Alzheimer's disease risk assessment system, method and apparatus based on aβ oligomers, apo e ε4 gene and multi-domain risk factors

Abstract

This invention provides an Alzheimer's disease risk assessment system, comprising: a data acquisition module for acquiring test data of subjects, the test data including at least the concentration of Aβ oligomers, APOEε4 gene carrier status data, and multi-domain risk factor data; the multi-domain risk factor data including demographic indicators and clinical data related to Alzheimer's disease risk; a risk assessment module, communicatively connected to the data acquisition module, for taking the test data from the data acquisition module as input, and outputting a risk probability value of the subject having Alzheimer's disease through a pre-trained machine learning model; and a result output module, communicatively connected to the risk assessment module, for comparing the risk probability value with a preset threshold and outputting risk classification information based on the comparison result. This invention is the first to achieve high-precision and high-sensitivity clinical cognitive impairment screening by fusing multidimensional conventional data through a nonlinear model.

Description

Technical Field

[0001] This invention relates to the field of biomedical technology, and in particular to an Alzheimer's disease risk assessment system, method and device based on Aβ oligomers, APOEε4 gene and multi-domain risk factors. Background Technology

[0002] Alzheimer's disease (AD) is a neurodegenerative disease with an insidious onset. Currently, there is no effective cure for advanced AD. Newer therapies approved in recent years, such as monoclonal antibodies targeting β-amyloid protein, primarily target patients in the early stages of the disease or even the preclinical phase. However, due to the subtle nature of early symptoms, they are often mistaken for normal aging, causing many patients to miss the optimal intervention period. Therefore, developing an accurate and cost-effective early screening tool applicable to a large population has become an urgent need in the field of AD prevention and treatment.

[0003] Currently, early screening and diagnosis of Alzheimer's disease (AD) mainly rely on the following types of technologies, but they all have significant limitations when used for large-scale screening: Neuropsychological screening scales and their limitations: Primary screening commonly employs neurocognitive scales such as the Mini-Mental State Examination (MMSE) or the Montreal Cognitive Assessment (MoCA). While these scales are easy to use, they have inherent limitations as screening tools: (1) Insufficient sensitivity: They are not sensitive to early-stage mild cognitive impairment (MCI), especially mild cognitive decline in highly educated individuals; (2) Poor specificity: Their results cannot etiologically differentiate between Alzheimer's disease (AD), vascular cognitive impairment, or other types of dementia; (3) Susceptibility to interference: The reliability and validity fluctuate significantly due to the influence of education level, cultural background, emotional state during testing, and the subjective judgment of the assessor on severely affected subjects. More importantly, these screening methods are time-consuming and labor-intensive, requiring professional personnel, and are difficult to meet the practical needs of large-scale rapid initial screening in primary healthcare and communities.

[0004] Advances and challenges in peripheral blood biomarker detection: In recent years, research on blood biomarkers has made groundbreaking progress. Indicators such as Tau protein phosphorylated at threonine-181 / 217 sites (p-tau181, p-tau217) and glial fibrillary acidic protein (GFAP) have shown extremely high diagnostic value in differentiating AD pathology from other neurodegenerative diseases, and their performance is approaching that of cerebrospinal fluid (CSF) analysis and even amyloid PET imaging. However, there are still huge challenges in transforming these cutting-edge biomarkers into universal screening tools: (1) High technical threshold and cost: Their accurate detection relies heavily on high-end platforms such as Simoa technology and mass spectrometry. These devices are expensive and complex to operate, making them difficult to popularize in primary hospitals; (2) The detection performance still needs to be uniformly verified: Although many studies have reported the excellent performance of biomarkers such as p-tau217, there are differences in sensitivity and specificity reported by different research cohorts and different detection platforms. The results are not uniform, and the optimal cutoff value still needs to be further verified and standardized in prospective, large-scale community cohorts.

[0005] Limitations of the gold standard diagnostic method: According to the 2024 National Institute on Aging-Alzheimer's Disease Association (NIA-AA) criteria, cerebrospinal fluid (CSF) biomarker analysis and amyloid positron emission tomography (PET) scans are the reference standards for diagnosing the biological pathology of Alzheimer's disease (AD). However, their inherent properties prevent them from being used for screening: CSF testing involves invasive lumbar puncture, which carries discomfort and risks of infection; Amyloid PET scans are extremely expensive and require complex technical support and professional interpretation personnel. These technologies are only available in a few top-tier medical centers, making them completely inaccessible and unsuitable as first-line screening tools.

[0006] In summary, a significant challenge currently exists in the field of early Alzheimer's disease (AD) screening: on the one hand, simple and easy-to-use cognitive scales have low sensitivity and poor specificity; on the other hand, precise detection methods (including blood p-tau and PET / CSF) are unsuitable for large-scale screening scenarios due to either high technical costs or the potential to identify a large number of preclinical individuals. Therefore, a new technological solution is urgently needed. Summary of the Invention

[0007] In view of the shortcomings of the prior art described above, this invention provides an Alzheimer's disease risk assessment system, method, and device based on Aβ oligomers (AβO), the APOEε4 gene, and multi-domain risk factors. The purpose of this invention is to construct a predictive model directly targeting the current risk of Alzheimer's disease, rather than simply detecting AD pathology. Such a tool can more accurately identify the groups most in need of immediate clinical intervention and referral, thereby achieving efficient allocation of limited medical resources. This invention cleverly integrates readily available clinical information with low-cost, widely accessible blood indicators.

[0008] like Figure 1 As shown, the first aspect of the present invention provides an Alzheimer's disease risk assessment system, the system comprising: The data acquisition module 11 is used to acquire the test data of the subjects. The test data includes at least the concentration of Aβ oligomers, APOEε4 gene carrier status data, and multi-domain risk factor data. The multi-domain risk factor data includes demographic indicators and clinical data related to the risk of Alzheimer's disease. The risk assessment module 12 is communicatively connected to the data acquisition module and is used to take the detection data of the data acquisition module as input, and output the risk probability value of the subject having Alzheimer's disease through a pre-trained machine learning model. The result output module 13 is communicatively connected to the risk assessment module and is used to compare the risk probability value with a preset threshold and output risk classification information based on the comparison result.

