Brain age estimation method and system based on multilayer weighted modeling
By employing multi-layer weighted modeling technology, core features are screened using three-dimensional brain MRI images, a hybrid function model is established, and weights are optimized. This solves the accuracy and stability problems of existing brain age prediction technologies and achieves efficient brain age assessment.
Patent Information
- Application Number
- CN202511391194.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-26
- Publication Date
- 2026-02-13
AI Technical Summary
Existing brain age prediction technologies suffer from problems such as low accuracy, poor diversity and stability of examination methods, long clinical examination time, and poor applicability, which limit their widespread application in clinical practice.
A multi-layer weighted modeling approach was adopted. Three-dimensional brain MRI images were acquired and preprocessed to screen core aging-related features, establish a mixture function model, and generate a global brain age prediction model through deep learning iteration optimization. The complex relationship between gender differences and physiological processes was considered, and the weight parameters were dynamically adjusted.
It improves the accuracy and stability of brain age prediction, reduces errors, enhances adaptability, and meets the needs of large-scale brain health screening and early diagnosis.
Smart Images
Figure CN121528558A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of brain age prediction technology, specifically relating to a brain age prediction method and system based on multi-layer weighted modeling. Background Technology
[0002] In the fields of neuroscience and medicine, brain age research is gradually becoming a key means to reveal the mechanisms of brain aging, assess the risk of neurodegenerative diseases, and explore new directions for delaying aging and preventing and treating brain diseases. Through brain age research, multi-dimensional data can be integrated to quantify the heterogeneity of brain aging rates, thereby providing quantitative standards for assessing the risk of neurodegenerative diseases. This research helps to promote the transformation of brain health examinations from traditional qualitative analysis to quantitative analysis, bringing new ideas and methods to clinical practice.
[0003] Currently, brain age research mainly relies on two approaches: data-driven methods and biomarker exploration.
[0004] In terms of data-driven methods, this approach leverages neuroimaging features such as multimodal MRI and fMRI, combined with advanced algorithms like random forests and deep neural networks to construct predictive models for calculating the brain age deviation (BrainAGE). This method fully utilizes the rich information contained in neuroimaging data, using complex algorithmic models to uncover the intrinsic relationship between brain structure and functional characteristics and brain age, providing an effective technical means for brain age prediction.
[0005] Biomarker research, on the other hand, starts at the molecular level, using cutting-edge technologies such as proteomics and metabolomics to screen for biomarkers associated with brain aging. For example, studies have confirmed an early link between abnormal lipid metabolism pathways and neurodegenerative diseases such as Parkinson's disease, providing a theoretical basis for assessing the risk of brain aging and neurodegenerative diseases by detecting specific biomarkers. However, while such methods have high scientific value, they face many challenges in practical application. Complex experimental procedures and high costs hinder their widespread clinical application, limiting their large-scale use in brain age research and disease diagnosis.
[0006] Despite advancements in existing brain age research techniques, numerous challenges remain in practical application. Current brain age prediction technologies generally suffer from low accuracy, potentially leading to biased assessments of brain aging and consequently affecting the accurate judgment of neurodegenerative disease risk. Poor diversity and stability of examination methods are also significant issues; different methods may yield inconsistent results, and the stability of the same method across different individuals or time points needs improvement, posing considerable challenges to clinical diagnosis and research. Furthermore, the lengthy clinical examination times and limited applicability restrict the widespread application of existing technologies in clinical practice, failing to meet the demands for large-scale brain health screening and early diagnosis. Summary of the Invention
[0007] To overcome the problems in the prior art, this invention proposes a brain age prediction method and system based on multi-layer weighted modeling.
[0008] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides a brain age prediction method based on multi-level weighted modeling, comprising the following steps: Step 100: Acquire three-dimensional brain MRI image data and perform preprocessing; Step 200: Based on the preprocessed three-dimensional brain MRI image data, core aging-related features are screened; Step 300: Based on the selected core aging-related features, establish a mixed function model of multiple nuclei and brain regions to fit the mapping relationship between the core aging-related features and physiological age; Step 400: Integrate the mixed function models of each nucleus and brain region, iteratively optimize them through deep learning, dynamically adjust the weight parameters, and generate a global brain age prediction model to predict brain age.
