A method and system for constructing an early warning model of AD based on multi-dimensional testing

Abstract

The application relates to the technical field of medical health, and provides an AD early warning model construction method and system based on multi-dimensional testing, which comprises the following steps: collecting neural image data, biomarker data and clinical feature data of a plurality of subjects, screening abnormal data of the neural image data set, performing data cleaning on the biomarker data set, performing data coding on the clinical feature data set, respectively extracting features from the pretreated neural image data set and the pretreated biomarker data set to obtain a neural image feature set and a biomarker feature set, constructing a multi-dimensional feature matrix set, obtaining a weighted multi-dimensional feature matrix set by using a training set matrix subset and a test set matrix subset, training a clinical prediction model by using the weighted multi-dimensional feature matrix set, obtaining an early warning experimental model, and testing the early warning experimental model to obtain an early warning model. The application can solve the problems of low diagnosis precision of a single data source and abnormal data interference.

Description

Technical Field

[0001] This invention relates to the field of medical and health technology, and in particular to a method, system, electronic device, and computer-readable storage medium for constructing an early warning model for Alzheimer's disease (AD) based on multidimensional testing. Background Technology

[0002] With the rapid development of the healthcare industry, the prediction of Alzheimer's disease faces new challenges.

[0003] Currently, Alzheimer's disease prediction still relies mainly on single biomarker detection or clinical-cognitive assessment, and a multimodal integrated system has not yet been formed.

[0004] While traditional methods for predicting Alzheimer's disease can indicate whether a patient will develop the disease, they suffer from limitations such as considering only a single characteristic factor and low predictive accuracy. Therefore, optimizing Alzheimer's disease prediction methods is crucial for improving predictive accuracy. Summary of the Invention

[0005] This invention provides a method for constructing an early warning model for Alzheimer's disease (AD) based on multidimensional testing, and a computer-readable storage medium. Its main purpose is to solve the problems of low diagnostic accuracy from a single data source and interference from abnormal data.

[0006] To achieve the above objectives, this invention provides a method for constructing an early warning model for Alzheimer's disease (AD) based on multidimensional testing, comprising:

[0007] Neuroimaging data, biomarker data, and clinical characteristic data of each of multiple subjects identified by subject tags are collected to obtain a neuroimaging dataset, a biomarker dataset, and a clinical characteristic dataset, wherein the subject tags are healthy subjects or diseased subjects.

[0008] Anomaly screening was performed on the neuroimaging dataset to obtain a preprocessed neuroimaging dataset.

[0009] Data cleaning was performed on the biomarker dataset to obtain a preprocessed biomarker dataset;

[0010] The clinical feature dataset is encoded to obtain a preprocessed clinical feature dataset;

[0011] Feature extraction was performed on the preprocessed neural image dataset and the preprocessed biomarker dataset to obtain the neural image feature set and the biomarker feature set, respectively.

[0012] A multidimensional feature matrix set was constructed using neuroimaging feature sets, biomarker feature sets, and preprocessed clinical feature datasets;

[0013] Randomly sample from the multidimensional feature matrix set to obtain a subset of the training set matrix and a subset of the test set matrix, and use the subset of the training set matrix and the subset of the test set matrix to obtain a weighted multidimensional feature matrix set;

[0014] The pre-constructed clinical prediction model was trained using a weighted multidimensional feature matrix set to obtain an early warning experimental model. The early warning experimental model was then tested to obtain test evaluation results.

[0015] The early warning experimental model was optimized based on the test and evaluation results to obtain the early warning model.

[0016] Optionally, the step of screening the neural image dataset for abnormal data to obtain a preprocessed neural image dataset includes:

[0017] Aligning multiple neuroimaging data points in the neuroimaging dataset with a pre-constructed standard template space to obtain a standardized neuroimaging dataset, which includes multiple standardized neuroimaging data points.

[0018] Intensity normalization was performed on all standardized neuroimaging data in the standardized neuroimaging dataset to obtain a normalized neuroimaging dataset.

[0019] The normalized neural image dataset is denoised to obtain the denoised normalized neural image dataset.

[0020] Anomaly screening is performed on the denoised normalized neural image dataset to obtain one or more abnormal image data. The one or more abnormal image data are then removed from the denoised normalized neural image dataset to obtain the preprocessed neural image dataset.

[0021] Optionally, the step of cleaning the biomarker dataset to obtain a preprocessed biomarker dataset includes:

[0022] Missing data detection is performed on the biomarker dataset to obtain a missing data detection report;

[0023] Based on the missing data detection report, the missing data was identified, and the missing data was imputed using a pre-built random forest imputation method to obtain a complete biomarker dataset.

[0024] Feature signals are extracted from the complete biomarker dataset to obtain the original biomarker sequence set;

[0025] The original biomarker sequence set was analyzed using a pre-constructed local outlier factor algorithm to calculate multiple local outlier factor values. The calculation formula for the local outlier factor algorithm is shown below:

[0026]

[0027] in, Represents the data points of the original biomarker sequence set. The corresponding local outlier factor value, This represents data points in the original biomarker sequence set. This represents the preset neighborhood value. Indicates the first Individual and The nearest original biomarker sequence neighbor, Indicates the first Neighboring points Locally achievable density, express Locally achievable density;

[0028] Threshold determination is performed on all the local outlier factor values ​​to obtain an outlier detection report;

[0029] Extreme outliers were identified based on the outlier detection report, and the extreme outliers were corrected using the random forest filling method to obtain the corrected biomarker dataset.

[0030] The modified biomarker dataset was normalized to obtain the preprocessed biomarker dataset.

[0031] Optionally, the step of encoding the clinical feature dataset to obtain a preprocessed clinical feature dataset includes:

[0032] The clinical feature dataset is classified to obtain a cumulative feature set and a judgment feature set.

[0033] Continuous encoding is performed on multiple cumulative features in the cumulative feature set to obtain a continuous variable dataset;

[0034] Discrete encoding is performed on multiple judgment features in the judgment feature set to obtain a discrete variable dataset;

[0035] The continuous variable dataset and the discrete variable dataset are merged to obtain the clinical feature coding dataset;

[0036] The clinical feature encoding dataset is normalized to obtain a preprocessed clinical feature dataset.

[0037] Optionally, the step of extracting features from the preprocessed neural image dataset and the preprocessed biomarker dataset to obtain a neural image feature set and a biomarker feature set includes:

[0038] Morphological and texture features were extracted from the preprocessed neural image data in the preprocessed neural image dataset to obtain morphological and texture feature subsets.

[0039] The morphological feature subset and the texture feature subset are merged to obtain the neuroimaging feature set;

[0040] Blood biomarker data and cerebrospinal fluid biomarker data were extracted from the preprocessed biomarker dataset to obtain the blood biomarker dataset and the cerebrospinal fluid biomarker dataset.

[0041] Concentration features were extracted from blood biomarker data in the blood biomarker dataset and cerebrospinal fluid biomarker data in the cerebrospinal fluid biomarker dataset to obtain biomarker feature sets.

[0042] Optionally, the construction of a multidimensional feature matrix set using neuroimaging feature sets, biomarker feature sets, and preprocessed clinical feature datasets includes:

[0043] The neuroimaging feature set, biomarker feature set, and preprocessed clinical feature dataset are concatenated to obtain multiple original multidimensional feature matrices. Each original multidimensional feature matrix corresponds to the neuroimaging feature data, biomarker feature data, and clinical feature data of a subject.

