An alzheimer's disease auxiliary screening method based on whole brain spatial variation pattern

Abstract

The application provides an Alzheimer's disease auxiliary screening method based on whole brain spatial variation patterns, and belongs to the technical field of medical image data analysis. The method continuously calculates the spatial variation values of each ROI in the time dimension, constructs a spatial variation time sequence reflecting the time evolution law of the spatial heterogeneity in the brain region, so as to realize the expansion of the brain function spatial structure from “static description” to “dynamic process modeling”. On this basis, the spatial variation time sequence is decomposed in multiple time scales through a sliding time window mechanism, and a spatial heterogeneity dynamic feature capable of describing the whole brain spatial heterogeneity is constructed. The spatial heterogeneity dynamic feature is input into a classification model constructed based on supervised learning, so that the model can learn the nonlinear mapping relationship between the brain function spatial heterogeneity pattern and the clinical state of Alzheimer's disease, thereby outputting a continuous risk score and realizing the quantitative prediction from the image feature to the disease risk.

Description

Technical Field

[0001] This invention belongs to the field of medical image data analysis technology, specifically relating to an auxiliary screening method for Alzheimer's disease based on whole-brain spatial variation patterns. Background Technology

[0002] Alzheimer's disease (AD) is a common neurodegenerative disease, with main clinical manifestations including memory impairment, cognitive decline, and behavioral changes. With the increasing aging of the global population, the prevalence of Alzheimer's disease is rising year by year, making early screening and risk assessment of significant clinical importance.

[0003] Functional magnetic resonance imaging (fMRI) can non-invasively record changes in brain functional activity and has been widely used in Alzheimer's disease screening research. Existing fMRI-based disease identification methods mainly include functional connectivity analysis, local consistency analysis, low-frequency amplitude analysis, and brain network topology feature analysis. These methods extract brain functional features and combine them with machine learning models to classify and identify healthy individuals and Alzheimer's patients. However, most of these methods, which analyze average signals from brain regions or statistical relationships between brain regions, have certain limitations. First, averaging brain regions converges functional signals from different voxels within a brain region, potentially causing positive and negative changes to cancel each other out, thus weakening the expressive power of disease-related signals. Second, during feature construction, the voxel signals within a brain region are often compressed by averaging, resulting in the neglect of a large amount of spatial distribution information at the voxel level, making it difficult to reflect the fine-grained differences in brain functional activity in the spatial dimension. Furthermore, the above analytical methods typically assume that there are relatively consistent functional activity patterns within the same brain region. However, numerous neuroimaging studies have shown that brain functional activity is not uniformly distributed spatially; even within the same brain region, significant functional heterogeneity can exist between different voxels. This spatial heterogeneity not only reflects the complexity of brain functional organization but is also closely related to individual cognitive states, neurodegenerative pathological processes, and the degree of disease progression. Especially in the early stages of Alzheimer's disease, pathological changes often initially manifest as abnormalities in local brain functional organization patterns, rather than necessarily causing significant changes in the average signal level of the brain region. Therefore, traditional averaging features may be insufficient to capture such early abnormal information.

[0004] Therefore, the key is to design an auxiliary screening method for Alzheimer's disease risk assessment that can make full use of the spatial distribution information of voxels within brain regions to achieve accurate assessment and automated auxiliary screening of Alzheimer's disease risk. Summary of the Invention

[0005] To address the problem that existing Alzheimer's disease screening methods based on functional connectivity or local activity indicators fail to fully utilize spatial distribution information of brain functional activity, resulting in insufficient expression of disease-related features, this invention provides an Alzheimer's disease screening method based on whole-brain spatial variation patterns. By constructing an analytical model that reflects the spatial variation characteristics of brain function, it achieves automated assessment of Alzheimer's disease risk.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] An Alzheimer's disease screening method based on whole-brain spatial variation patterns includes the following steps:

[0008] S1: Acquire resting-state functional magnetic resonance imaging data of the subject;

[0009] S2: Preprocessing resting-state functional magnetic resonance imaging data to obtain standardized brain functional images;

[0010] S3: Constructing a whole-brain spatial variation feature vector based on standardized brain functional imaging data. The specific process is as follows:

[0011] S31. Based on a preset brain function partitioning template, voxels are divided into regions within a standardized brain function image space constrained by a brain gray matter mask to obtain several regions of interest (ROIs).