[0009] Further, in step 100, the preprocessing includes: registering the acquired three-dimensional brain MRI image data to the MNI152 standard template space, then performing brain structure segmentation and labeling brain region structures, and calculating the quantitative values of morphological indicators of each brain region.
[0010] Further, in step 200, based on the preprocessed three-dimensional brain MRI image data, core aging-related features are screened, including: The quantitative values of morphological indicators of each brain region were statistically analyzed separately for different genders to show the changes with age. Calculate the first derivative of the changes in morphological indicators of each brain region with physiological age, and retain the morphological indicators of brain regions that show a monotonically decreasing or increasing trend. Calculate the Pearson correlation coefficient between morphological indicators of each brain region and retain morphological indicators with correlation coefficients higher than a preset threshold. Sort by annual rate of change in descending order to generate a feature priority list, which is the core aging-related feature selected.
[0011] Further, in step 200, the first derivative of the morphological indicators of each brain region with physiological age is calculated, and morphological indicators of brain regions showing a monotonically decreasing or increasing trend are retained, including: Let the morphological indicators of the brain region be z Calculate the first derivatives of the changes in morphological indices of each brain region with physiological age: ; In the above formula, , , , , , This indicates parameters associated with that brain region; x Indicates physiological age.
[0012] Further, in step 300, based on the selected core aging-related features, a mixture function model is established to fit the mapping relationship between the core aging-related features and physiological age, including: Constructing core aging-related features y i With physiological age x The mixture function model is as follows: ; In the above formula, a, b, c, d, e, f, g This represents the fitted parameters.
[0013] Furthermore, in step 400, the mixed function models of various nuclei and brain regions are integrated, and the weight parameters are dynamically adjusted through deep learning iterative optimization to generate a global brain age prediction model, including: Based on feature correlation and data dispersion, the initial brain age contribution parameters of the mixed function model for each brain region are set; With the constraint that the sum of brain age contribution parameters is 1, the brain age contribution parameters of each brain region are adjusted to obtain the optimal value of brain age contribution parameters that is closest to the physiological age. By using deep learning for iterative optimization and employing the gradient descent algorithm with mean squared error as the loss function, and dynamically adjusting the weights through 5-fold cross-validation, a global prediction model incorporating the contributions of each brain region is constructed.
[0014] Secondly, a brain age prediction system based on multi-level weighted modeling is provided, including: The preprocessing module acquires and preprocesses three-dimensional brain MRI images. The feature selection module performs core aging-related feature selection based on preprocessed three-dimensional brain MRI image data. The mixed function model module establishes mixed function models of multiple nuclei and brain regions based on the selected core aging-related features, and fits the mapping relationship between the core aging-related features and physiological age; The global brain age prediction module integrates the mixed function models of various nuclei and brain regions, and dynamically adjusts the weight parameters through deep learning iteration optimization to generate a global brain age prediction model.
[0015] Compared with the prior art, the present invention has the following technical effects: (1) This invention separately analyzes the changes in morphological indicators of each brain region with age according to different genders, fully considering the differences between men and women in brain structure and aging process, avoiding feature confusion, and making the screening more targeted. At the same time, multi-dimensional feature screening retains indicators that show monotonic changes to reflect the stable process of brain aging, while selecting highly correlated indicators to reflect the close relationship with brain age, and then generates a priority list by sorting by annual change rate. This comprehensive and accurate feature screening method provides a reliable basis for the subsequent establishment of accurate brain age prediction models, improving the quality of brain age prediction from the source.