[0044] Dimensionality reduction is performed on multiple original multidimensional feature matrices to obtain multiple dimensionality-reduced original multidimensional feature matrices.

[0045] Normalize the original multidimensional feature matrices after dimensionality reduction to obtain a set of multidimensional feature matrices.

[0046] Optionally, the step of randomly sampling from the multidimensional feature matrix set to obtain a subset of the training set matrix and a subset of the test set matrix, and using the subset of the training set matrix and the subset of the test set matrix to obtain a weighted multidimensional feature matrix set, includes:

[0047] According to a preset ratio, the multidimensional feature matrix set is randomly sampled to obtain a training set matrix subset and a test set matrix subset;

[0048] Standardize a subset of the training set matrix to obtain a standardized subset of the training feature matrix.

[0049] The global weight vector matrix is ​​obtained by using a pre-constructed weight optimization algorithm and a standardized training feature matrix subset;

[0050] The global weight vector matrix is ​​weighted and fused with the training set matrix subset and the test set matrix subset respectively to obtain a weighted training feature matrix set and a weighted test feature matrix set;

[0051] By summing the weighted training feature matrix set and the weighted test feature matrix set, a weighted multidimensional feature matrix set is obtained.

[0052] Optionally, the step of training a pre-constructed clinical prediction model using a weighted multidimensional feature matrix set to obtain an early warning experimental model, testing the early warning experimental model, and obtaining test evaluation results includes:

[0053] The weighted multidimensional feature matrix set is merged with the subject labels to obtain the overall weighted dataset;

[0054] A one-time random stratified sampling is performed on the entire weighted dataset according to a preset ratio to obtain a weighted training set and a weighted test set;

[0055] A pre-constructed clinical prediction model was trained using a weighted training set to obtain an early warning experimental model;

[0056] Multiple weighted multidimensional feature matrices from the weighted test set are input into the early warning experimental model to obtain test evaluation results.

[0057] Optionally, optimizing the early warning experimental model based on the test evaluation results to obtain the early warning model includes:

[0058] The prediction accuracy is obtained using the test evaluation results. This prediction accuracy is then compared to a preset accuracy threshold. If the prediction accuracy is confirmed to be lower than the threshold, a pre-built adaptive learning rate adjustment algorithm is used to calculate the optimized learning rate. The calculation formula is shown below:

[0059]

[0060] in, This indicates optimizing the learning rate. This represents the learning rate of the early warning experimental model. Represents the sensitivity coefficient. Indicates the magnification factor. Indicates the accuracy threshold. Indicates prediction accuracy. Represents the forgetting factor coefficient. This represents the learning rate from the previous round, used during the first optimization. =0, Represents the hyperbolic tangent function;

[0061] The learning rate of the early warning experimental model is adjusted using the optimized learning rate. The adjusted early warning experimental model is then used as the early warning experimental model. The process of inputting multiple weighted multidimensional feature matrices from the weighted test set into the early warning experimental model to obtain the test evaluation results is repeated until the prediction accuracy is greater than or equal to the accuracy threshold, at which point the early warning experimental model is confirmed as an early warning model.

[0062] To achieve the above objectives, the present invention also provides an AD early warning model construction system based on multidimensional testing, comprising:

[0063] The data acquisition module is used to collect neuroimaging data, biomarker data, and clinical characteristic data of each of multiple subjects identified by subject tags, to obtain a neuroimaging dataset, a biomarker dataset, and a clinical characteristic dataset, wherein the subject tags are healthy subjects or diseased subjects.

[0064] The data processing module is used to screen for abnormal data in the neuroimaging dataset to obtain a preprocessed neuroimaging dataset.

[0065] Data cleaning was performed on the biomarker dataset to obtain a preprocessed biomarker dataset;

[0066] The clinical feature dataset is encoded to obtain a preprocessed clinical feature dataset;

[0067] Feature extraction was performed on the preprocessed neural image dataset and the preprocessed biomarker dataset to obtain the neural image feature set and the biomarker feature set, respectively.

[0068] The weighted fusion module is used to construct a multidimensional feature matrix set using neuroimaging feature sets, biomarker feature sets, and preprocessed clinical feature datasets;

[0069] Randomly sample from the multidimensional feature matrix set to obtain a subset of the training set matrix and a subset of the test set matrix, and use the subset of the training set matrix and the subset of the test set matrix to obtain a weighted multidimensional feature matrix set;

[0070] The model building module is used to train a pre-built clinical prediction model using a weighted multidimensional feature matrix set to obtain an early warning experimental model, test the early warning experimental model, and obtain test evaluation results.

[0071] The early warning experimental model was optimized based on the test and evaluation results to obtain the early warning model.

[0072] To address the above problems, the present invention also provides an electronic device, the electronic device comprising:

[0073] Memory, storing at least one instruction;

[0074] The processor executes the instructions stored in the memory to implement the AD early warning model construction method based on multidimensional testing described above.

[0075] To address the aforementioned problems, the present invention also provides a computer-readable storage medium storing at least one instruction, which is executed by a processor in an electronic device to implement the aforementioned method for constructing an AD early warning model based on multidimensional testing.

[0076] To address the problems described in the background section, this invention collects neuroimaging data, biomarker data, and clinical feature data from each of multiple subjects tagged with subject labels, resulting in a neuroimaging dataset, a biomarker dataset, and a clinical feature dataset. The subject labels are either healthy or diseased subjects. This invention eliminates data fragmentation and label mismatch at the source by simultaneously collecting multimodal data and assigning unified labels, laying a highly consistent and traceable data foundation for subsequent fusion and modeling. Based on this, the invention performs anomaly screening on the neuroimaging dataset to obtain a preprocessed neuroimaging dataset, cleans the biomarker dataset to obtain a preprocessed biomarker dataset, and encodes the clinical feature dataset to obtain a preprocessed clinical feature dataset. Feature extraction is then performed on both the preprocessed neuroimaging dataset and the preprocessed biomarker dataset to obtain neuroimaging feature sets and biomarker feature sets, respectively. This invention effectively removes noise, fills in missing data, and aligns heterogeneous features under the same dimension by implementing a standardized cleaning-encoding-normalization pipeline for the three types of data, significantly improving the robustness and accuracy of subsequent fusion and modeling. Next, this invention constructs a multidimensional feature matrix set using neuroimaging feature sets, biomarker feature sets, and preprocessed clinical feature datasets. Random sampling is performed from this multidimensional feature matrix set to obtain training set matrix subsets and test set matrix subsets. A weighted multidimensional feature matrix set is then obtained using these subsets. This invention demonstrates that while preserving key information, it achieves sample balance and quantifies feature importance, significantly enhancing the model's generalization ability and prediction accuracy. Furthermore, this invention trains a pre-constructed clinical prediction model using the weighted multidimensional feature matrix set to obtain an early warning experimental model. The early warning experimental model is then tested to obtain test evaluation results. Based on these results, the early warning experimental model is optimized to obtain an early warning model. This invention, through a closed-loop training-evaluation-optimization mechanism, enables the model to maintain training efficiency while dynamically adjusting the learning rate and weights based on test feedback, significantly improving accuracy and robustness. The final output early warning model possesses high generalization ability and clinical interpretability, and can be directly deployed in real-world scenarios to achieve personalized, real-time Alzheimer's disease risk prediction. Therefore, this invention can solve the problems of low diagnostic accuracy from a single data source and interference from abnormal data. Attached Figure Description

[0077] Figure 1 A flowchart illustrating an embodiment of the AD early warning model construction method based on multidimensional testing provided by the present invention;

[0078] Figure 2 This is a functional block diagram of an AD early warning model construction system based on multidimensional testing provided in an embodiment of the present invention;

[0079] Figure 3 This is a schematic diagram of the structure of an electronic device that implements the AD early warning model construction method based on multidimensional testing, according to an embodiment of the present invention.