[0012] S32. For any ROI, at time point t, extract the set of functional signals of all voxels within that ROI, denoted as . ,in, It is the functional signal value of the Nth voxel within the rth ROI at time point t;

[0013] The standard deviation is calculated based on the functional signal set and used as a characterization value for spatial variability. ;

[0014] S33. Repeat the calculation process in S32 for all time points. The set of spatial variation characteristics at all time points constitutes the ROI-level spatial variation time series, denoted as... ;

[0015] S4: Based on the ROI-level spatial variation time series, solve for its spatial variation mean, spatial variation standard deviation and spatial variation dynamic fluctuation index, and concatenate the three to obtain the spatial variation feature vector corresponding to each ROI;

[0016] Then, the feature vectors of all ROIs are concatenated to obtain the whole-brain spatial variation feature vector, which is used as the input of the Alzheimer's disease risk prediction model.

[0017] S5: Constructing an Alzheimer's disease risk prediction model:

[0018] The whole-brain spatial variation feature vector corresponding to each subject is used as the model input feature, and the clinical diagnosis result of the subject is used as the supervised learning label; the subjects should include a healthy control group and an Alzheimer's disease patient group;

[0019] A classification model was constructed using machine learning algorithms to learn the mapping relationship between spatial variability features and Alzheimer's disease;

[0020] When the model training error converges or reaches the preset maximum number of iterations, training is stopped, and a well-trained Alzheimer's disease risk prediction model is obtained.

[0021] S6: Input the whole-brain spatial variation feature vector obtained after processing the subject according to S1 to S4 into the Alzheimer's disease risk prediction model constructed in S5 to obtain the Alzheimer's disease risk score, and output the auxiliary screening results according to the preset threshold.

[0022] Furthermore, the resting-state functional magnetic resonance imaging data in S1 are brain functional images collected when the subject is awake and not performing a specific task, which are used to reflect the spatiotemporal changes in brain functional activity.

[0023] Resting-state functional magnetic resonance imaging (fMRI) data are obtained from clinically acquired data or publicly available neuroimaging databases. Clinical acquisition can be performed using magnetic resonance imaging (MRI) equipment, and the scanning sequence uses echo planar imaging (EPI) sequences. Brain imaging data at multiple time points are usually obtained during the scanning process, with each time point corresponding to a set of three-dimensional brain voxel signals.

[0024] Furthermore, the preprocessing in S2 adopts the functional magnetic resonance imaging data processing workflow commonly used in the field to reduce the impact of acquisition noise and individual differences on subsequent analysis, including sequential temporal correction, head motion correction, spatial normalization, noise signal regression and spatial smoothing.

[0025] Furthermore, the formula for calculating the spatial variation value in S3 is as follows:

[0026] ,

[0027] In the formula, This represents the spatial variation value of the r-th ROI at time point t; This represents the functional signal value of the i-th voxel within the r-th ROI at time t; N represents the average signal value of the r-th ROI at time point t; r This represents the total number of voxels contained in the r-th ROI.

[0028] Furthermore, the formula for calculating the mean of spatial variation is:

[0029] ,

[0030] in, Let represent the mean spatial variation of the r-th ROI; T represents the total number of time points.

[0031] The formula for calculating the standard deviation of spatial variation is:

[0032] ,

[0033] in, This represents the standard deviation of spatial variation for the r-th ROI;

[0034] The calculation process for the spatial variation dynamic fluctuation index is as follows:

[0035] Spatial variation time series for each ROI The time window is segmented according to a preset time window length W and a sliding step size S, resulting in K time windows, each corresponding to a spatial variant sequence. Where k = 1,2,…,K; the variance of the spatially variable subsequence within each time window is calculated to obtain the dynamic volatility index, the formula of which is: ,

[0036] in, Represents the dynamic spatial variation index of the r-th ROI; This indicates that the r-th ROI is at the k-th time.

[0037] The spatial variant sequence within the sliding window; Var(·) represents the variance operation; K represents the number of sliding time windows.

[0038] Furthermore, the machine learning algorithms in S5 include, but are not limited to, support vector machines, random forests, logistic regression, or gradient boosting trees.