[0016] (2) When establishing the hybrid function model, this invention comprehensively considers multiple physiological processes, such as introducing an exponential term to simulate neuronal apoptosis, using a cubic polynomial to fit changes in different stages of the brain, and using linear and square root terms to model the reduction in the thickness of brain functional areas. This allows for a more accurate fit of the complex relationship between core aging-related features and physiological age. When constructing the global brain age prediction model, feature correlation and data dispersion are considered, and deep learning is used for iterative optimization. The gradient descent algorithm is adopted, with mean squared error (MSE) as the loss function. The weights are dynamically adjusted through 5-fold cross-validation, which improves the model's prediction accuracy. Attached Figure Description
[0017] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a schematic diagram of the process of the present invention; Figure 2 is a schematic diagram showing the monotonic increase of structural quantification values with age; Figure 3 is a schematic diagram showing that the structural quantification value decreases monotonically with age; Figure 4 illustrates that the structural quantification values do not exhibit monotonicity in the selected age range; Figure 5 shows a schematic diagram of the monotonic relationship between structural quantification values and age, but with a low linear correlation, and the Left_Lateral_Ventricle fitting parameters and p-values for women; Figure 6 Fitting parameters and p-values for male Left_Lateral_Ventricle; Figure 7 This is a schematic diagram of the structure of the present invention. Detailed Implementation
[0019] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the specific implementation methods, structures, features, and effects of the technical solutions proposed according to the present invention are described in detail below with reference to the accompanying drawings and preferred embodiments. Specific features, structures, or characteristics in one or more embodiments may be combined in any suitable form. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0020] In one embodiment of the present invention, refer to 1- Figure 7 This paper presents a brain age prediction method based on multi-level weighted modeling, which includes the following steps: Step 100: Acquire three-dimensional brain MRI image data and perform preprocessing; Step 200: Based on the preprocessed three-dimensional brain MRI image data, core aging-related features are screened; Step 300: Based on the selected core aging-related features, establish a mixed function model of multiple nuclei and brain regions to fit the mapping relationship between the core aging-related features and physiological age; Step 400: Integrate multiple nucleus and brain region mixed function models, iteratively optimize them through deep learning, use the gradient descent algorithm with mean squared error as the loss function, dynamically adjust the weight parameters through 5-fold cross-validation, and generate a global brain age prediction model to predict brain age.
[0021] The following is a detailed explanation of each of the above steps: Step 100: Acquire three-dimensional brain MRI image data and perform preprocessing.
[0022] As an example, step 100 specifically includes the following sub-steps: Step 110: Acquire three-dimensional brain MRI images.
[0023] Three-dimensional brain MRI images of the subjects were collected using mainstream clinical 3.0T and 1.5T MRI equipment (such as Siemens SKYRA) to acquire T1-weighted imaging data. The scanning parameters were fixed, with a slice thickness of 1 mm and a pixel pitch of less than 1.5 × 1.5 to ensure that the acquired image data had high resolution and quality, providing an accurate data basis for subsequent brain structure analysis.
[0024] Step 120: Preprocess the acquired three-dimensional brain MRI image data. The preprocessing includes data registration, brain structure segmentation and annotation, and quantization calculation.
[0025] The data registration process includes: registering the acquired 3D brain MRI images to the internationally used MNI152 standard template space based on multiple open-source graphics analysis libraries. This unifies the 3D brain MRI images of different individuals into the same standard template space, eliminating the influence of individual brain morphology and location differences on subsequent analysis, and enabling accurate comparison and analysis of brain structures between different individuals.
[0026] The brain structure segmentation and annotation process includes: segmenting the brain structure using spatial probe annotation and an improved 3D watershed algorithm that combines a thresholding algorithm with a 3D watershed algorithm, and then annotating the brain regions. The segmented brain structure clearly distinguishes different brain regions, providing a foundation for subsequent calculation of quantitative indicators for each brain region.
[0027] The improved 3D watershed algorithm combines thresholding and 3D watershed methods. For example, it binarizes brain images using the Otsu algorithm or adaptive thresholding to initially separate gray matter, white matter, and cerebrospinal fluid. The brain image is treated as a 3D terrain, and gradient magnitudes (such as the Sobel operator) are calculated. Regions with high gradients correspond to brain region boundaries ("peaks"), while regions with low gradients correspond to the interior of brain regions ("valleys"). Water is injected starting from local minima (valleys), and watersheds (boundaries) are constructed at the confluence of different water bodies as the water level rises. Combined with spatial probe annotation results, the segmentation boundaries are dynamically adjusted to ensure that the brain region division conforms to prior anatomical knowledge.
[0028] The quantitative value calculation includes: calculating the quantitative values of morphological indicators for each brain region based on the quantitative data of spatial voxels, as detailed in Tables 1-4. These morphological indicators include the volume of brain nuclei, the thickness of the cerebral cortex, and surface area. These quantitative indicators objectively reflect the size, shape, and other structural characteristics of the brain, providing important data for analyzing brain aging and neurodegenerative diseases. For example, the thickness of the cerebral cortex can be obtained by calculating the distance between different points on the surface of the cerebral cortex, while the volume of brain nuclei can be calculated by multiplying the number of voxels contained in the brain nuclei by the volume of a single voxel.