[0080] Explanation of reference numerals in the attached figures:

[0081] 10. Electronic device; 11. Processor; 12. Memory; 13. Bus.

[0082] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0083] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0084] This application provides a method for constructing an early warning model for Alzheimer's disease (AD) based on multidimensional testing. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0085] Reference Figure 1 The diagram shown is a flowchart illustrating a method for constructing an early warning model for Alzheimer's disease (AD) based on multidimensional testing, according to an embodiment of the present invention. In this embodiment, the method for constructing an early warning model for AD based on multidimensional testing includes:

[0086] S1. Collect neuroimaging data, biomarker data, and clinical characteristic data for each of the multiple subjects identified by subject tags to obtain a neuroimaging dataset, a biomarker dataset, and a clinical characteristic dataset, wherein the subject tags are healthy subjects or diseased subjects.

[0087] It should be explained that neuroimaging data refers to brain structural imaging data obtained using techniques such as magnetic resonance imaging (MRI) and positron emission tomography (PET). A neuroimaging dataset is a collection of neuroimaging data from multiple subjects. Biomarker data is data obtained through bodily fluid analysis, including blood and cerebrospinal fluid indicators. For example, blood indicators include: plasma... The ratio, p-tau181 concentration, and cerebrospinal fluid indicators include t-tau protein and NfL neurofilament light chain protein concentration. Body fluid testing is a detection technique that obtains specific molecular markers by analyzing human blood and cerebrospinal fluid samples. Optionally, a digital single-molecule immunoassay analyzer is used for the body fluid testing. A biomarker dataset is a collection of biomarker data from multiple subjects.

[0088] Furthermore, clinical characteristic data refers to the subject's age, gender, marital status, family history of Alzheimer's disease, years of education, and other characteristic data. A clinical characteristic dataset is a collection of clinical characteristic data from multiple subjects. Subject labels are assigned based on whether a subject has Alzheimer's disease. Multiple subjects with subject labels are identified using these labels to indicate the disease status of each subject. For example, subject labels can be set to [0 / 1], where a healthy subject has a label of [0], and a subject with Alzheimer's disease has a label of [1]. This allows for comparison in subsequent data analysis. Here, a healthy subject refers to an otherwise healthy person, and a patient subject refers to someone with Alzheimer's disease. AD stands for Alzheimer's disease, a degenerative disease of the central nervous system.

[0089] S2. Perform anomaly screening on the neuroimaging dataset to obtain a preprocessed neuroimaging dataset.

[0090] It should be explained that the process of screening for abnormal data in the neuroimaging dataset to obtain a preprocessed neuroimaging dataset includes:

[0091] Aligning multiple neuroimaging data points in the neuroimaging dataset with a pre-constructed standard template space to obtain a standardized neuroimaging dataset, which includes multiple standardized neuroimaging data points.

[0092] Intensity normalization was performed on all standardized neuroimaging data in the standardized neuroimaging dataset to obtain a normalized neuroimaging dataset.

[0093] The normalized neural image dataset is denoised to obtain the denoised normalized neural image dataset.

[0094] Anomaly screening is performed on the denoised normalized neural image dataset to obtain one or more abnormal image data. The one or more abnormal image data are then removed from the denoised normalized neural image dataset to obtain the preprocessed neural image dataset.

[0095] Furthermore, multiple neuroimaging datasets within the neuroimaging dataset are aligned with a pre-constructed standard template space to obtain a standardized neuroimaging dataset. This involves aligning the neuroimaging data with the standard template space through rotation or scaling. The aim is to eliminate individual differences in brain size, orientation, and scanning location, placing all images within the same anatomical framework to facilitate subsequent feature extraction, machine learning, and clinical comparison. The pre-constructed standard template space is a universal coordinate reference system based on the brain anatomy of healthy individuals, containing standardized brain region divisions and spatial coordinate systems. Optionally, Talelash coordinates are used as the standard template space. Standardized neuroimaging data refers to the image data obtained after aligning the neuroimaging data. A standardized neuroimaging dataset is a collection containing multiple standardized neuroimaging datasets.

[0096] Understandably, intensity normalization is a process to eliminate grayscale value deviations caused by differences in scanning equipment. The purpose of intensity normalization is to eliminate grayscale value deviations caused by different scanning equipment, sequence parameters, or gain settings. Optionally, histogram matching can be used as the intensity normalization method. The normalized neuroimaging dataset is a collection obtained by summarizing standardized neuroimaging data that has undergone multiple intensity normalizations. Noise reduction is a method to improve image quality by eliminating noise in the normalized neuroimaging data. Optionally, spatial filtering can be used to perform noise reduction on the normalized neuroimaging data in the normalized neuroimaging dataset.

[0097] It should be explained that anomaly screening is a detection method that classifies neuroimages into (normal / abnormal) categories according to a preset anomaly screening standard method. This standard method includes determining whether scan integrity, the presence of motion artifacts, anatomical structure integrity, and alignment accuracy meet preset reference conditions. Abnormal image data refers to images identified as abnormal during anomaly screening. The preprocessed neuroimage dataset is a collection of neuroimage data from multiple subjects, preserved after spatial alignment, intensity normalization, noise reduction, and anomaly screening. For example, the set reference conditions are as follows: scan integrity greater than or equal to 90%, anatomical structure integrity greater than or equal to 85%, no motion artifacts in the normalized neuroimage data, and alignment accuracy greater than or equal to 95%. Therefore, if the normalized neuroimage data contains motion artifacts, has a scan integrity less than 90%, an anatomical structure integrity less than 85%, or an alignment accuracy less than 95%, then the normalized neuroimage data is considered abnormal image data.

[0098] S3. Perform data cleaning on the biomarker dataset to obtain a preprocessed biomarker dataset.

[0099] It should be explained that the data cleaning of the biomarker dataset to obtain the preprocessed biomarker dataset includes:

[0100] Missing data detection is performed on the biomarker dataset to obtain a missing data detection report;

[0101] Based on the missing data detection report, the missing data was identified, and the missing data was imputed using a pre-built random forest imputation method to obtain a complete biomarker dataset.

[0102] Feature signals are extracted from the complete biomarker dataset to obtain the original biomarker sequence set;

[0103] The original biomarker sequence set was analyzed using a pre-constructed local outlier factor algorithm to calculate multiple local outlier factor values. The calculation formula for the local outlier factor algorithm is shown below:

[0104]

[0105] in, Represents the data points of the original biomarker sequence set. The corresponding local outlier factor value, This represents data points in the original biomarker sequence set. This represents the preset neighborhood value. Indicates the first Individual and The nearest original biomarker sequence neighbor, Indicates the first Neighboring points Locally achievable density, express Locally achievable density;

[0106] Threshold determination is performed on all the local outlier factor values ​​to obtain an outlier detection report;

[0107] Extreme outliers were identified based on the outlier detection report, and the extreme outliers were corrected using the random forest filling method to obtain the corrected biomarker dataset.

[0108] The modified biomarker dataset was normalized to obtain the preprocessed biomarker dataset.