[0039] Furthermore, the auxiliary screening results include normal, mild cognitive impairment risk, and high risk of Alzheimer's disease; when the risk score is below the first threshold, it is judged as normal; when the risk score is between the first threshold and the second threshold, it is judged as mild cognitive impairment risk; when the risk score is above the second threshold, it is judged as high risk of Alzheimer's disease.

[0040] The first threshold and the second threshold are determined based on the receiver operating characteristic (ROC) curve of the training set.

[0041] Furthermore, the specific process for determining the first threshold and the second threshold is as follows:

[0042] The continuous risk score output by the classification model is between 0 and 1. Within this range, multiple point values ​​are obtained by dividing the range with a fixed step size. Any two point values ​​are used as a candidate threshold. Sensitivity and specificity are calculated for each candidate threshold condition, and the corresponding Youden index J is further calculated: J = Sensitivity + Specificity − 1. The candidate threshold that maximizes the Youden index is selected as the optimal classification threshold. The two point values ​​in this candidate threshold are the first threshold and the second threshold, respectively. Sensitivity represents the model's ability to correctly identify diseased subjects, and its calculation formula is: Specificity represents the model's ability to correctly identify healthy subjects, and its calculation formula is: Where: TP (True Positive) represents the number of true positives, that is, the number of samples that are actually disease subjects and are correctly predicted by the model as disease subjects; TN (True Negative) represents the number of true negatives, that is, the number of samples that are actually healthy subjects and are correctly predicted by the model as healthy subjects; FP (False Positive) represents the number of false positives, that is, the number of samples that are actually healthy subjects but are incorrectly predicted by the model as disease subjects; FN (False Negative) represents the number of false negatives, that is, the number of samples that are actually disease subjects but are incorrectly predicted by the model as healthy subjects.

[0043] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0044] This invention utilizes whole-brain spatial variation analysis of resting-state brain functional imaging data to construct characteristic indicators that reflect the spatial heterogeneity of brain function. Compared to traditional functional connectivity or local signal indicators, these indicators can more comprehensively characterize the spatial variation patterns of brain functional activity, thereby improving the sensitivity of detecting early functional abnormalities in Alzheimer's disease. Furthermore, this invention combines machine learning methods to construct a disease risk prediction model, achieving automated assessment of Alzheimer's disease risk, improving the accuracy and stability of assisted screening, and providing reliable technical support for early clinical diagnosis and intervention. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the Schaefer 100 brain region template.

[0046] Figure 2 This is a schematic diagram of the ROI-level spatial variation time series obtained in Example 1.

[0047] Figure 3 This is a brain region distribution map of the set of variant features of one of the subjects in Example 1.

[0048] Figure 4 This is a schematic diagram of the prediction model construction in Example 1.

[0049] Figure 5 The graph shows the prediction performance results of the prediction model in Example 1. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings.

[0051] This invention, based on resting-state functional magnetic resonance imaging (fMRI) data, divides the entire brain into Regions of Interest (ROIs) using brain region partitioning templates under spatial constraints of brain gray matter. Within each ROI, the spatial dispersion of voxel-level functional signals is quantified to obtain spatial variability values ​​characterizing the spatial heterogeneity within local brain regions. Unlike existing methods that characterize brain function features only at a single time point or based on average brain region signals, this invention continuously calculates the spatial variability values ​​of each ROI over time, constructing a spatial variability time series reflecting the evolution of spatial heterogeneity within brain regions over time. This expands the understanding of brain functional spatial structure from "static description" to "dynamic process modeling." Furthermore, this invention uses a sliding time window mechanism to decompose the spatial variability time series across multiple time scales, extracting multidimensional features reflecting overall levels, local fluctuations, and dynamic trends. This constructs a feature system capable of characterizing the dynamic patterns of whole-brain spatial heterogeneity, achieving a refined representation of the complex changes in brain functional spatial structure.

[0052] This invention inputs the aforementioned spatial variation dynamic features into a classification model constructed based on supervised learning, enabling the model to learn the nonlinear mapping relationship between the spatial heterogeneity pattern of brain function and the clinical state of Alzheimer's disease, thereby outputting a continuous risk score and realizing quantitative prediction of disease risk from imaging features.