[0029] Table 1
[0030] Table 2
[0031] Table 3
[0032] Table 4
[0033] Step 200: Based on the preprocessed three-dimensional brain MRI images, core aging-related features are screened.
[0034] As an example, referring to Figures 2(a)(b), 3(a)(b), and 4(a)(b), this step 200 specifically includes the following sub-steps: Step 210: Analyze the changes in the quantitative values of morphological indicators of each brain region with age separately for different genders.
[0035] To focus on core aging-related characteristics, the data were divided into separate datasets for men and women. Then, the changes in quantitative values such as the volume, surface area, and thickness of each brain region with age were statistically analyzed separately for each gender.
[0036] Since men and women may have differences in brain structure and aging process, gender-based analysis can more accurately identify gender-related brain aging characteristics, improving the targeting and accuracy of feature selection.
[0037] Step 220: Perform global monotonicity analysis, calculate the first derivative of the morphological indicators of each brain region with physiological age, and retain the morphological indicators of brain regions that show a monotonically decreasing or increasing trend.
[0038] Let z be the morphological index of a brain region. Calculate the first derivative of the morphological index of each brain region as a function of physiological age: ; In the above formula, , , , , , This indicates parameters associated with that brain region; x Indicates physiological age.
[0039] Monotonic variations better reflect the stable process of brain aging, while fluctuating variations may be affected by various factors, making it difficult to accurately analyze the relationship between brain aging and age. Monotonicity selection can screen out more representative brain aging features.
[0040] Step 230: Calculate the Pearson correlation coefficient between morphological indicators of each brain region, and retain morphological indicators of brain regions with correlation coefficients higher than a preset threshold.
[0041] The Pearson correlation coefficient measures the degree of linear correlation between two variables. Suppose we have a dataset containing 50 samples. The calculated Pearson correlation coefficient is r = 0.75. In this study, morphological features of brain regions with correlation coefficients higher than 0.60 are retained, while those with low correlation are removed. Therefore, this brain region structural volume feature will be retained because it shows a strong positive linear correlation with age. Conversely, if the calculated Pearson correlation coefficient is r = 0.30, since 0.30 < 0.60, this feature will be removed because its linear correlation with age is weak, contributing little to the analysis of brain aging.
[0042] Step 240: Sort by annual rate of change in descending order to generate a feature priority list, which is the core aging-related features selected.
[0043] Step 300: Based on the selected core aging-related features, establish a mixed function model of multiple nuclei and brain regions to fit the mapping relationship between the core aging-related features and physiological age.
[0044] Hippocampal volume changes with age due to a variety of physiological processes. Neuronal apoptosis exhibits an exponential decay process, and an exponential term is introduced into the model to simulate this phenomenon. At the same time, considering the different stages of brain development, maturation, and aging, a cubic polynomial is used to fit the three-stage nonlinear changes. Processes such as the reduction of brain functional area thickness can be modeled using linear and square root terms.
[0045] Therefore, constructing core aging-related features y With physiological age x The mixed function model of nuclei and brain regions is as follows: ; In the above formula, a, b, c, d, e, f, g Indicates the fitted parameters; This represents an exponential decay process, simulating neuronal apoptosis, such as the exponential shrinkage of the hippocampus with age; a cubic polynomial. This represents the fitting of three-stage nonlinear changes in brain development (0-20 years), maturation (20-60 years), and aging (60+ years); linear and square root terms. This is used to model physiological processes such as the reduction in thickness and surface area (square root growth) of brain functional areas. Refer to Figures 5(a), (b), and (c) for details. Figure 6 .
[0046] Figure 5 shows the fitting parameters and p-values for the Left_Lateral_Ventricle female model. a = -0.0019059323699313069 b=0.01564288095882408, c = 109.9351451248012 d = -626.2242932976116 e = 5829.533347901055 p={0.7230121218451906}.
[0047] Figure 6 In the data, the Left_Lateral_Ventricle fitting parameters and p-values for males are as follows: a = -0.976199215842316 b = 0.026703207709571658 c = 25.068529806209376 d = 145.62804547161167 e = 5491.187383093582 p = {0.7619887281694626}.