[0109] Furthermore, missing data detection refers to methods for detecting null values ​​in biomarker data within a biomarker dataset. Optionally, a Python library can be used to perform null value detection on the biomarker data in the biomarker dataset. The missing data detection report is a report obtained through missing data detection, including the type and location of missing data. For example, biomarker data may include plasma... The ratio, p-tau181 concentration, t-tau protein, and NfL neurofilament light chain protein concentration indicate the plasma... The ratio, p-tau181 concentration, t-tau protein, and NfL neurofilament light chain protein concentration can all be considered as missing data types. The missing location refers to the position where the null value is missing, i.e., the location of the missing data for a specific patient. Missing data are null values ​​confirmed to exist in the biomarker dataset by a missing data detection report.

[0110] It should be explained that using a pre-built random forest imputation method to fill in missing data and obtain a complete biomarker dataset refers to the process of using a random forest imputation method to fill in missing data in a biomarker dataset. The pre-built random forest imputation method is a machine learning-based missing value imputation method. Optionally, random forest can be used as the missing value imputation method. The complete biomarker dataset is the complete dataset obtained after imputing the missing data in the biomarker dataset using the random forest imputation method.

[0111] Understandably, feature signal extraction is a method of screening biomarker data that are significantly associated with disease risk from a complete biomarker dataset based on the pathological mechanisms of Alzheimer's disease and clinical validation evidence, and converting it into a raw biomarker sequence set that can be used for subsequent modeling. Optionally, a single-molecule immunoassay platform can be used as the instrument for the feature signal extraction. The raw biomarker sequence set refers to the set of data sequences obtained after the feature signal extraction step.

[0112] It should be explained that, to facilitate understanding of the process of using a pre-constructed local outlier factor algorithm to calculate outlier factors on the original biomarker sequence set and obtain multiple local outlier factor values, the following example is given: for instance, plasma from multiple subjects... The concentration value was used as a sequence in the original biomarker sequence set for calculating the local outlier factor. The sequence was assumed to be (A: 800, B: 820, C: 810, D: 805, E: 815, F: 807, G: 812, H: 808, I: 803, J: 1000). =3 (3 nearest neighbors), using J as the original biomarker sequence set data point. Calculate the absolute difference between each data point in the sequence and the target point, and identify 3 points that are related to the target point. The most recent original biomarker sequence neighbors, of which 3 are related to The nearest neighbors of the original biomarker sequence are nearest neighbor 1: 820, nearest neighbor 2: 815, and nearest neighbor 3: 812. The nearest neighbors are determined by calculating the absolute difference between each data point in the sequence and the target point, and then sorting these absolute differences in descending order to obtain the absolute difference sequence. If the absolute difference is 3, then the data points corresponding to the top 3 absolute differences in the absolute difference sequence, sorted from smallest to largest, are taken as the nearest neighbors. The local reachability density of the first neighbor is calculated to be 0.10 using the nearest neighbor ratio of 1:820. The technique for calculating local reachability density is existing technology and will not be elaborated here. The reachability densities of neighbor points 1, 2, 3, and... are calculated respectively. The local reachability density was calculated, and the local outlier factor was 18.8. The local outlier factor is a quantitative indicator that measures the difference between a single data point and the density of its local neighborhood. The larger the value, the more the point deviates from the normal neighborhood and the more likely it is to be judged as an anomaly.

[0113] Furthermore, threshold determination is applied to all the aforementioned local outlier values ​​to obtain an outlier detection report. This means setting a threshold for the calculated local outlier values ​​for all data points. If the local outlier value corresponding to a data point exceeds the threshold, the data point is determined to be an outlier. All outlier data points are then aggregated to obtain the outlier detection report. Extreme outliers refer to biomarker data in the original biomarker sequence set whose local outlier values ​​are confirmed to be higher than the set threshold according to the outlier detection report. For example, if the local outlier value is 18.8 and the threshold is 1, then J:1000 is confirmed as an extreme outlier.

[0114] Understandably, using random forest imputation to correct extreme outliers to obtain a corrected biomarker dataset means correcting extreme outliers to obtain a corrected biomarker dataset, confirming that the biomarker data in the corrected biomarker dataset maintains an interpretable relationship with the actual pathological mechanism. Random forest imputation is an existing technique and will not be elaborated upon here. The corrected biomarker dataset refers to the dataset after extreme outlier correction. Normalization refers to scaling the values ​​of each feature in the corrected biomarker dataset to a range of (0-1). Optionally, Min-Max normalization can be used as the normalization method. The preprocessed biomarker dataset refers to the biomarker dataset after missing data detection and imputation, outlier detection and correction, and normalization.

[0115] S4. Encode the clinical feature dataset to obtain a preprocessed clinical feature dataset.

[0116] It should be explained that the process of encoding the clinical feature dataset to obtain a preprocessed clinical feature dataset includes:

[0117] The clinical feature dataset is classified to obtain a cumulative feature set and a judgment feature set.

[0118] Continuous encoding is performed on multiple cumulative features in the cumulative feature set to obtain a continuous variable dataset;

[0119] Discrete encoding is performed on multiple judgment features in the judgment feature set to obtain a discrete variable dataset;

[0120] The continuous variable dataset and the discrete variable dataset are merged to obtain the clinical feature coding dataset;

[0121] The clinical feature encoding dataset is normalized to obtain a preprocessed clinical feature dataset.

[0122] Furthermore, feature classification refers to dividing clinical features into cumulative features or judgmental features based on the processing methods used in medical or statistical analysis. Cumulative features include, but are not limited to, age and years of education, while judgmental features include, but are not limited to, marital status and family history of Alzheimer's disease. A cumulative feature set is a collection of cumulative features from multiple subjects. A judgmental feature set is a collection of judgmental features from multiple subjects.

[0123] Understandably, encoding multiple cumulative features in a cumulative feature set continuously to obtain a continuous variable dataset means preserving the original numerical form of the cumulative features for each subject in the cumulative feature set (e.g., 10 years of education is preserved as a numerical value of 10), resulting in multiple subsets of continuous variable data. These subsets are then combined to form the continuous variable dataset. A subset of continuous variable data is the dataset obtained by continuously encoding the cumulative features of a single subject. A continuous variable dataset is the dataset obtained by summing multiple subsets of continuous variable data.

[0124] Furthermore, discrete coding transforms the original categorical feature values ​​into a numerical form suitable for processing by machine learning algorithms.

[0125] For example, the marital status of the subjects is discretely encoded, where marital status is set as [married / unmarried], and the discretely encoded marital status is [0 / 1]. When the discretely encoded marital status is [0], it indicates that the subject is unmarried, and when the discretely encoded marital status is [1], it indicates that the subject is married. The family history of Alzheimer's disease of the subjects is discretely encoded, where family history of Alzheimer's disease is set as [family history of Alzheimer's disease exists / family history of Alzheimer's disease does not exist], and the discretely encoded family history of Alzheimer's disease is [0 / 1]. When the discretely encoded family history of Alzheimer's disease is [0], it indicates that there is a history of Alzheimer's disease in the subject's family, and when the discretely encoded family history of Alzheimer's disease is [1], it indicates that there is no history of Alzheimer's disease in the subject's family.

[0126] It needs to be explained that merging the continuous variable dataset and the discrete variable dataset to obtain the clinical feature coding dataset refers to merging multiple subsets of continuous variable data from the continuous variable dataset and multiple subsets of judgment features from the judgment feature set according to the subject dimension (one subject is considered as one dimension). These multiple clinical feature coding subsets are then aggregated to obtain the clinical feature coding dataset. The judgment feature subset is obtained by discretely encoding the judgment features of a single subject.