[0053] Finally, by determining the optimal classification threshold based on receiver operating characteristic (ROC) curve analysis, adaptive stratification of continuous risk scores is achieved, classifying subjects into normal, mild cognitive impairment risk, and high risk of Alzheimer's disease, thereby constructing a complete automated assisted screening mechanism for Alzheimer's disease.

[0054] Example 1

[0055] An Alzheimer's disease screening method based on whole-brain spatial variation patterns includes the following steps:

[0056] S1: Acquire resting-state functional magnetic resonance imaging (fMRI) data of the subject. The resting-state fMRI data refers to the spontaneous neural activity signals of the brain in the resting state of the subject with eyes closed or fixated on a fixed point in the magnetic resonance scanner, without performing any specific cognitive tasks. This data is used to reflect the spatiotemporal changes in brain functional activity. The data is acquired using a 3.0T magnetic resonance scanning device, and the scanning sequence uses echo planar imaging (EPI) sequence. Brain imaging data at multiple time points are usually obtained during the scanning process, and each time point corresponds to a set of three-dimensional brain voxel signals.

[0057] S2: Preprocessing resting-state functional magnetic resonance imaging data to eliminate scanning noise and individual differences, improve data consistency, and obtain standardized brain functional imaging data;

[0058] Preprocessing steps can be implemented using existing neuroimaging software, such as SPM, DPABI, FSL, or AFNI. Specific preprocessing procedures include, but are not limited to, the following steps: First, slice timing correction is performed to eliminate the impact of differences in acquisition time between different slices on the signal; then, motion correction is performed, using rigid body transformation methods to correct for minor head movements generated by the subject during the scan; next, spatial normalization is performed, mapping individual brain images to a unified standard brain space, such as the MNI standard space, to facilitate comparisons between different individuals; then, noise signal regression is performed, including removing white matter signals, cerebrospinal fluid signals, and motion parameter-related signals; finally, spatial smoothing is performed, smoothing the images using Gaussian kernels to improve the signal-to-noise ratio and spatial consistency.

[0059] After the above preprocessing, standardized brain functional imaging data can be obtained for subsequent analysis.

[0060] S3: Constructing a whole-brain spatial variation representation vector based on standardized brain functional imaging data. The specific process is as follows:

[0061] S31. Based on the brain functional partitioning template, the standardized brain functional imaging data is divided into several Regions of Interest (ROIs) to reduce the influence of non-neural tissue signals on spatial variation calculations. The brain functional partitioning template can be the Schaefer brain functional partitioning template. In this embodiment, the Schaefer 100 template is used to divide the whole brain into 100 Regions of Interest (ROIs). A schematic diagram of the Schaefer 100 brain partitioning template is shown below. Figure 1 As shown;

[0062] S32. For any ROI, extract the functional signal values ​​of all voxels within the ROI at time point t to obtain the functional signal set, denoted as . ,in, It is the functional signal value of the Nth voxel within the rth ROI at time point t;

[0063] The standard deviation of the functional signal set based on voxels is used as the spatial variability value. This is used to reflect the overall degree of variation of the ROI in the spatial dimension at time point t;

[0064] The formula for calculating spatial variation is:

[0065] ,

[0066] In the formula, The expression represents the spatial variation value of the r-th ROI at time point t; This represents the functional signal value of the i-th voxel within the r-th ROI at time t; This represents the average signal value of the r-th ROI at time point t; N represents the number of voxels contained in the r-th ROI.

[0067] S33. Repeat the calculation process in S32 for all time points, and arrange the spatial variation values ​​of all time points in chronological order to obtain the ROI-level spatial variation time series. That is, each ROI obtains a spatial variation representation sequence that changes over time at all time points, which is used to describe the dynamic change pattern of spatial functional distribution within different brain regions of the whole brain over time. A schematic diagram of the time series of spatial variation at the ROI level is shown below. Figure 2 As shown;

[0068] S4. Based on the ROI-level spatial variation time series, solve for its spatial variation mean, spatial variation standard deviation and spatial variation dynamic fluctuation index, and concatenate the three to obtain the spatial variation feature vector corresponding to each ROI. The spatial variation feature vector is a 1×3-dimensional vector used to characterize the spatial dynamic characteristics of brain function.