[0009] This invention provides a nonlinear fusion machine learning screening system and method based on serum Aβ oligomers (AβO), APOEε4 carrier status, and multi-domain risk factor data. It uses a trained machine learning model to capture the complex interactions between a single blood biomarker (Aβ oligomer) and readily available data, achieving low-cost, high-sensitivity prediction of Alzheimer's disease (AD). This technical solution has been successfully deployed as an online web application at: https: / / ad-screening-app-az0909.streamlit.app / Furthermore, the demographic and clinical data related to Alzheimer's disease risk include one or more of the following: age, sex, years of education, body mass index, history of alcohol consumption, family history of dementia, hypertension, diabetes, hyperlipidemia, and depressive state.

[0010] Preferably, the machine learning model is a non-linear machine learning model.

[0011] The nonlinear machine learning model involved in this invention can be one or more of the following: decision tree, support vector machine, random forest, lightweight gradient booster, extreme gradient booster, k-nearest neighbor algorithm, multilayer perceptron classifier, and Gaussian Naive Bayes classifier model.

[0012] Nonlinear machine learning models significantly outperform traditional linear models and baseline models. This invention innovatively uses machine learning to model Aβ oligomers, APOEε4 carrier status, and multi-domain risk factor data (age, sex: female, body mass index, years of education, alcohol consumption history, family history of dementia, hypertension, diabetes, hyperlipidemia, and depressive state). A systematic comparison of nine different algorithms revealed, as shown in Table 1, that nonlinear models generally performed better, traditional linear models (such as logistic regression) performed second best, and conventional baseline models (such as those using only age + APOE4) performed the worst. This finding confirms the complex nonlinear interactions between Aβ oligomers and other clinical characteristics, and that using nonlinear models is a necessary technique for achieving high-precision prediction.

[0013] Table 1. The ability of the 12 models to distinguish between cognitively normal and cognitively impaired.

[0014] More preferably, the machine learning model is a random forest model. The random forest model of the present invention has demonstrated excellent screening capabilities in both the total population and key subgroups. Combining the results of AUC and sensitivity, the present invention finally selected the random forest model, which showed an extremely high discriminative power of 0.9279 (95% confidence interval: 0.8854 - 0.9589) on an independent test set of 192 people.

[0015] The model also demonstrated an AUC of 0.8828 when distinguishing between MCI patients and cognitively normal patients, and an AUC of 0.8840 when distinguishing between patients diagnosed with AD pathology by Amyloid PET or CSF and cognitively normal patients. At the Youden index (0.3969), the model achieved a sensitivity of 93.3% and a specificity of 78.8%.

[0016] Furthermore, in the result output module, when the risk probability value is not less than a preset threshold, "non-low risk" classification information is output; when the risk probability value is less than the preset threshold, "low risk" classification information is output.

[0017] Preferably, the preset threshold is between 0.38 and 0.45.

[0018] More preferably, the preset threshold is 0.3969 or 0.43.

[0019] This threshold is derived by training the system with 764 samples in the training set.

[0020] On the other hand, preferably, the present invention employs a "dual threshold" strategy to optimize clinical triage efficiency. Dual thresholds (i.e., finding 90% sensitivity and 90% specificity as upper and lower limits respectively) are used as feasible value ranges.

[0021] In the result output module, the preset thresholds include a first preset threshold and a second preset threshold, where the first preset threshold is less than the second preset threshold. When the risk probability value is less than the first preset threshold, "low risk" classification information is output; when the risk probability value is greater than the second preset threshold, "high risk" classification information is output; and when the risk probability value is neither less than the first preset threshold nor greater than the second preset threshold, "medium risk" classification information is output. The medium risk zone represents the early risk zone of Alzheimer's disease (AD).

[0022] Preferably, the first preset threshold value ranges from 0.38 to 0.45, and the second preset threshold value ranges from 0.55 to 0.61.

[0023] More preferably, the first preset threshold is 0.3969 or 0.43, and the second preset threshold is 0.6.

[0024] Specifically, such as Figure 18 and Figure 19 As shown, preset exclusion thresholds (Low Cutoff) and confirmation thresholds (High Cutoff) are defined: The exclusion threshold (first preset threshold) is set at 0.4300. This setting is based on the clinical goal of ensuring high sensitivity (sensitivity ≥ 90%), aiming to minimize missed diagnoses. At this threshold, test set data shows a negative predictive value (NPV) as high as 94.74%, meaning that subjects with scores below this value (approximately 60% of the population) can be confidently classified as "low-risk," thus avoiding further expensive PET or cerebrospinal fluid examinations and significantly saving medical resources.

[0025] The diagnostic threshold (second preset threshold) was set at 0.6000. This threshold was determined based on the clinical goal of ensuring high specificity (≥ 90%), aiming to accurately identify high-risk individuals. At this threshold, test set data showed a specificity of 90.15% and a positive predictive value (PPV) of 77.19%. Subjects with scores higher than this value (approximately 30% of the population) were considered "high-risk" and were strongly recommended for referral to a specialist for gold-standard diagnosis.

[0026] Gray zone (the area between the exclusion threshold and the diagnosis threshold): Probability values ​​between 0.4300 and 0.6000 are defined as the gray zone. Test data shows that approximately 10.9% of the subjects fall into this range. For this group, the system outputs a "moderate risk or follow-up recommended" message, indicating an early risk zone for Alzheimer's disease (AD). It is recommended to use the Mini-Cognitive Assessment (MCA) scale for further assessment and to have regular follow-ups.

[0027] The random forest model used in this embodiment, combined with the above-mentioned dual-threshold strategy, successfully diverted 89.1% of the subjects to a clearly defined "low-risk" or "high-risk" group.

[0028] Furthermore, the system also includes a data preprocessing module, which is communicatively connected to the data acquisition module, for preprocessing the raw data of the detection data so that the detection data can be used as input to the risk assessment module.

[0029] Preferably, the preprocessing includes imputing missing values ​​in the original data of the detection data, and / or standardizing the continuous feature data in the original data to generate a standardized feature vector.