[0048] The least squares method was used to fit the parameters. Collected hippocampal volume and physiological age data were substituted into the model, and the optimal parameters were determined by minimizing the sum of squared errors between the predicted and actual values. The Shapiro-Wilk test was used to ensure the residuals were normally distributed; if the p-value of the test was greater than the significance level, the hypothesis of a normally distributed residual distribution could not be rejected. An R² threshold was set to ensure the effectiveness of the fit.
[0049] Step 400: Integrate multiple nucleus and brain region mixed function models, iteratively optimize them through deep learning, dynamically adjust the weight parameters, and generate a global brain age prediction model.
[0050] As an example, step 400 specifically includes the following sub-steps: Step 410: Integrate the multiple hybrid function models established in step 300 and set the initial contribution parameters.
[0051] Based on the mixed function model of nuclei and brain regions, global feature fusion is achieved through two-layer weight allocation: the correlation between each brain region feature and physiological age is calculated, and the initial contribution parameter of the brain region with higher correlation should be set higher; the data dispersion (e.g., standard deviation) of each brain region feature is calculated, and the brain region with lower dispersion has higher data stability, and the initial contribution parameter can be appropriately increased.
[0052] Suppose there are three brain regions, denoted as region A, region B, and region C. First, calculate the correlation between the features of each brain region and physiological age, as well as the data dispersion. Correlation calculation: The correlation between brain region A and physiological age is 0.8, brain region B is 0.6, and brain region C is 0.4. Data dispersion calculation: The standard deviation of brain region A is 0.2, brain region B is 0.3, and brain region C is 0.4. Considering both correlation and data dispersion, the initial contribution parameters can be set as follows: the initial contribution parameter for brain region A is 0.5, the initial contribution parameter for brain region B is 0.3, and the initial contribution parameter for brain region C is 0.2.
[0053] With the constraint that the sum of brain age contribution parameters is 1, the contribution parameters of each brain region are initially set as described above. The gradient descent algorithm is used with mean squared error (MSE) as the loss function, and the weights are dynamically adjusted through 5-fold cross-validation. In each iteration, the contribution parameters of each brain region are adjusted according to the gradient direction of the loss function, while ensuring that the sum of the parameters is 1. After updating the parameters, the parameters need to be normalized to ensure that the sum of the parameters is 1. When the loss function converges or reaches the preset number of iterations, the iteration stops, and the optimal contribution parameters are obtained.
[0054] Through the above iterative optimization process, the optimal contribution parameters for each brain region were obtained. These parameters were then substituted into the global prediction model: ; Each of them The prediction results for a specific brain region are weighted and fused to obtain the final prediction result.
[0055] Internal validation was conducted using a sample of 1459 individuals across age groups (5-100 years old, including 751 women and 708 men), split into training and test sets at an 8:2 ratio. Results showed a mean absolute error (MAE) ≤ 3.2 years, a root mean square error (RMSE) ≤ 4.1 years, and a coefficient of determination (R²) ≥ 0.89, demonstrating superior accuracy compared to similar methods. See Tables 5 and 6 for details.
[0056] Validated on the UK Biobank dataset (n=500, GE device), the MAE was 3.5 years, reducing the generalization error by 20% compared to traditional methods, demonstrating cross-device data adaptability. In clinical applications, the system inputs the brain structural characteristics of the subject, and outputs the difference between the estimated brain age and physiological age within 30 minutes, providing doctors with an intuitive basis for assessing the degree of brain aging.
[0057] Table 5 Results of Brain Age Test in Normal Individuals
[0058] Table 6. Brain age test results of AD patients
[0059] Based on the same inventive concept, this application also provides a brain age prediction system based on multi-level weighted modeling for implementing the aforementioned brain age prediction method. The solution provided by this system is similar to the implementation described in the above method; therefore, the specific limitations in one or more system embodiments provided below can be found in the limitations of the brain age prediction method based on multi-level weighted modeling described above, and will not be repeated here.