[0069] The feature vectors of all ROIs are concatenated to form a high-dimensional whole-brain spatial variation feature vector that characterizes the whole-brain spatial variation pattern, and this vector is used as the input to the Alzheimer's disease risk prediction model.

[0070] Unlike traditional functional connectivity analysis methods, this invention analyzes the spatial dispersion of voxel signals within a region of interest (ROI) in the spatial dimension, which can characterize the spatial organization features and dynamic changes of brain functional activities at the brain region level.

[0071] The mean spatial variation reflects the average level of spatial variation of the corresponding ROI throughout the entire scanning process, and its calculation formula is as follows:

[0072] ,

[0073] in, Let represent the mean spatial variation of the r-th ROI; T represents the total number of time points.

[0074] The standard deviation of spatial variation is used to reflect the range of fluctuation of the spatial variation of a corresponding ROI over time. Its calculation formula is...

[0075] for: ,

[0076] in, This represents the standard deviation of spatial variation for the r-th ROI;

[0077] The spatial variability dynamic fluctuation index is used to characterize the dynamic change magnitude of the spatial variability value of ROI in a time series; specifically, for each ROI, the spatial variability time series... A sliding time window is used to construct spatially variant subsequences over local time periods;

[0078] The spatially mutated time series is segmented according to a preset time window length W and a sliding step size S to obtain K time windows, each time window corresponding to a spatially mutated subsequence. Where k = 1,2,…,K; the variance of the spatially variable subsequence within each time window is calculated to characterize the volatility intensity within a local time period, thus obtaining the dynamic volatility index, the formula of which is: ,

[0079] in, Represents the dynamic spatial variation index of the r-th ROI; This indicates that the r-th ROI is at the k-th time.

[0080] Spatial variation sequence within the sliding window; Var(·) represents variance operation; K represents the number of sliding time windows;

[0081] One set of variability features for a subject includes the spatial variability mean, spatial variability standard deviation, spatial variability dynamic fluctuation index, and brain region distribution map, as shown in the figure. Figure 3 As shown;

[0082] S5: Using a set of spatial variation features with known diagnostic labels as training data, an Alzheimer's disease risk prediction model is constructed. In this embodiment, 100 subjects each of Alzheimer's disease (AD), mild cognitive impairment (MCI), and normal control (HC) are selected from the public dataset OASIS-3 as experimental data sources.

[0083] The prediction model is trained using a machine learning algorithm (support vector machine), and the trained Alzheimer's disease risk prediction model is obtained after training.

[0084] During model training, cross-validation can be used to evaluate model performance in order to improve model stability and generalization ability.

[0085] A schematic diagram of the constructed prediction model is shown below. Figure 4 As shown;

[0086] To verify the effectiveness of the risk prediction model constructed in this invention, this embodiment selects independent test samples from the ADNI (Alzheimer's Disease Neuroimaging Initiative) dataset as subjects to be tested, including 30 subjects with Alzheimer's disease (AD), 30 subjects with mild cognitive impairment (MCI), and 30 healthy controls (HC), for a total of 90 subjects.

[0087] Specifically, firstly, resting-state functional magnetic resonance imaging (fMRI) data of the aforementioned 90 subjects were collected according to steps S1 to S4.

[0088] The data are processed to obtain the corresponding ROI-level spatial variation feature set; the spatial variation feature set includes features such as the mean spatial variation, standard deviation of spatial variation, and dynamic fluctuation index of spatial variation for each ROI.

[0089] Subsequently, the set of spatial variation features of the subjects to be tested is input into the Alzheimer's disease risk prediction dataset trained in step S5.

[0090] In the testing model, a corresponding disease risk score is calculated; the risk score is used to characterize the probability or risk level of the subject belonging to an Alzheimer's disease-related pathological state.

[0091] In this embodiment, the risk classification threshold is determined based on the ROC curve analysis results of the training set, wherein the ROC curve...

[0092] Schematic diagram as follows Figure 5As shown in the figure, AUC represents the area under the curve, which serves as an indicator of model reliability. The point with the largest Youden index is used as the optimal threshold. Let the first risk threshold be T1 and the second risk threshold be T2, satisfying T1 < T2. Then, based on the risk score, subjects are divided into different risk levels:

[0093] (1) When the risk score is less than T1, it is judged as normal risk (HC);

[0094] (2) When the risk score is greater than or equal to T1 and less than T2, it is determined to be at risk of mild cognitive impairment (MCI).