[0030] Optionally, a pre-trained Iterative Imputer model can be used to impute any missing items in the original data.

[0031] Optionally, the continuous feature data in the original data can be standardized using a fitted Z-score standardization (StandardScaler) model. Continuous features can be one or more of age, body mass index, Aβ oligomer concentration, and years of education.

[0032] Optionally, the system may further include a data input module, which is communicatively connected to the data preprocessing module, for inputting the raw data of the detection data.

[0033] Preferably, the Alzheimer's disease risk assessment is an early Alzheimer's disease risk assessment.

[0034] The machine learning model is trained on a training dataset, where each training sample's data includes: Aβ oligomer concentration, APOEε4 gene carrier status data, multi-domain risk factor data, and a binary label characterizing whether the individual corresponding to the training sample has Alzheimer's disease; the multi-domain risk factor data includes demographic indicators and clinical data related to Alzheimer's disease risk.

[0035] The risk probability value in the risk assessment module is a continuous probability value between 0.0 and 1.0. This probability value represents the individual's current risk of having Alzheimer's disease (AD).

[0036] The data on Aβ oligomers were obtained from serum samples of the subjects.

[0037] The APOEε4 gene carrier status data were obtained from the subjects' blood samples.

[0038] like Figure 2 As shown, a second aspect of the present invention provides a method for assessing the risk of Alzheimer's disease, the method comprising the following steps: S1, Obtain the subject's test data, which includes at least the concentration of Aβ oligomers, APOEε4 gene carrier status data, and multi-domain risk factor data; the multi-domain risk factor data includes demographic indicators and clinical data related to Alzheimer's disease risk. S2, taking the detection data described in step S1 as input, and using a pre-trained machine learning model, outputs a risk probability value representing the subject's Alzheimer's disease. S3, compare the risk probability value with a preset threshold, and output risk classification information based on the comparison result.

[0039] Furthermore, the demographic and clinical data related to Alzheimer's disease risk include one or more of the following: age, sex, years of education, body mass index, history of alcohol consumption, family history of dementia, hypertension, diabetes, hyperlipidemia, and depressive state.

[0040] The assessment results provided by this invention can serve as an auxiliary reference tool for clinicians, but are not directly equivalent to clinical diagnosis. The Alzheimer's disease risk assessment method described in this invention is for non-disease diagnosis and treatment purposes.

[0041] Preferably, the machine learning model is a non-linear machine learning model.

[0042] The nonlinear machine learning model involved in this invention can be one or more of the following: decision tree, support vector machine, random forest, lightweight gradient booster, extreme gradient booster, k-nearest neighbor algorithm, multilayer perceptron classifier, and Gaussian Naive Bayes classifier model.

[0043] Further preferably, the machine learning model is a random forest model. The random forest model of this invention demonstrates excellent screening capabilities in both the total population and key subgroups. Considering both AUC and sensitivity results, this invention ultimately selected the random forest model, which exhibited extremely high discriminative power of 0.9279 (95% confidence interval: 0.8854–0.9589) on an independent test set of 192 individuals. Simultaneously, it was found that the model achieved an AUC of 0.8828 when distinguishing between MCI patients and the cognitively normal group (e.g., ...). Figure 11 (D); In distinguishing patients diagnosed with AD pathology by Amyloid PET or CSF from cognitively normal patients, the AUC reached 0.8840. At the Youden index (0.3969), the model achieved 93.3% sensitivity and 78.8% specificity. (e.g. Figure 11 As shown in E).

[0044] Furthermore, in step S3, when the risk probability value is not less than a preset threshold, “non-low risk” classification information is output; when the risk probability value is less than the preset threshold, “low risk” classification information is output.

[0045] Preferably, the preset threshold is between 0.38 and 0.45.

[0046] More preferably, the preset threshold is 0.3969 or 0.43.

[0047] This threshold is derived by training the system with 764 samples in the training set.

[0048] On the other hand, preferably, the present invention employs a "dual threshold" strategy to optimize clinical triage efficiency. Dual thresholds (i.e., finding 90% sensitivity and 90% specificity as upper and lower limits respectively) are used as feasible value ranges.

[0049] In step S3, the preset threshold includes a first preset threshold and a second preset threshold, wherein the first preset threshold is less than the second preset threshold; when the risk probability value is less than the first preset threshold, “low risk” classification information is output; when the risk probability value is greater than the second preset threshold, “high risk” classification information is output; when the risk probability value is not less than the first preset value and not greater than the second preset threshold, “medium risk” classification information is output.

[0050] Preferably, the first preset threshold value ranges from 0.38 to 0.45, and the second preset threshold value ranges from 0.55 to 0.61.

[0051] More preferably, the first preset threshold is 0.3969 or 0.43, and the second preset threshold is 0.6.

[0052] Furthermore, the method also includes the following steps: S11, preprocessing the raw data of the detection data so that the detection data of step S1 can be used in step S2.

[0053] Preferably, the preprocessing includes imputing missing values ​​in the original data of the detection data, and / or standardizing the continuous feature data in the original data to generate a standardized feature vector.

[0054] Optionally, a pre-trained Iterative Imputer model can be used to impute any missing items in the original data.

[0055] Optionally, the continuous feature data in the original data can be standardized using a fitted Z-score standardization (StandardScaler) model. Continuous features can be one or more of age, body mass index, Aβ oligomer concentration, and years of education.

[0056] Preferably, the Alzheimer's disease risk assessment is an early Alzheimer's disease risk assessment.

[0057] The machine learning model is trained on a training dataset, where each training sample's data includes: Aβ oligomer concentration, APOEε4 gene carrier status data, multi-domain risk factor data, and a binary label characterizing whether the individual corresponding to the training sample has Alzheimer's disease; the multi-domain risk factor data includes demographic indicators and clinical data related to Alzheimer's disease risk.

[0058] The risk probability value in step S2 is a continuous probability value between 0.0 and 1.0. This probability value represents the individual's current risk of having AD.

[0059] The data on Aβ oligomers were obtained from serum samples of the subjects.

[0060] The APOEε4 gene carrier status data were obtained from the subjects' blood samples.

