the third area, the crested porcupine secretes a foul odor that spreads in the area around it to prevent predators from approaching it, represented by:
[0099]
[0100] The vector is used To simulate the three possible scenarios in this strategy, r3 is a parameter between [1, N] used to control the search direction, γ t and represent the defense factor and odor diffusion factor, respectively, as shown in formulas (13)-(15),
[0101]
[0102] represents the objective function value of the i-th individual at the t-th iteration, ∈ is a small number to avoid division by zero, is a vector containing randomly generated values between 0 and 1, and rand is a random variable between 0 and 1;
[0103] Tangential flight strategy improves the fourth defense strategy: The calculation formula of the tangential flight strategy is as follows,
[0104]
[0105] Where v = randn(1, dim), applying the tangent branch operator to the fourth defense strategy is expressed as:
[0106]
[0107] Where α is the convergence speed factor, τ 4 and τ 5 is a random value between [0,1], is the average force affecting the ith individual, m i is the mass of the ith individual, is a vector containing generated random values between 0 and 1.
[0108] The prediction module 40 combines feature selection and prediction, and calculates the score for the imaging gene features after the data preprocessor through the MCPO module. We incorporate the principle of sparse CCA into the fitness function and introduce sparsity penalties to limit the number of variables, allowing us to effectively utilize complementary information from different data modes, thereby encouraging the algorithm to select the features with the highest relevance to the target and improving the accuracy and robustness of the optimization process.
[0109] In the prediction module 40, the problem is modeled as maximizing modal correlation and minimizing error, and the fitness function is constructed as,
[0110] fitness=MSE(Z,Z′)+ω(-U′X′YV+‖U‖ 1 +‖V‖ 1 ) (20)
[0111] Among them, X and Y are image and gene features, U and V are their corresponding weight coefficients, Z and Z' are the actual scores and SVR predicted scores of the subjects, respectively. 1 is the l1 norm, ω is the regularization coefficient, and through formula (20), MCPO iteratively optimizes the weights of the features and finally selects the optimal features to input into SVR for prediction.
[0112] (1) MCPO performance evaluation
[0113] In order to evaluate the optimization ability and stability of the proposed MCPO algorithm, this chapter compares MCPO with other intelligent optimization algorithms, including CPO, Dung Beetle Optimizer (DBO), Harrier Hawk Optimizer (HHO), Particle Swarm Optimizer (PSO), Whale Optimizer (WOA) and Average-based Optimizer (SABO). Table 1 shows the specific parameter settings of each algorithm. In the experiment, 29 benchmark functions from CEC2017 are used to test the performance of MDBO.
[0114] Table 1 Algorithm parameter settings
[0115]
[0116] The population size of all algorithms is set to 30, and the number of iterations is 1000. In order to reduce the impact of random disturbances and enhance the persuasiveness of the experiment, each objective function is executed 30 times, and the average value (Avg), standard deviation / stability (Std) and optimal solution (Best) of each algorithm are statistically analyzed to reflect the optimization ability and stability of each algorithm. Tables 2 and 3 show the optimization effects of algorithms with significant differences when dim = 30 and dim = 100, respectively. Figure 2 and 3 The corresponding convergence curves are drawn respectively.
[0117] Table 2 Comparison results of MCPO and other algorithms at CEC2017 (dim=30)
[0118]
[0119]
[0120] Table 3 Comparison results of MCPO and other algorithms at CEC2017 (dim=100)
[0121]
[0122]
[0123] When dim=30, MCPO is effective in single-peak functions F5, F7, F8, F11, F12, F14, F15,
[0124] The performance is outstanding on F17-F19, F21-F23, F26 and F29. When dim = 100, MCPO shows similar results to dim = 30, but there are also some differences. Specifically, on F1, F7, F12,
[0125] It performs well on F13, F15, F17, F20-F21, F23, F24, F27, F29 and F30. In general, MCPO performs well in most cases, especially when dim=30. MCPO has good search ability, stability and convergence on the four types of functions.
[0126] (2) MCPO-SVR prediction and evaluation
[0127] In order to verify the performance of the model in score prediction, we learned the feature subset from the training set for test set validation and repeated it 30 times to reduce experimental bias. We selected 10 genes and imaging features for 10 predictions of Mini-Mental Exam Scale (MMSE) and Alzheimer's Disease Assessment Scale (ADAS). Table 4 shows the prediction results of MDBO and other heuristic algorithms and the correlation coefficients of the prediction results.
[0128] Table 4. Comparison of MCPO prediction results with other algorithms
[0129]
[0130]
[0131] When predicting the MMSE score, MCPO achieved the best results in terms of both mean and stability, and was second only to DBO in terms of the optimal solution. When predicting the ADAS score, MCPO achieved the optimal stability for ADAS11 and the optimal mean and stability for ADAS13. The original CPO algorithm achieved the minimum mean and the optimal solution for ADAS11 and the optimal solution for ADAS13. Generally speaking, MCPO showed high stability and accuracy in different evaluation indexes, providing strong support for the AD prediction evaluation score. Mini-Mental State Examination (MMSE): It is concise in content, short in measurement time, and easy to be accepted by the elderly. It is the most common scale in clinical practice at present. The score of MMSE is related to the level of education. If the score of illiterate people is ≤ 17 points, primary school degree ≤ 20 points, middle school degree ≤ 22 points, and university degree ≤ 23 points, it indicates cognitive function impairment.