[0095] (3) When the risk score is greater than or equal to T2, it is determined to be high risk of Alzheimer's disease (AD).

[0096] The prediction results of 90 independent test subjects were statistically analyzed, and the accuracy, sensitivity, specificity, AUC value and F1-score of the model were calculated to evaluate the disease screening performance of the method proposed in this invention.

[0097] In this embodiment, the trained risk prediction model was validated using 30 independent test samples each from HC, MCI and AD from the ADNI dataset. The results are shown in Table 1.

[0098] Table 1

[0099]

[0100] As can be seen from Table 1, the method of the present invention can classify subjects with relatively high accuracy.

[0101] The risk prediction model trained in this application was evaluated using the metrics shown in Table 2. The evaluation metrics included accuracy, sensitivity, specificity, area under the receiver operating characteristic curve (AUC), and F1 score. The evaluation results are shown in Table 2.

[0102] Table 2

[0103]

[0104] As shown in Table 2, the risk prediction model constructed in this application achieved good classification performance. Specifically, the model accuracy reached 84.4%, indicating that the model can accurately distinguish subjects with different disease states; the sensitivity was 86.7%, indicating that the model can effectively identify subjects with Alzheimer's disease and mild cognitive impairment with a low false negative rate; and the specificity was 80.0%, indicating that the model has good identification ability for healthy subjects with a low false negative rate. Meanwhile, the model's AUC reached 0.892, indicating that the model has strong overall discriminative ability and can effectively distinguish subjects with different risk levels. The F1 score was 0.842, further indicating that the model maintained a good balance between accuracy and recall.

[0105] The above results indicate that the Alzheimer's disease auxiliary screening method based on ROI spatial variation characteristics proposed in this invention can effectively distinguish between HC, MCI and AD subjects, and has good generalization ability and predictive performance.

[0106] The above description is merely a specific embodiment of the present invention. Any feature disclosed in this specification may be replaced by other equivalent or similar features unless otherwise specified. All disclosed features, or steps in all methods or processes, may be combined in any way except for mutually exclusive features and / or steps.

Claims

1. An Alzheimer's disease auxiliary screening method based on whole brain spatial variation pattern, characterized in that, Includes the following steps: S1: Acquire resting-state functional magnetic resonance imaging data of the subject; S2: Preprocessing resting-state functional magnetic resonance imaging data to obtain standardized brain functional images; S3: Constructing a whole-brain spatial variation feature vector based on standardized brain functional imaging data. The specific process is as follows: S31. Based on a preset brain function partitioning template, voxels are divided into regions within a standardized brain function image space constrained by a brain gray matter mask to obtain several regions of interest (ROIs). S32. For any ROI, extract the set of functional signals of all voxels within the ROI at time point t, denoted as wherein, is the functional signal value of the Nth voxel within the rth ROI at time point t. calculating standard deviation based on the functional signal set as spatial variation representation value ; S33. Repeat the calculation process in S32 for all time points. The set of spatial variation characteristics at all time points constitutes the ROI-level spatial variation time series, denoted as... ; S4: Based on the ROI-level spatial variation time series, solve for its spatial variation mean, spatial variation standard deviation and spatial variation dynamic fluctuation index, and concatenate the three to obtain the spatial variation feature vector corresponding to each ROI; Then, the feature vectors of all ROIs are concatenated to obtain the whole-brain spatial variation feature vector, which is used as the input of the Alzheimer's disease risk prediction model. S5: Constructing an Alzheimer's disease risk prediction model: The whole-brain spatial variation feature vector corresponding to each subject is used as the model input feature, and the clinical diagnosis result of the subject is used as the supervised learning label; the subjects should include a healthy control group and an Alzheimer's disease patient group; A classification model was constructed using machine learning algorithms to learn the mapping relationship between spatial variability features and Alzheimer's disease; When the model training error converges or reaches the preset maximum number of iterations, training is stopped, and a well-trained Alzheimer's disease risk prediction model is obtained. S6: Input the whole-brain spatial variation feature vector obtained after processing the subject according to S1 to S4 into the Alzheimer's disease risk prediction model constructed in S5 to obtain the Alzheimer's disease risk score, and output the auxiliary screening results according to the preset threshold.