[0061] like Figure 3 As shown, a third aspect of the present invention provides a method for constructing an Alzheimer's disease risk assessment model, comprising at least the following steps: SS1, obtain training sample data, which includes: Aβ oligomer concentration, APOEε4 gene carrier status data, and multi-domain risk factor data, as well as a binary label characterizing whether the individual corresponding to the training sample has Alzheimer's disease; the multi-domain risk factor data includes demographic indicators and clinical data related to Alzheimer's disease risk; SS2 uses the training sample data obtained from SS1 to build a machine learning model.

[0062] Furthermore, the demographic and clinical data related to Alzheimer's disease risk include one or more of the following: age, sex, years of education, body mass index, history of alcohol consumption, family history of dementia, hypertension, diabetes, hyperlipidemia, and depressive state.

[0063] This invention confirms that a predictive model constructed using only one feature of serum Aβ oligomers, such as... Figure 4 As shown, when distinguishing between cognitively normal subjects and those with Alzheimer's disease (AD), the area under the receiver operating characteristic (AUC) curve on the test set reached 0.8604 (95% confidence interval: 0.8018 - 0.9100). This indicates that Aβ oligomers are a potent and independent biomarker of clinical cognitive status, but also suggests that relying solely on Aβ oligomers is insufficient for achieving high-precision screening.

[0064] Preferably, the machine learning model is a non-linear machine learning model.

[0065] The nonlinear machine learning model involved in this invention can be one or more of the following: decision tree, support vector machine, random forest, lightweight gradient booster, extreme gradient booster, k-nearest neighbor algorithm, multilayer perceptron classifier, and Gaussian Naive Bayes classifier.

[0066] More preferably, the machine learning model is a random forest model.

[0067] Furthermore, the method also includes the following steps: SS11, preprocessing the original data of the training sample data so that the training sample data of step SS1 can be used in step SS2.

[0068] Preferably, the preprocessing includes imputing missing values ​​in the original data of the training sample data, and / or standardizing the continuous feature data in the original data to generate standardized feature vectors.

[0069] Optionally, a pre-trained Iterative Imputer model can be used to impute any missing items in the original data.

[0070] Optionally, the continuous feature data in the original data can be standardized using a fitted Z-score standardization (StandardScaler) model. Continuous features can be one or more of age, body mass index, Aβ oligomer concentration, and years of education.

[0071] The data on Aβ oligomers were obtained from serum samples of the subjects.

[0072] The APOEε4 gene carrier status data were obtained from the subjects' blood samples.

[0073] Total Aβ in the blood includes various forms such as Aβ monomers (mainly Aβ1-40 and Aβ1-42), oligomers (Aβ oligomers), and fibrils.

[0074] There is a certain correlation between the increase in Aβ oligomers and the increase in total Aβ levels, but the relationship between the two is quite complex, specifically reflected in the following two aspects.

[0075] Aβ oligomers are part of the total Aβ pool. If the total amount of Aβ produced and released into the bloodstream by the brain or other tissues increases, the concentration of Aβ oligomers, as an intermediate form, may also increase. In this case, an increase in total Aβ may indicate a synchronous increase in oligomer levels.

[0076] The Aβ molecule is not a static mixture in the body, but a dynamic equilibrium system:

[0077] In the early stages of Alzheimer's disease, total Aβ production may not increase significantly. The more crucial change may lie in a decreased Aβ clearance capacity or an increased tendency for Aβ to aggregate. This leads to Aβ monomers more easily aggregating into oligomers. During this process, total Aβ levels may not show significant fluctuations, or even decrease due to oligomers more readily depositing in brain tissue or forming insoluble aggregates that escape the "detectable pool." Because Aβ oligomers are far more neurotoxic than monomeric and fibrillary forms, even if total Aβ levels remain stable, a shift in equilibrium towards oligomers will significantly enhance their toxic effects. This is the main basis for the current academic community's view of Aβ oligomers as a key pathogenic factor.

[0078] In summary, while there is a correlation between increased Aβ oligomer levels and increased total Aβ levels, Aβ oligomer levels are primarily determined by the dynamic aggregation process of Aβ molecules, rather than simply by their total amount. In the pathological progression of Alzheimer's disease, an increase in the proportion of Aβ oligomers (even if changes in total Aβ levels are not significant) is considered a core mechanism driving neurotoxicity.

[0079] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the program is executed by a processor, it implements the aforementioned Alzheimer's disease risk assessment method and / or the aforementioned method for constructing an Alzheimer's disease risk assessment model.

[0080] A fifth aspect of the present invention provides a computer processing device, including a processor and the aforementioned computer-readable storage medium, wherein the processor executes a computer program on the computer-readable storage medium to implement the aforementioned Alzheimer's disease risk assessment method, and / or the steps of the aforementioned method for constructing an Alzheimer's disease risk assessment model.

[0081] A sixth aspect of the present invention provides an electronic terminal, comprising: a processor, a memory, and a communicator; the memory is used to store a computer program, the communicator is used to communicate with an external device, and the processor is used to execute the computer program stored in the memory to cause the terminal to execute the aforementioned Alzheimer's disease risk assessment method, and / or the aforementioned method for constructing an Alzheimer's disease risk assessment model.

[0082] As described above, the Alzheimer's disease risk assessment system, method, and device based on Aβ oligomers, APOEε4 gene, and multi-domain risk factors of the present invention have the following beneficial effects: (I) High Precision and Innovative Strategy: This invention overcomes the subjectivity of traditional scale assessments. Test set evaluation results show that the 12-feature random forest model of this invention achieved excellent performance with an AUC of 0.9279 (95% confidence interval: 0.8854 - 0.9589). This high precision is not achieved by simply piling up risk factors, but rather through a nonlinear machine learning model, achieving for the first time a deep, nonlinear fusion between a single blood biomarker (Aβ oligomer) and multidimensional clinical information. This fusion can capture the synergistic amplification effect of complex combinations such as "advanced age and carrying the APOEε4 gene with high Aβ oligomer concentration" on risk, which is impossible with traditional linear models (such as logistic regression) or single biomarkers. Therefore, this performance is significantly better than the simple model using only Aβ oligomers (AUC = 0.8604), and even better than the conventional risk model of age + APOEε4 carrier status (AUC = 0.6858), breaking through the bottleneck of single-indicator performance and producing synergistically enhanced screening efficacy.