[0127] Furthermore, normalization refers to scaling the values ​​of each feature in the clinical feature encoding dataset to a range of (0-1). Optionally, Min-Max normalization can be used as the normalization method. The preprocessed clinical feature dataset is the dataset obtained after normalizing the clinical feature encoding dataset.

[0128] S5. Perform feature extraction on the preprocessed neural image dataset and the preprocessed biomarker dataset respectively to obtain the neural image feature set and the biomarker feature set.

[0129] It should be explained that the step of extracting features from the preprocessed neural image dataset and the preprocessed biomarker dataset to obtain the neural image feature set and the biomarker feature set includes:

[0130] Morphological and texture features were extracted from the preprocessed neural image data in the preprocessed neural image dataset to obtain morphological and texture feature subsets.

[0131] The morphological feature subset and the texture feature subset are merged to obtain the neuroimaging feature set;

[0132] Blood biomarker data and cerebrospinal fluid biomarker data were extracted from the preprocessed biomarker dataset to obtain the blood biomarker dataset and the cerebrospinal fluid biomarker dataset.

[0133] Concentration features were extracted from blood biomarker data in the blood biomarker dataset and cerebrospinal fluid biomarker data in the cerebrospinal fluid biomarker dataset to obtain biomarker feature sets.

[0134] Furthermore, morphological feature extraction refers to quantifying the geometric properties of brain structures associated with the risk of Alzheimer's disease from preprocessed neuroimaging data in a preprocessed neuroimaging dataset. Examples include hippocampal volume, cortical thickness, ventricular dilation rate, and surface curvature. The aim is to capture macroscopic structural changes associated with Alzheimer's disease. Optionally, voxel-based morphological analysis can be used as the method for morphological feature extraction.

[0135] Understandably, texture feature extraction refers to the process of quantifying the spatial distribution pattern of grayscale intensity in preprocessed neuroimaging data. The aim is to identify and quantify the heterogeneous changes in brain tissue microstructure associated with Alzheimer's disease prevalence, such as reduced gray matter microstructural complexity, changes in the directionality of white matter fiber bundles, and increased local tissue inhomogeneity. Optionally, texture feature extraction can be achieved by constructing a gray-level co-occurrence matrix. The morphological feature subset refers to the set of morphological features for each subject obtained through morphological feature extraction. The texture feature subset refers to the set of texture features for each subject obtained through texture feature extraction. Merging the morphological and texture feature subsets to obtain the neuroimaging feature set refers to merging the morphological and texture feature subsets along the subject dimension. The neuroimaging feature set is a set containing both morphological and texture features for each subject.

[0136] Furthermore, a blood biomarker dataset refers to biomarker data extracted from the blood of multiple subjects from a preprocessed biomarker dataset. For example: plasma. Examples of biomarker datasets include plasma pTau181, serum neurofilament light chain protein, and serum glial fibrillary acidic protein. Cerebrospinal fluid (CSF) biomarker datasets refer to biomarker data extracted from the CSF of multiple subjects from a preprocessed biomarker dataset. For example: CSF... CSF pTau, CSFtTau, and CSF NfL, etc.

[0137] Understandably, concentration feature extraction is an operation that directly uses the original concentration measurements of biomarkers or their simple, biologically / clinically significant transformations or combinations as features. For example, the washed, normalized concentration value of each biomarker can be used as a feature. For certain combinations of biomarkers, their ratios can be used as new features, as ratios have stronger disease indicative significance. For example, CSF pTau / Aβ42 and CSF tTau / Aβ42. The biomarker feature set was obtained with reference to the Chinese Clinical Application Guidelines for Humoral Biomarkers in Alzheimer's Disease.

[0138] S6. Construct a multidimensional feature matrix set using neuroimaging feature sets, biomarker feature sets, and preprocessed clinical feature datasets.

[0139] It should be explained that the construction of a multidimensional feature matrix set using neuroimaging feature sets, biomarker feature sets, and preprocessed clinical feature datasets includes:

[0140] The neuroimaging feature set, biomarker feature set, and preprocessed clinical feature dataset are concatenated to obtain multiple original multidimensional feature matrices. Each original multidimensional feature matrix corresponds to the neuroimaging feature data, biomarker feature data, and clinical feature data of a subject.

[0141] Dimensionality reduction is performed on multiple original multidimensional feature matrices to obtain multiple dimensionality-reduced original multidimensional feature matrices.

[0142] Normalize the original multidimensional feature matrices after dimensionality reduction to obtain a set of multidimensional feature matrices.

[0143] Furthermore, the neuroimaging feature set, biomarker feature set, and preprocessed clinical feature dataset are concatenated to obtain multiple original multidimensional feature matrices. This means that the neuroimaging features, biomarker features, and preprocessed clinical feature data of each subject are mapped one-to-one along the subject dimension, resulting in multiple corresponding feature datasets. These corresponding feature datasets are then used to construct multiple original multidimensional feature matrices. The original multidimensional feature matrix is ​​a matrix constructed using the subject's neuroimaging features, biomarker features, and preprocessed clinical feature data.

[0144] Understandably, dimensionality reduction involves reducing the dimensionality of the original multidimensional matrix while preserving its statistical characteristics. Optionally, principal component analysis (PCA) can be used as the dimensionality reduction method. Normalization refers to scaling values ​​with the same characteristics in multiple dimensionality-reduced original multidimensional feature matrices to a range of (0-1). Optionally, Min-Max normalization can be used as the normalization method. For example, normalizing CSF pTau / Aβ42 and CSF tTau / Aβ42 in the three original feature matrices after dimensionality reduction means extracting three CSF pTau / Aβ42 values ​​from the three original feature matrices, normalizing the extracted three CSF pTau / Aβ42 values, and mapping the normalized CSF pTau / Aβ42 values ​​back into the original feature matrices. Similarly, extracting three CSF tTau / Aβ42 values ​​from the three original feature matrices, normalizing the extracted three CSF tTau / Aβ42 values, and mapping the normalized CSF tTau / Aβ42 values ​​back into the original feature matrices.

[0145] S7. Randomly sample from the multidimensional feature matrix set to obtain a training set matrix subset and a test set matrix subset, and use the training set matrix subset and the test set matrix subset to obtain a weighted multidimensional feature matrix set.

[0146] It should be explained that the step of randomly sampling from the multidimensional feature matrix set to obtain a subset of the training set matrix and a subset of the test set matrix, and using the subset of the training set matrix and the subset of the test set matrix to obtain a weighted multidimensional feature matrix set, includes:

[0147] According to a preset ratio, the multidimensional feature matrix set is randomly sampled to obtain a training set matrix subset and a test set matrix subset;

[0148] Standardize a subset of the training set matrix to obtain a standardized subset of the training feature matrix.

[0149] The global weight vector matrix is ​​obtained by using a pre-constructed weight optimization algorithm and a standardized training feature matrix subset;

[0150] The global weight vector matrix is ​​weighted and fused with the training set matrix subset and the test set matrix subset respectively to obtain a weighted training feature matrix set and a weighted test feature matrix set;

[0151] By summing the weighted training feature matrix set and the weighted test feature matrix set, a weighted multidimensional feature matrix set is obtained.