2. The method of claim 1, wherein the method is used for the secondary screening of Alzheimer's disease. The resting-state functional magnetic resonance imaging data in S1 are brain functional images acquired by subjects while they are awake and not performing tasks, and are used to reflect the spatiotemporal changes in brain functional activity. Resting-state functional magnetic resonance imaging (fMRI) data were obtained from clinically acquired data or publicly available neuroimaging databases. Clinical acquisition was performed using magnetic resonance imaging equipment, and the scanning sequence used was echo-planar imaging sequence. Brain imaging data at multiple time points were obtained during the scanning process, with each time point corresponding to a set of three-dimensional brain voxel signals.

3. The method of claim 1, wherein the method is used for the screening of Alzheimer's disease. The preprocessing in S2 includes sequential temporal correction, head motion correction, spatial normalization, noise signal regression, and spatial smoothing.

4. The method of claim 1, wherein the method is used for the screening of Alzheimer's disease. The formula for calculating the spatial variation value in S3 is: ， In the formula, This represents the spatial variation value of the r-th ROI at time point t; This represents the functional signal value of the i-th voxel within the r-th ROI at time point t; N represents the average signal value of the r-th ROI at time point t; r This represents the total number of voxels contained in the r-th ROI.

5. The method of claim 1, wherein the method is used for the screening of Alzheimer's disease. Calculation of the mean of spatial variation The formula is: ， wherein, denotes the spatial variation mean of the rth ROI; T denotes the total number of time points; The formula for calculating the standard deviation of spatial variation is: ， wherein, denotes the spatial variation standard deviation of the rth ROI; The calculation process for the spatial variation dynamic fluctuation index is as follows: Spatial variation time series for each ROI The time window is segmented according to a preset time window length W and a sliding step size S, resulting in K time windows, each corresponding to a spatial variant sequence. Where k = 1,2,…,K; the variance of the spatially variable subsequence within each time window is calculated to obtain the dynamic volatility index, the formula of which is: , wherein, denotes the dynamic spatial variation index of the rth ROI; denotes the kth time point of the rth ROI. The spatial variant sequence within the sliding window; Var(·) represents the variance operation; K represents the number of sliding time windows.

6. The method of claim 1, wherein the method is used for the screening of Alzheimer's disease. Machine learning algorithms in S5 include, but are not limited to, support vector machines, random forests, logistic regression, or gradient boosting trees.

7. The method of claim 1, wherein the method is used for the screening of Alzheimer's disease. The auxiliary screening results include normal, mild cognitive impairment risk, and high risk of Alzheimer's disease; when the risk score is below the first threshold, it is judged as normal; When the risk score is between the first and second thresholds, it is considered to be at risk of mild cognitive impairment. When the risk score is higher than the second threshold, the individual is considered to be at high risk of Alzheimer's disease. The first threshold and the second threshold are determined based on the subject operating characteristic curves of the training set.

8. The method of claim 7, wherein the method is used for the secondary screening of Alzheimer's disease. The specific process for determining the first and second thresholds is as follows: The continuous risk score output by the classification model is between 0 and 1. Within this range, multiple point values ​​are obtained by dividing the range with a fixed step size. Any two point values ​​are used as a set of candidate thresholds. The sensitivity and specificity under each candidate threshold condition are calculated, and the corresponding Youden index J is further calculated, J = sensitivity + specificity − 1. The set of candidate thresholds that maximizes the Youden index is selected as the optimal classification threshold. The two point values ​​in this set of candidate thresholds are the first threshold and the second threshold, respectively. The calculation formula of the sensitivity is: , The formula for calculating specificity is: , In the formula: TP represents the number of true positives, which is the number of samples that are actually disease subjects and are correctly predicted by the model to be disease subjects; TN represents the number of true negatives, which is the number of samples that are actually healthy subjects and are correctly predicted by the model to be healthy subjects; FP represents the number of false positives, which is the number of samples that are actually healthy subjects but are incorrectly predicted by the model to be disease subjects; FN represents the number of false negatives, which is the number of samples that are actually disease subjects but are incorrectly predicted by the model to be healthy subjects.

Images