[0083] (II) Model Advancement, High Sensitivity, and High Practicality: The random forest model used in this invention inherently possesses advantages such as handling nonlinear relationships, automatically assessing feature importance, and insensitivity to outliers and noise, ensuring high stability of the model when facing complex data from community populations. An AUC value exceeding 0.92 on the independent test set strongly demonstrates that the model is not overfitting the training data, but rather possesses strong generalization ability, effectively applicable to new and unseen individuals. The model exhibits superior performance in identifying the early stages of Alzheimer's disease (MCI). As shown in the technical validation section, the model achieved excellent identification accuracy (AUC > 0.88) in subgroups including MCI patients and Alzheimer's disease patients diagnosed by Amyloid PET / CSF. This not only confirms its high sensitivity to Alzheimer's pathology but also highlights its powerful ability to accurately capture early clinical signals.

[0084] (III) Minimally Invasive, Convenient, and Scalable: This screening system boasts a dual advantage in convenience. At the core level, its key biomarker, Aβ oligomer, can be obtained simply through routine venous blood sampling (based on a basic ELISA platform). Compared to invasive lumbar puncture and the complex procedures of Amyloid PET, the operation is extremely simple and minimally invasive, significantly improving subject acceptance and compliance. At the system level, this invention combines this minimally invasive blood test with 10 routine clinical and demographic characteristics that can be quickly obtained through simple inquiries, greatly lowering the barrier to promotion and eliminating the need for cutting-edge equipment in top medical centers or highly scarce professional personnel. The system's final output is not a difficult-to-interpret probability number, but rather a binary classification result of "high risk" or "low risk" based on the optimal clinical decision threshold (0.3969) determined in advance through the Youden index, delivered directly through a publicly accessible online application. This provides primary care physicians with a clear and intuitive basis for decision-making, avoiding the subjectivity and uncertainty of result interpretation, making it highly suitable for large-scale promotion and application by non-specialist physicians in community and primary healthcare institutions.

[0085] (iv) Systematization, Standardization, and Interpretability: This invention is a complete decision support system that minimizes human error. The system has fully solidified and automated the data preprocessing (including missing value imputation and data standardization) and model prediction processes. Users only need to input the raw data, and the system can automatically complete all subsequent complex calculations and output the final risk classification. This design eliminates errors introduced by manual operation or calculation mistakes, ensuring the objectivity and repeatability of the screening results. Attached Figure Description

[0086] Figure 1 This is an embodiment of the Alzheimer's disease risk assessment system of the present invention.

[0087] Figure 2 This is an embodiment of the Alzheimer's disease risk assessment method of the present invention.

[0088] Figure 3 This is a method for constructing an Alzheimer's disease risk assessment model according to an embodiment of the present invention.

[0089] Figure 4 This is the receiver operating characteristic curve (ROC) of serum Aβ in distinguishing between cognitively normal individuals and those with AD cognitive impairment.

[0090] Figure 5 This is the calibration curve for the random forest model.

[0091] Figure 6 It is a decision curve analysis of the random forest model (DCA).

[0092] Figure 7 It is a SHAP dependency graph with twelve features.

[0093] Figure 8 This is a schematic diagram of an electronic terminal according to an embodiment of the present invention.

[0094] Figure 9 The descriptive distribution (A) and ROC curve (B) of serum Aβ oligomers are shown.

[0095] Figure 10 It compares the ROC curves of all machine learning models and the baseline model on the independent test set.

[0096] Figure 11 This involves the clinical utility and performance validation of the random forest model. (A. Confusion Matrix) This section demonstrates the model's specific classification performance at the optimal classification threshold.

[0097] Content Interpretation: True Negative (Top Left 104): 104 cognitively normal subjects were correctly predicted as normal by the model. False Positive (Top Right 28): 28 normal subjects were incorrectly predicted as having the disease. False Negative (Bottom Left 4): Only 4 patients with cognitive impairment (MCI / AD) were incorrectly predicted as normal (this is the most critical indicator of the screening tool, with an extremely low false negative rate). True Positive (Bottom Right 56): 56 patients were correctly identified by the model. Clinical Significance: The calculated sensitivity of the model is as high as 93.3% (56 / 60), and the specificity is 78.8% (104 / 132). This indicates that the model is very suitable as an "exclusionary screening tool" because it rarely misses true patients.

[0098] B. Decision Curve Analysis (DCA) This section assesses the model's net benefit in clinical practice.

[0099] Content Interpretation: Green line (Random Forest Model): Represents the net benefit of using this model for screening. Yellow dashed line (Treat All): Assumes everyone needs intervention / further testing. Gray dashed line (Treat None): Assumes no intervention for anyone. Clinical Significance: The green line is significantly higher than the yellow and gray lines over a wide range of probability thresholds. This means that using this model to determine who needs further testing, compared to "testing everyone" or "not testing anyone," provides the greatest net benefit to patients and the healthcare system (i.e., benefiting patients by identifying those who need further testing without increasing unnecessary testing).

[0100] C. Calibration Curve This section assesses whether the probabilities predicted by the model are accurate and reliable.

[0101] Content Interpretation: X-axis: Average probability predicted by the model. Y-axis: Actual observed positive rate. Diagonal dashed line: Represents perfect calibration (predicted probability = actual probability). Purple line: Actual performance of the model. Clinical significance: The purple line is very close to the diagonal dashed line, and the Brier Score is only 0.1086 (lower is better). This indicates that the model is neither severely overconfident nor underconfident. For example, when the model predicts a 70% risk of disease, approximately 70% of people actually develop the disease.

[0102] D. ROC curves of the model distinguishing between MCI patients and cognitively normal groups in an independent test set; ROC curves of the MCI subgroup (ROC Curve: MCI vs. CU). This part validates the model's performance in early, mild cases.