[0152] Furthermore, the preset ratio refers to the pre-determined ratio used to divide the training set matrix subset and the test set matrix subset, for example, 7. 3. The training set matrix subset is a subset of the multidimensional feature matrix set used for training the model, obtained through random sampling. The test set matrix subset is a subset of the multidimensional feature matrix set used for independently evaluating model performance, obtained through random sampling. Standardization refers to Z-score standardization of densely distributed continuous numerical features (such as CSFtTau concentrations concentrated between 0.4 and 0.6 after normalization) in the training set matrix subset. The aim is to eliminate the problem of difficulty in splitting the weight optimization algorithm due to the dense distribution of values ​​in the training set matrix subset, thereby improving the efficiency and stability of the subsequent weight optimization algorithm. The standardized training feature matrix subset refers to the training set matrix subset after the above standardization process.

[0153] It should be explained that the pre-built weight optimization algorithm refers to the pre-configured weight optimization model; optionally, XGBoost is used as the weight optimization algorithm. The global weight vector matrix refers to the set of numerical weights representing the contribution of each feature to the prediction of Alzheimer's disease incidence probability, obtained after training on a standardized subset of the training feature matrix using XGBoost. Weighted fusion refers to multiplying the global weight vector matrix element-wise with each element of both the training set matrix subset and the test set matrix subset. The weighted training feature matrix set refers to the matrix obtained by weighted fusion of the training set matrix subset and the global weight vector matrix. The weighted test feature matrix set refers to the new feature matrix obtained by weighted fusion of the test set matrix subset and the global weight vector matrix.

[0154] S8. Train the pre-constructed clinical prediction model using a weighted multidimensional feature matrix set to obtain an early warning experimental model. Test the early warning experimental model and obtain the test evaluation results.

[0155] It should be explained that the process of training a pre-constructed clinical prediction model using a weighted multidimensional feature matrix set to obtain an early warning experimental model, and then testing the early warning experimental model to obtain test evaluation results, includes:

[0156] The weighted multidimensional feature matrix set is merged with the subject labels to obtain the overall weighted dataset;

[0157] A one-time random stratified sampling is performed on the entire weighted dataset according to a preset ratio to obtain a weighted training set and a weighted test set;

[0158] A pre-constructed clinical prediction model was trained using a weighted training set to obtain an early warning experimental model;

[0159] Multiple weighted multidimensional feature matrices from the weighted test set are input into the early warning experimental model to obtain test evaluation results.

[0160] Further, the weighted multidimensional feature matrix set is merged with the subject labels to obtain the overall weighted dataset. This means that the weighted multidimensional feature matrix of each subject and its corresponding subject label (e.g., the subject label is set to [0 / 1], where the subject label is [0] for a healthy subject and [1] for a sick subject) are concatenated into a complete dataset, forming the overall weighted dataset that can be used to build the model. The overall weighted dataset is the dataset obtained by merging the weighted multidimensional feature matrix set with the subject labels. The preset ratio refers to the pre-determined division ratio used to divide the overall weighted dataset into a weighted training set and a weighted test set, for example, 7. 3. One-time random stratified sampling refers to stratified sampling based on the category distribution of subject labels. The weighted training set is the dataset obtained through one-time random stratified sampling and used to train the final clinical prediction model. The weighted test set is the dataset obtained through one-time random stratified sampling and used to independently evaluate the trained early warning experimental model. In both the weighted training set and the weighted test set, the proportion of healthy subjects to diseased subjects is equal.

[0161] Understandingly, training a pre-built clinical prediction model using a weighted training set to obtain an early warning experimental model refers to using a deep learning model trained on a weighted training set to obtain an early warning experimental model capable of predicting the risk of Alzheimer's disease. The pre-built clinical prediction model is a pre-selected and configured deep learning model. Optionally, a convolutional neural network can be used as the clinical prediction model. The early warning experimental model is a deep learning model trained on a weighted training set, used to predict an individual's risk of developing Alzheimer's disease. The test evaluation results refer to the set of model performance metrics calculated and output after inputting a weighted test set into the early warning experimental model. These performance metrics include prediction accuracy, recall, and specificity, used to quantify the early warning experimental model's predictive ability regarding the risk of Alzheimer's disease.

[0162] S9. Optimize the early warning experimental model based on the test and evaluation results to obtain the early warning model.

[0163] It should be explained that the process of optimizing the early warning experimental model based on the test evaluation results to obtain the early warning model includes:

[0164] The prediction accuracy is obtained using the test evaluation results. This prediction accuracy is then compared to a preset accuracy threshold. If the prediction accuracy is confirmed to be lower than the threshold, a pre-built adaptive learning rate adjustment algorithm is used to calculate the optimized learning rate. The calculation formula is shown below:

[0165]

[0166] in, This indicates optimizing the learning rate. This represents the learning rate of the early warning experimental model. Represents the sensitivity coefficient. Indicates the magnification factor. Indicates the accuracy threshold. Indicates prediction accuracy. Represents the forgetting factor coefficient. This represents the learning rate from the previous round, used during the first optimization. =0, Represents the hyperbolic tangent function;

[0167] The learning rate of the early warning experimental model is adjusted using the optimized learning rate. The adjusted early warning experimental model is then used as the early warning experimental model. The process of inputting multiple weighted multidimensional feature matrices from the weighted test set into the early warning experimental model to obtain the test evaluation results is repeated until the prediction accuracy is greater than or equal to the accuracy threshold, at which point the early warning experimental model is confirmed as an early warning model.

[0168] Further, prediction accuracy is the proportion of samples correctly predicted as healthy [0] and sick [1] on the weighted test set by the early warning experimental model. The preset accuracy threshold refers to the threshold set beforehand for model performance to be considered acceptable. The optimized learning rate is the learning rate obtained by optimizing the learning rate of the early warning experimental model through an adaptive learning rate adjustment algorithm. The learning rate of the early warning experimental model is the initial learning rate used during the training of the early warning model. The learning rate adjustment step size is a preset scaling factor that controls the adjustment magnitude of the learning rate based on the accuracy gap. Adjusting the learning rate of the early warning experimental model using the optimized learning rate means setting the calculated optimized learning rate as the new learning rate of the early warning experimental model. The early warning model refers to the first early warning experimental model that satisfies the prediction accuracy greater than or equal to the accuracy threshold after the above optimization cycle.