[0060] In one embodiment, a brain age prediction system based on multi-level weighted modeling is provided, comprising: The preprocessing module acquires and preprocesses three-dimensional brain MRI images. The feature selection module performs core aging-related feature selection based on preprocessed three-dimensional brain MRI image data. The mixed function model module establishes mixed function models of multiple nuclei and brain regions based on the selected core aging-related features, and fits the mapping relationship between the core aging-related features and physiological age; The global brain age prediction module integrates the mixed function models of various nuclei and brain regions, and dynamically adjusts the weight parameters through deep learning iteration optimization to generate a global brain age prediction model.
[0061] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.
Claims
1. A brain age prediction method based on multi-level weighted modeling, characterized in that, Includes the following steps: Step 100: Acquire three-dimensional brain MRI image data and perform preprocessing; Step 200: Based on the preprocessed three-dimensional brain MRI image data, core aging-related features are screened; Step 300: Based on the selected core aging-related features, establish a mixed function model of multiple nuclei and brain regions to fit the mapping relationship between the core aging-related features and physiological age; Step 400: Integrate the mixed function models of each nucleus and brain region, iteratively optimize them through deep learning, dynamically adjust the weight parameters, and generate a global brain age prediction model to predict brain age.
2. The brain age prediction method based on multi-level weighted modeling according to claim 1, characterized in that, In step 100, the preprocessing includes: registering the acquired three-dimensional brain MRI image data to the MNI152 standard template space, then performing brain structure segmentation and labeling brain region structures, and calculating the quantitative values of morphological indicators of each brain region.
3. The brain age prediction method based on multi-level weighted modeling according to claim 2, characterized in that, In step 200, based on the preprocessed three-dimensional brain MRI image data, core aging-related features are screened, including: The quantitative values of morphological indicators of each brain region were statistically analyzed separately for different genders to show the changes with age. Calculate the first derivative of the changes in morphological indicators of each brain region with physiological age, and retain the morphological indicators of brain regions that show a monotonically decreasing or increasing trend. Calculate the Pearson correlation coefficient between morphological indicators of each brain region and retain morphological indicators with correlation coefficients higher than a preset threshold. Sort by annual rate of change in descending order to generate a feature priority list, which is the core aging-related feature selected.
4. The brain age prediction method based on multi-level weighted modeling according to claim 3, characterized in that, In step 200, the first derivative of the morphological indices of each brain region with physiological age is calculated, and morphological indices of brain regions showing a monotonically decreasing or increasing trend are retained, including: Let the morphological indicators of the brain region be z i Calculate the first derivatives of the changes in morphological indices of each brain region with physiological age: ； In the above formula, , , , , , Indicates the fitted parameters; x Indicates physiological age.
5. The brain age prediction method based on multi-level weighted modeling according to claim 3, characterized in that, In step 300, based on the selected core aging-related features, a mixed function model of brain nuclei and regions is established to fit the mapping relationship between the core aging-related features and physiological age, including: Constructing core aging-related features With physiological age x The mixture function model is as follows: ； In the above formula, a, b, c, d, e, f, g This represents the fitted parameters.
6. The brain age prediction method based on multi-level weighted modeling according to claim 1, characterized in that, In step 400, the mixed function models of various nuclei and brain regions are integrated, and the weight parameters are dynamically adjusted through deep learning iterations to generate a global brain age prediction model, including: Based on feature correlation and data dispersion, the initial brain age contribution parameters of the mixed function model for each brain region are set; With the constraint that the sum of brain age contribution parameters is 1, the brain age contribution parameters of each brain region are adjusted to obtain the optimal value of brain age contribution parameters that is closest to the physiological age. By using deep learning for iterative optimization and employing the gradient descent algorithm with mean squared error as the loss function, and dynamically adjusting the weight parameters through 5-fold cross-validation, a global prediction model incorporating the contributions of each brain region is constructed.
7. A brain age prediction system based on multi-level weighted modeling, characterized in that, include: The preprocessing module acquires and preprocesses three-dimensional brain MRI images. The feature selection module performs core aging-related feature selection based on preprocessed three-dimensional brain MRI image data. The mixed function model module establishes mixed function models of multiple nuclei and brain regions based on the selected core aging-related features, and fits the mapping relationship between the core aging-related features and physiological age. The global brain age prediction module integrates the mixed function models of various nuclei and brain regions, and dynamically adjusts the weight parameters through deep learning iteration optimization to generate a global brain age prediction model.