[0103] Interpretation: This figure only compares patients with mild cognitive impairment (MCI) with cognitively normal individuals (CU). AUC = 0.883 (95% CI: 0.791–0.956). Clinical significance: Even in the milder and more difficult-to-diagnose stage of MCI, the model maintains a high discriminative ability. This demonstrates the model's potential for early screening.

[0104] E. ROC curves of the model distinguishing between pathologically confirmed AD groups and cognitively normal groups on independent test sets; ROC curves of biomarker-confirmed subgroups (ROC Curve: PET / CSF Confirmed vs. CU). This part validates the model's performance in the "gold standard" of confirmed cases.

[0105] Content Interpretation: This figure compares patients with confirmed AD pathology via PET scan or cerebrospinal fluid (CSF) testing with cognitively normal individuals. AUC = 0.884 (95% CI: 0.796–0.947). Clinical Significance: This is a crucial validation. It demonstrates that the model not only distinguishes between "people with cognitive symptoms" but also specifically identifies patients with the biological pathology of Alzheimer's disease.

[0106] Figure 12 This is a SHAP summary diagram of a random forest model used to predict cognitive impairment.

[0107] Figure 13 It is the SHAP main effect dependency graph of 12 features in the random forest model.

[0108] Figure 14 This is a SHAP interaction diagram between Aβ oligomers and age.

[0109] Figure 15 This is a flowchart of the Alzheimer's disease risk assessment method according to an embodiment of the present invention.

[0110] Figure 16 This is verified by digital western blot.

[0111] Figure 17 This is a comparison of the Aβ oligomer content in the peripheral blood and cerebral cortex tissue of APP / PS1 mice and control mice.

[0112] Figure 18 X-axis (Threshold Probability): The probability of the disease predicted by the model (0.0 to 1.0), Y-axis (Metric Value): The values ​​of each evaluation indicator (0.0 to 1.0, i.e., 0% to 100%).

[0113] Curve Meanings: Solid green line (Sensitivity): Sensitivity. Decreases as the threshold increases; Solid blue line (Specificity): Specificity. Increases as the threshold increases; Dashed light green line (NPV): Negative predictive value. Remains at a very high level in the low threshold region; Dashed light blue line (PPV): Positive predictive value. Increases as the threshold increases. Vertical dashed lines (Cutoffs): Green dotted vertical line (Low Cutoff): Set at 0.43; Blue dotted vertical line (High Cutoff): Set at 0.60. Low Risk / Rule-out Zone – Green background on the left; High Risk / Rule-in Zone – Red background on the right; Grey Zone / Indeterminate Zone – Yellow background in the middle.

[0114] Figure 19 Threshold division between non-AD and AD. Detailed Implementation

[0115] The research in this invention is based on the National Key Research and Development Program of China, "Research on the Mechanism and Intervention of Psychological-Sleep-Cognitive Interaction in the Aging Process" (2023YFC36003200, 2023YFC3603201).

[0116] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0117] Before further describing specific embodiments of the present invention, it should be understood that the scope of protection of the present invention is not limited to the specific embodiments described below; it should also be understood that the terminology used in the embodiments of the present invention is for describing specific embodiments and not for limiting the scope of protection of the present invention; in the specification and claims of the present invention, unless otherwise expressly stated in the text, the singular forms "a", "an" and "this" include the plural forms.

[0118] When numerical ranges are given in the embodiments, it should be understood that, unless otherwise stated in the present invention, both endpoints of each numerical range and any value between the two endpoints may be selected. Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. In addition to the specific methods, apparatus, and materials used in the embodiments, based on the knowledge of the prior art possessed by one of ordinary skill in the art and the description of this invention, any prior art methods, apparatus, and materials similar to or equivalent to those described, apparatus, and materials in the embodiments of this invention may be used to implement the present invention.

[0119] Unless otherwise stated, the experimental methods, detection methods, and preparation methods disclosed in this invention all employ conventional techniques in molecular biology, biochemistry, chromatin structure and analysis, analytical chemistry, cell culture, recombinant DNA technology, and related fields.

[0120] Since the systems and methods in this invention are based on the same principle, the definitions, calculation methods, implementation methods, and preferred methods of the same features can be used interchangeably and will not be repeated.

[0121] This invention confirms the predictive value of Aβ oligomers and, for the first time, achieves high-precision (AUC > 0.92) and high-sensitivity (>93%) clinical cognitive impairment screening by fusing multidimensional conventional data through a nonlinear model.

[0122] The reliability of the system described in this invention has also been fully verified: such as Figure 5 The calibration curve (Brill score = 0.1086) shows that the probability values ​​output by the model are accurate and reliable. Figure 6 The decision curve analysis (DCA) shows that the model demonstrates a higher net benefit than the "all-treatment" or "no-treatment" strategies at almost all relevant clinical thresholds. Regarding interpretability, as... Figure 7 As shown, Shapley Additive exPlanations reveal that Aβ oligomers, family history of dementia, and APOE ε4 carrier status are strong risk predictors, allowing physicians to intuitively understand the specific factors leading to a patient's high risk.

[0123] It should be noted that the division of modules in the system of this invention is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. These modules can be implemented entirely in software through processing element calls; they can also be fully implemented in hardware; or some modules can be implemented by processing element calls to software, while others are implemented in hardware. For example, the acquisition module can be a separate processing element, or it can be integrated into a chip. Alternatively, it can be stored in memory as program code, and its functions can be called and executed by a processing element. The implementation of other modules is similar. Furthermore, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element mentioned here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above modules can be completed through the integrated logic circuits in the hardware of the processor element or through software instructions.

[0124] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more digital signal processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs, or Graphics Processing Units, GPUs). As another example, when a module is implemented through processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together to form a system-on-a-chip (SOC).

[0125] like Figure 8 The diagram illustrates an electronic terminal provided by the present invention. The electronic terminal includes a processor 31, a memory 32, a communicator 33, a communication interface 34, and a system bus 35. The memory 32 and the communication interface 34 are connected to the processor 31 and the communicator 33 via the system bus 35 and communicate with each other. The memory 32 stores computer programs, the communicator 34 and the communication interface 34 communicate with other devices, and the processor 31 and the communicator 33 run the computer programs, enabling the electronic terminal to execute the various steps of the image analysis method described above.