[0169] To address the problems described in the background section, this invention collects neuroimaging data, biomarker data, and clinical feature data from each of multiple subjects tagged with subject labels, resulting in a neuroimaging dataset, a biomarker dataset, and a clinical feature dataset. The subject labels are either healthy or diseased subjects. This invention eliminates data fragmentation and label mismatch at the source by simultaneously collecting multimodal data and assigning unified labels, laying a highly consistent and traceable data foundation for subsequent fusion and modeling. Based on this, the invention performs anomaly screening on the neuroimaging dataset to obtain a preprocessed neuroimaging dataset, cleans the biomarker dataset to obtain a preprocessed biomarker dataset, and encodes the clinical feature dataset to obtain a preprocessed clinical feature dataset. Feature extraction is then performed on both the preprocessed neuroimaging dataset and the preprocessed biomarker dataset to obtain neuroimaging feature sets and biomarker feature sets, respectively. This invention effectively removes noise, fills in missing data, and aligns heterogeneous features under the same dimension by implementing a standardized cleaning-encoding-normalization pipeline for the three types of data, significantly improving the robustness and accuracy of subsequent fusion and modeling. Next, this invention constructs a multidimensional feature matrix set using neuroimaging feature sets, biomarker feature sets, and preprocessed clinical feature datasets. Random sampling is performed from this multidimensional feature matrix set to obtain training set matrix subsets and test set matrix subsets. A weighted multidimensional feature matrix set is then obtained using these subsets. This invention demonstrates that while preserving key information, it achieves sample balance and quantifies feature importance, significantly enhancing the model's generalization ability and prediction accuracy. Furthermore, this invention trains a pre-constructed clinical prediction model using the weighted multidimensional feature matrix set to obtain an early warning experimental model. The early warning experimental model is then tested to obtain test evaluation results. Based on these results, the early warning experimental model is optimized to obtain an early warning model. This invention, through a closed-loop training-evaluation-optimization mechanism, enables the model to maintain training efficiency while dynamically adjusting the learning rate and weights based on test feedback, significantly improving accuracy and robustness. The final output early warning model possesses high generalization ability and clinical interpretability, and can be directly deployed in real-world scenarios to achieve personalized, real-time Alzheimer's disease risk prediction. Therefore, this invention can solve the problems of low diagnostic accuracy from a single data source and interference from abnormal data.

[0170] like Figure 2 The diagram shown is a functional block diagram of an AD early warning model construction system based on multidimensional testing provided in an embodiment of the present invention.

[0171] The AD early warning model construction system 100 based on multidimensional testing described in this invention can be installed in an electronic device. Depending on the functions implemented, the AD early warning model construction system 100 based on multidimensional testing may include a data acquisition module 101, a data processing module 102, a weighted fusion module 103, and a model construction module 104. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0172] The data acquisition module 101 is used to collect neuroimaging data, biomarker data and clinical feature data of each of the multiple subjects identified with subject tags, to obtain a neuroimaging dataset, a biomarker dataset and a clinical feature dataset, wherein the subject tags are healthy subjects or sick subjects.

[0173] The data processing module 102 is used to perform abnormal data screening on the neuroimaging dataset to obtain a preprocessed neuroimaging dataset.

[0174] Data cleaning was performed on the biomarker dataset to obtain a preprocessed biomarker dataset;

[0175] The clinical feature dataset is encoded to obtain a preprocessed clinical feature dataset;

[0176] Feature extraction was performed on the preprocessed neural image dataset and the preprocessed biomarker dataset to obtain the neural image feature set and the biomarker feature set, respectively.

[0177] The weighted fusion module 103 is used to construct a multidimensional feature matrix set using neuroimaging feature set, biomarker feature set and preprocessed clinical feature dataset;

[0178] Randomly sample from the multidimensional feature matrix set to obtain a subset of the training set matrix and a subset of the test set matrix, and use the subset of the training set matrix and the subset of the test set matrix to obtain a weighted multidimensional feature matrix set;

[0179] The model building module 104 is used to train a pre-built clinical prediction model using a weighted multidimensional feature matrix set to obtain an early warning experimental model, test the early warning experimental model, and obtain test evaluation results.

[0180] The early warning experimental model was optimized based on the test and evaluation results to obtain the early warning model.

[0181] In detail, the modules in the AD early warning model construction system 100 based on multidimensional testing described in this embodiment of the invention adopt the same methods as described above when in use. Figure 1The method used is the same as the AD early warning model construction method based on multidimensional testing described in the article, and can produce the same technical effect, so it will not be repeated here.

[0182] like Figure 3 The diagram shown is a structural schematic of an electronic device that implements a method for constructing an AD early warning model based on multidimensional testing, according to an embodiment of the present invention.

[0183] The electronic device 1 may include a processor 10, a memory 11 and a bus 12, and may also include a computer program stored in the memory 11 and capable of running on the processor 10, such as a method program for constructing an AD early warning model based on multidimensional testing.

[0184] The memory 11 includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device 1, such as a portable hard drive. In other embodiments, the memory 11 can be an external storage device of the electronic device 1, such as a plug-in portable hard drive, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device 1. Furthermore, the memory 11 includes both internal storage units and external storage devices of the electronic device 1. The memory 11 can be used not only to store application software and various types of data installed on the electronic device 1, such as the code of a method program for constructing an AD early warning model based on multidimensional testing, but also to temporarily store data that has been output or will be output.

[0185] In some embodiments, the processor 10 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory 11 (e.g., a method for constructing an AD early warning model based on multidimensional testing), and calls data stored in the memory 11 to perform various functions of the electronic device 1 and process data.

[0186] The bus 12 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus 12 can be divided into an address bus, a data bus, a control bus, etc. The bus 12 is configured to realize the connection and communication between the memory 11 and at least one processor 10, etc.

[0187] Figure 3 Only electronic devices with components are shown; it will be understood by those skilled in the art that... Figure 3 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0188] For example, although not shown, the electronic device 1 may also include a power supply (such as a battery) to power the various components. Preferably, the power supply can be logically connected to the at least one processor 10 through a power management system, thereby enabling functions such as charging management, discharging management, and power consumption management through the power management system. The power supply may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device 1 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0189] Furthermore, the electronic device 1 may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, a Bluetooth interface, etc.), which is typically used to establish communication connections between the electronic device 1 and other electronic devices.

[0190] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), and optionally, a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device 1 and to display a visual user interface.

[0191] It should be understood that the embodiments described are for illustrative purposes only and are not limited to this structure in the scope of the patent application.

[0192] The AD early warning model construction method program based on multidimensional testing, stored in the memory 11 of the electronic device 1, is a combination of multiple instructions. When run in the processor 10, it can achieve the following:

[0193] Neuroimaging data, biomarker data, and clinical characteristic data of each of multiple subjects identified by subject tags are collected to obtain a neuroimaging dataset, a biomarker dataset, and a clinical characteristic dataset, wherein the subject tags are healthy subjects or diseased subjects.

[0194] Anomaly screening was performed on the neuroimaging dataset to obtain a preprocessed neuroimaging dataset.

[0195] Data cleaning was performed on the biomarker dataset to obtain a preprocessed biomarker dataset;

[0196] The clinical feature dataset is encoded to obtain a preprocessed clinical feature dataset;

[0197] Feature extraction was performed on the preprocessed neural image dataset and the preprocessed biomarker dataset to obtain the neural image feature set and the biomarker feature set, respectively.

[0198] A multidimensional feature matrix set was constructed using neuroimaging feature sets, biomarker feature sets, and preprocessed clinical feature datasets;

[0199] Randomly sample from the multidimensional feature matrix set to obtain a subset of the training set matrix and a subset of the test set matrix, and use the subset of the training set matrix and the subset of the test set matrix to obtain a weighted multidimensional feature matrix set;

[0200] The pre-constructed clinical prediction model was trained using a weighted multidimensional feature matrix set to obtain an early warning experimental model. The early warning experimental model was then tested to obtain test evaluation results.

[0201] The early warning experimental model was optimized based on the test and evaluation results to obtain the early warning model.

[0202] Specifically, the processor 10's implementation method for the above instructions can be found in [reference needed]. Figures 1 to 3 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0203] Furthermore, if the modules / units integrated in the electronic device 1 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium may include: any entity or system capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0204] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor of an electronic device, can perform the following:

[0205] Neuroimaging data, biomarker data, and clinical characteristic data of each of multiple subjects identified by subject tags are collected to obtain a neuroimaging dataset, a biomarker dataset, and a clinical characteristic dataset, wherein the subject tags are healthy subjects or diseased subjects.