[0132] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. An Alzheimer's disease scale score prediction device based on image genome, characterized in that: It includes a data acquisition device, a data preprocessor, a MCPO module and a prediction module connected in sequence, wherein: The data acquired by the data acquirer include magnetic resonance imaging images, single nucleotide polymorphism data and psychological test scores; The data preprocessor normalizes, strips and segments the magnetic resonance imaging images, performs standard quality control, estimates missing values and digitally encodes the single nucleotide polymorphism data; The MCPO module integrates multiple data sources and updates the results; The prediction module is combined with the MCPO module to obtain the optimal features.
2. The device for predicting Alzheimer's disease scale score based on image genome according to claim 1, characterized in that: The data preprocessor processes the magnetic resonance imaging images using CAT12.
3. The device for predicting Alzheimer's disease scale score based on image genome according to claim 1, characterized in that: The data preprocessor uses PLINK to perform standard quality control on the single nucleotide polymorphism data.
4. The device for predicting Alzheimer's disease scale score based on image genome according to claim 1, characterized in that: The MCPO module includes an initial submodule, a dynamic reverse submodule, a loop reduction submodule and an improved defense submodule which are connected in sequence.
5. The device for predicting Alzheimer's disease scale score based on image genome according to claim 4, characterized in that: The initial submodule is initialized based on the good point set strategy, specifically: Assume G s is the unit cube in s-dimensional Euclidean space, r∈G s , Its deviation satisfies Where C(r,ε)n -1+ε is a constant that is only related to r and ε, P n (k) represents the set of optimal points, where r is the optimal point; represents the decimal part, and n represents the number of points; make Mapping it to the search space is expressed as, where x i (j) is the jth dimension of the i-th crested porcupine, is its advantage, j and lb j denotes the upper and lower bounds of the j-th dimension.
6. The device for predicting Alzheimer's disease scale score based on image genome according to claim 5, characterized in that: The dynamic reverse submodule is specifically represented by a mathematical model as follows: X d =X+c1×(c2×(ub+lb-X)-X) (4) Where X d For dynamic reverse solution, c1 and c2 are random numbers between 0 and 1, and ub and lb are their upper and lower limits.
7. The device for predicting Alzheimer's disease scale score based on image genome according to claim 6, characterized in that: The cycle reduction submodule is specifically expressed by a mathematical model as follows: Where N is the population size, N min is the minimum number of individuals in the newly generated population, so that the population size cannot be less than N min , T c Used to determine the number of loops, T is the maximum number of function evaluations.
8. The device for predicting Alzheimer's disease scale score based on image genome according to claim 7, characterized in that: The improved defense submodule includes four defense strategies, specifically: The inverse cumulative distribution function of the Cauchy distribution improves the first defense strategy: The probability density function of the Cauchy distribution is Where x0 is the position parameter that defines the peak position of the distribution, and is the scale parameter of the half width at half the maximum value. The random variable X obeys the Cauchy distribution as X~C(γ, x0), and the special case of γ = 1 and x0 = 0 is called the standard Cauchy distribution. Its probability density function and corresponding cumulative distribution function are shown in formulas (7) and (8). In the case where the cumulative distribution function of the Cauchy distribution is calculated as an inverse function, the inverse cumulative distribution function of the Cauchy distribution is defined as, Where p = randn(1, dim), the first defense strategy is expressed as formula (10), Where τ1 is a random number based on normal distribution, τ2 is a random number in the interval [0, 1], is the optimal solution, is the position of the i-th individual at the t-th iteration; Second defense strategy: Crested porcupines threaten predators by making noise. When a predator approaches, the crested porcupine's voice will become louder, represented by: Where r1 and r2 are random integers between [1, N], τ3 is a random number between [0, 1], Represents the vector generated between the current individual and the randomly selected individual; binary vector Includes randomly generated 0s and 1s. When a cell in this vector includes 0, the predator stands still and does not move toward the crested porcupine. When a cell contains 1, it means that the predator may move closer to or away from the crested porcupine. Third defense strategy: Entering the third area, the crested porcupine secretes a foul odor that spreads in the area around it to prevent predators from approaching it, represented by: The vector is used To simulate the three possible scenarios in this strategy, r3 is a parameter between [1, N] used to control the search direction, γ t and represent the defense factor and odor diffusion factor, respectively, as shown in formulas (13)-(15), represents the objective function value of the i-th individual at the t-th iteration, ∈ is a small number to avoid division by zero, is a vector containing randomly generated values between 0 and 1, and rand is a random variable between 0 and 1; Tangential flight strategy improves the fourth defense strategy: The calculation formula of the tangential flight strategy is as follows, Where v = randn(1, dim), applying the tangent branch operator to the fourth defense strategy is expressed as: Where α is the convergence speed factor, τ4 and τ5 are random values between [0,1], is the average force affecting the ith individual, m i is the mass of the ith individual, is a vector containing generated random values between 0 and 1.
9. The device for predicting Alzheimer's disease scale score based on image genome according to claim 1, characterized in that: The prediction module combines feature selection and prediction, and calculates scores for the image gene features after the data preprocessor through the MCPO module.
10. The device for predicting Alzheimer's disease scale score based on image genome according to claim 1, characterized in that: In the prediction module, the problem is modeled as maximizing modal correlation and minimizing error, and the fitness function is constructed as: fitness=MSE(Z,Z′)+ω(-U′X′YV+‖U‖1+‖V‖1) (20) Among them, X, Y are image and gene features, respectively, U, V are their corresponding weight coefficients, Z, Z' are the subject's actual score and SVR predicted score, respectively, ||·||1 is the l1 norm, ω is the regularization coefficient, and through formula (20), MCPO iteratively optimizes the weight of the feature and finally selects the optimal feature to input into SVR for prediction.