[0126] The system bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. The communication interface is used to enable communication between the database access device and other devices (e.g., clients, read-write libraries, and read-only libraries). Memory may include Random Access Memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.

[0127] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), graphics processing units (GPUs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0128] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented using hardware related to a computer program. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; the computer-readable storage medium may include, but is not limited to, floppy disks, optical disks, CD-ROMs (Read-Only Optical Disk Memory), magneto-optical disks, ROMs (Read-Only Memory), RAMs (Random Access Memory), EPROMs (Erasable Programmable Read-Only Memory), EEPROMs (Electrically Erasable Programmable Read-Only Memory), magnetic cards or optical cards, flash memory, or other types of media / machine-readable media suitable for storing machine-executable instructions. The computer-readable storage medium can be a product not connected to a computer device or a component used in a computer device.

[0129] In practice, the computer program is a routine, program, object, component, data structure, etc., that performs a specific task or implements a specific abstract data type.

[0130] The baseline characteristics of the study cohort (the baseline characteristics of the entire dataset, which were subsequently divided into training and test sets) were analyzed. The study cohort included 958 participants, stratified by diagnosis into a cognitively normal (CU) group (N=658) and a cognitively impaired (CI) group (N=298). Significant differences were observed between the two diagnostic groups; compared to the CU group, the CI group had a higher mean age, fewer years of education, and exhibited a higher proportion of APOEε4 carriers and depressive symptoms (Table 2). This complete cohort was subsequently divided into a training set (N=770) and an independent test set (N=192). The stratification was confirmed to be balanced, with standardized mean differences (SMD) of all features between the two groups well below 0.1.

[0131] Table 2. Baseline demographics and clinical characteristics of the study cohort

[0132] Independent diagnostic performance of serum Aβ oligomers (ABO) Given its central role, we first assessed the distribution of the core biomarker, serum Aβ oligomers (Aβ oligomers). Distribution ( Figure 9 A) showed that the original Aβ oligomer level (median 134.9 pg / ml) was significantly higher in the CI group compared with the CU group (median 58.0 pg / ml). To quantify its independent diagnostic utility, we trained an Aβ oligomer-only logistic regression model on the training set and evaluated it on the independent test set. This independent biomarker model showed strong discriminative power, with an area under the curve (AUC) of 0.860 (95% CI: 0.802–0.910), achieving 85% sensitivity and 77% specificity at the optimal probability cutoff point (0.50). Figure 9 (B).

[0133] Development and performance comparison of machine learning models After evaluating the Aβ oligomer univariate model, we developed and compared 10 state-of-the-art machine learning (ML) models integrating all 12 pre-specified features, and compared them with two baseline logistic regression (LR) models (Table 3). The 12-feature model significantly outperformed both baseline models (Baseline LR (Age+APOE4) AUC = 0.686; Baseline LR (Aβ oligomers Only) AUC = 0.860). The superiority of this multi-feature approach integrating Aβ oligomers and clinical data is evident in the ROC curve comparison (Table 3). Figure 10 This was intuitively confirmed in the data. The best-performing models on the independent test set were Support Vector Machine (SVM) (AUC = 0.930) and Random Forest (RF) (AUC = 0.928). The Random Forest (RF) model was chosen as the final model for all subsequent in-depth analyses due to its combination of high accuracy, robust cross-validation performance, and excellent interpretability. This model demonstrated excellent generalization ability, with its test set AUC (0.928) showing only a slight difference compared to the training set AUC (0.985), indicating minimal overfitting, and its performance was statistically significantly better than the baseline model (DeLong test, p < 0.001).

[0134] Table 3. Performance comparison of machine learning models in cognitive impairment classification

[0135] In-depth validation of the selected random forest model The selected 12-feature RF model underwent a comprehensive validation process on an independent test set. Figure 11 For classification performance at the optimal Youden index threshold (0.397), please refer to the confusion matrix. Figure 11 The matrix (A) shows that the model correctly classified 104 out of 132 patients in the CU group (specificity 78.8%) and 56 out of 60 patients in the CI group (sensitivity 93.3%). Decision curve analysis (DCA) demonstrated its significant clinical applicability, as the RF model consistently provided a net benefit higher than the "all treatment" or "no treatment" strategies across a broad range of clinically relevant threshold probabilities. Figure 11 (B). Furthermore, the model was shown to have good calibration accuracy, with its calibration curve closely fitting the diagonal of perfect calibration, supported by a low Brier score (0.109) and a high BSS (0.494). Figure 11(C). This high level of reliability contrasts with the poor calibration of other high AUC models such as SVM. The model's robustness was demonstrated in key clinical subgroups, both in differentiating MCI patients (AUC = 0.883) Figure 11 (D) or is it distinguishing patients diagnosed by PET / CSF (AUC = 0.884) Figure 11 Both the E and CU control groups showed excellent distinguishing ability.

[0136] The final model was explained using SHAP analysis. We employ SHAP (SHapley Additive exPlanations) analysis on the test set to explain the model's decision-making process. Global Feature Importance Summary Plot ( Figure 12 The results showed that serum Aβ oligomers were the most influential single feature to date (mean |SHAP| = 0.207), with a driving force far exceeding that of second-tier features (including BMI, age, APOEε4 carrier status, and years of education). Notably, the effects of some traditional clinical risk factors (such as hypertension, hyperlipidemia, and alcohol consumption) on the model output were almost negligible. The main effect dependency plot of SHAP for the 12 features is shown in the figure. Figure 13 These relationships were further clarified, showing the direction and magnitude of each feature's contribution. The analysis confirmed that higher Aβ oligomer levels, older age, and female sex were associated with increased risk, while higher education levels were protective. These plots also revealed complex nonlinear patterns, such as the U-shaped association between BMI and risk, confirming the model's ability to capture synergistic relationships beyond simple linear associations. This synergy was clearly visualized in a SHAP interaction plot, showing that the positive effect of Aβ oligomers on risk was amplified in older individuals. Figure 14 ).