[0206] Anomaly screening was performed on the neuroimaging dataset to obtain a preprocessed neuroimaging dataset.

[0207] Data cleaning was performed on the biomarker dataset to obtain a preprocessed biomarker dataset;

[0208] The clinical feature dataset is encoded to obtain a preprocessed clinical feature dataset;

[0209] Feature extraction was performed on the preprocessed neural image dataset and the preprocessed biomarker dataset to obtain the neural image feature set and the biomarker feature set, respectively.

[0210] A multidimensional feature matrix set was constructed using neuroimaging feature sets, biomarker feature sets, and preprocessed clinical feature datasets;

[0211] Randomly sample from the multidimensional feature matrix set to obtain a subset of the training set matrix and a subset of the test set matrix, and use the subset of the training set matrix and the subset of the test set matrix to obtain a weighted multidimensional feature matrix set;

[0212] The pre-constructed clinical prediction model was trained using a weighted multidimensional feature matrix set to obtain an early warning experimental model. The early warning experimental model was then tested to obtain test evaluation results.

[0213] The early warning experimental model was optimized based on the test and evaluation results to obtain the early warning model.

[0214] Applicant's Declaration:

[0215] All personal health data collected from the subjects in this application will be used solely for research and analysis purposes, and will strictly comply with relevant privacy protection regulations. The subjects' data will be anonymized, and no personally identifiable information of any kind will be disclosed along with the health data.

[0216] All participants in this study participated voluntarily with informed consent.

[0217] The research team is committed to complying with the Personal Information Protection Law and related laws and regulations, and will take strict technical and management measures to ensure the security of the subjects' personal health data and prevent unauthorized access, disclosure, alteration or damage.

[0218] All health data will be stored on secure servers, and the data storage and processing will comply with relevant laws and regulations. Upon project completion, the data will be destroyed or stored in accordance with applicable regulations and project requirements.

[0219] The research team will take all reasonable measures to protect the privacy and safety of participants. However, due to technical and unforeseen factors, this project cannot completely eliminate the risk of data leakage, misuse, or abuse. In such cases, the research team will not be liable for any losses suffered by participants due to data leakage or other uncontrollable events.

[0220] The data collected will be used only within the scope of this study and will not be used for commercial purposes. The data will not be transferred to any third party without the participant's consent, and any use of the data by a third party requires the participant's explicit prior consent.

[0221] To the extent permitted by law, the research team shall not be liable for data loss or disclosure caused by force majeure, technical failure or other unforeseen events.

[0222] In the embodiments provided by this invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and actual implementations may have other classification methods.

[0223] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0224] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0225] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0226] Furthermore, it is clear that the word "including" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or systems described in a system can also be implemented by a single unit or system through software or hardware. Second-class terms are used to indicate names and do not indicate any specific order.

[0227] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A system for constructing an early warning model for Alzheimer's disease (AD) based on multidimensional testing, characterized in that, The system includes: The data acquisition module is used to collect neuroimaging data, biomarker data, and clinical characteristic data of each of multiple subjects identified by subject tags, to obtain a neuroimaging dataset, a biomarker dataset, and a clinical characteristic dataset, wherein the subject tags are healthy subjects or diseased subjects. The data processing module is used to screen for abnormal data in the neuroimaging dataset to obtain a preprocessed neuroimaging dataset, including: Align multiple neuroimaging data points in the neuroimaging dataset with a pre-constructed standard template space to obtain a standardized neuroimaging dataset, which includes multiple standardized neuroimaging data points. Intensity normalization was performed on all standardized neuroimaging data in the standardized neuroimaging dataset to obtain a normalized neuroimaging dataset. The normalized neural image dataset is denoised to obtain the denoised normalized neural image dataset. Anomaly screening is performed on the denoised normalized neural image dataset to obtain one or more abnormal image data. The one or more abnormal image data are then removed from the denoised normalized neural image dataset to obtain a preprocessed neural image dataset. Data cleaning was performed on the biomarker dataset to obtain a preprocessed biomarker dataset, including: Missing data detection is performed on the biomarker dataset to obtain a missing data detection report; Based on the missing data detection report, the missing data was identified, and the missing data was imputed using a pre-built random forest imputation method to obtain a complete biomarker dataset. Feature signals are extracted from the complete biomarker dataset to obtain the original biomarker sequence set; The original biomarker sequence set was analyzed using a pre-constructed local outlier factor algorithm to calculate multiple local outlier factor values. The calculation formula for the local outlier factor algorithm is shown below: ； in, Represents the data points of the original biomarker sequence set. The corresponding local outlier factor value, This represents data points in the original biomarker sequence set. This represents the preset neighborhood value. Indicates the first Individual and The nearest original biomarker sequence neighbor, Indicates the first Neighboring points Locally achievable density, express Locally achievable density; Threshold determination is performed on all the local outlier factor values ​​to obtain an outlier detection report; Extreme outliers were identified based on the outlier detection report, and the extreme outliers were corrected using the random forest filling method to obtain the corrected biomarker dataset. The modified biomarker dataset was normalized to obtain the preprocessed biomarker dataset; The clinical feature dataset is encoded to obtain a preprocessed clinical feature dataset, including: The clinical feature dataset is classified to obtain a cumulative feature set and a judgment feature set. Continuous encoding is performed on multiple cumulative features in the cumulative feature set to obtain a continuous variable dataset; Discrete encoding is performed on multiple judgment features in the judgment feature set to obtain a discrete variable dataset; The continuous variable dataset and the discrete variable dataset are merged to obtain the clinical feature coding dataset; The clinical feature encoding dataset is normalized to obtain a preprocessed clinical feature dataset; Feature extraction was performed on the preprocessed neural image dataset and the preprocessed biomarker dataset to obtain the neural image feature set and the biomarker feature set, respectively. The weighted fusion module is used to construct a multidimensional feature matrix set using neuroimaging feature sets, biomarker feature sets, and preprocessed clinical feature datasets; Randomly sample from the multidimensional feature matrix set to obtain a subset of the training set matrix and a subset of the test set matrix, and use the subset of the training set matrix and the subset of the test set matrix to obtain a weighted multidimensional feature matrix set; The model building module is used to train a pre-built clinical prediction model using a weighted multidimensional feature matrix set to obtain an early warning experimental model, test the early warning experimental model, and obtain test evaluation results. The early warning experimental model was optimized based on the test and evaluation results to obtain the early warning model, including: The prediction accuracy is obtained using the test evaluation results. This prediction accuracy is then compared to a preset accuracy threshold. If the prediction accuracy is confirmed to be lower than the threshold, a pre-built adaptive learning rate adjustment algorithm is used to calculate the optimized learning rate. The calculation formula is shown below: ； in, This indicates optimizing the learning rate. This represents the learning rate of the early warning experimental model. Represents the sensitivity coefficient. Indicates the magnification factor. Indicates the accuracy threshold. Indicates prediction accuracy. Represents the forgetting factor coefficient. This represents the learning rate from the previous round, used during the first optimization. =0, Represents the hyperbolic tangent function; The learning rate of the early warning experimental model is adjusted using the optimized learning rate. The adjusted early warning experimental model is then used as the early warning experimental model. Multiple weighted multidimensional feature matrices are input into the early warning experimental model to obtain test evaluation results. This process continues until the prediction accuracy is greater than or equal to the accuracy threshold, at which point the early warning experimental model is confirmed as an early warning model.

Images