[0137] Example 1 like Figure 15 As shown, this embodiment includes the following steps: Data Collection: During community health checkups, subject Y (70 years old) was assessed, and 12 characteristic data required for this invention were collected: Age: 70; Gender: Female; Body Mass Index (BMI): 24; Years of Education: 9 (years); Serum Aβ oligomer: 125.0 (pg / mL); APOEε4 carrier status: Yes; History of alcohol consumption: Yes; Family history of dementia: None; Hypertension: No; Diabetes: No; Hyperlipidemia: No; Depression status: No.

[0138] The 12 core features incorporated into this system model were collected through the following two main approaches: Blood Collection and Analysis (2 core features): A blood sample is collected from a routine venous blood draw and used for the following two key tests: Serum Aβ oligomer assay: Approximately 5 mL of venous blood was collected from the subject and injected into a serum coagulation tube. After blood collection, the tube was left to stand at room temperature (approximately 20-25°C) for 30-60 minutes to allow the blood to coagulate fully. The coagulated sample tube was then placed in a centrifuge and centrifuged at a relative centrifugal force of 1500-1800 xg for 10-15 minutes at room temperature. The supernatant after centrifugation is the serum. The serum was carefully aspirated, avoiding contact with the lower clot. The separated serum was aliquoted into polypropylene cryovials and immediately frozen at -80°C until analysis. Before testing, the sample tubes were thawed in a 37°C water bath for 15 minutes, and then the serum sample was used to determine the Aβ oligomer concentration using an enzyme-linked immunosorbent assay (ELISA).

[0139] APOE ε4 carrier status determination (using anticoagulant tubes): During the same blood collection session, collect approximately 3-5 mL of venous blood and inject it into an EDTA anticoagulant tube (usually a purple-tipped tube). Immediately after blood collection, gently invert and mix 8-10 times to prevent clotting. Centrifuge the EDTA anticoagulant tube at room temperature at a relative centrifugation force of approximately 1500 xg for 10-15 minutes. After centrifugation, the sample separates into three layers: an upper plasma layer, a thin middle layer of leukocytes, and a lower layer of erythrocytes. Carefully aspirate the upper plasma layer and collect the leukocyte layer from which genomic DNA is extracted. Use the extracted DNA for APOE genotyping to determine whether the individual is an APOE ε4 carrier.

[0140] Routine clinical and demographic data (10 features in total): Obtained through inquiry: Demographics (4 items): age, sex, years of education, body mass index.

[0141] Clinical and lifestyle factors (6 items): history of alcohol consumption, family history of dementia, hypertension, diabetes, hyperlipidemia, and depressive state (based on the Geriatric Depression Scale-15, a score greater than 4 indicates a depressive state).

[0142] Inputting data: Enter the collected subject data on the online webpage https: / / ad-screening-app-az0909.streamlit.app / .

[0143] Data Preprocessing: Input the data of “Subject Y” into the data preprocessing module. Missing Value Imputation (if missing values ​​exist during data collection): If a variable is missing, the system will start IterativeImputer and use the regression model of the remaining 10 features to estimate the value of the missing variable; Standardization: The system starts StandardScaler and applies the mean and standard deviation saved during training to the four continuous features of age (70), body mass index (BMI) (24), serum Aβ oligomer: 125.0, and years of education: 9, converting them into standardized values.

[0144] Risk Prediction: The 12 preprocessed feature vectors are input into the risk prediction module (a trained random forest model). After calculation, the model outputs a probability value. Output risk probability: 61.42%.

[0145] Decision Support and Explanation 1: The system presets the first threshold to be 0.3969 and the second threshold to be 0.6. The risk probability of subject Y (0.6142) is compared with the preset two thresholds: 0.6142 > 0.6. Since the probability value is higher than the diagnostic threshold, the system determines that "subject Y" falls into the red high-risk zone and outputs the classification information as "AD high risk".

[0146] Decision Support and Explanation 2: The system presets a first threshold of 0.38 and a second threshold of 0.55. The risk probability of subject Y (0.6142) is compared with the preset thresholds: 0.6142 > 0.55. Since the probability value is higher than the diagnostic threshold, the system determines that "subject Y" falls into the red high-risk zone and outputs the classification information as "AD high risk".

[0147] Decision Support and Explanation 3: The system presets a first threshold of 0.45 and a second threshold of 0.61. The risk probability of subject Y (0.6142) is compared with the preset dual thresholds: 0.6142 > 0.61. Since the probability value is higher than the diagnostic threshold, the system determines that "subject Y" falls into the red high-risk zone and outputs the classification information as "AD high risk".

[0148] Decision Support and Explanation 4: The system presets a first threshold of 0.43 and a second threshold of 0.6. The risk probability of subject Y (0.6142) is compared with the preset dual thresholds: 0.6142 > 0.6. Since the probability value is higher than the diagnostic threshold, the system determines that "subject Y" falls into the red high-risk zone and outputs the classification information as "AD high risk".

[0149] Example 2 like Figure 15 As shown, this embodiment includes the following steps: Data Collection: During a community health check, subject "Subject X" (65 years old) was assessed, and 12 characteristic data required for this invention were collected: Age: 65; Gender: Female; Body Mass Index (BMI): 22; Years of Education: 12 (years); Serum Aβ oligomer: 50.0 (pg / mL); APOEε4 carrier status: No; History of alcohol consumption: Yes; Family history of dementia: None; Hypertension: No; Diabetes: No; Hyperlipidemia: No; Depression status: No.

[0150] The 12 core features incorporated into this system model were collected through the following two main approaches: Blood Collection and Analysis (2 core features): A blood sample is collected from a routine venous blood draw and used for the following two key tests: Serum Aβ oligomer assay: Approximately 5 mL of venous blood was collected from the subject and injected into a serum coagulation tube. After blood collection, the tube was left to stand at room temperature (approximately 20-25°C) for 30-60 minutes to allow the blood to coagulate fully. The coagulated sample tube was then placed in a centrifuge and centrifuged at a relative centrifugal force of 1500-1800 xg for 10-15 minutes at room temperature. The supernatant after centrifugation is the serum. The serum was carefully aspirated, avoiding contact with the lower clot. The separated serum was aliquoted into polypropylene cryovials and immediately frozen at -80°C until analysis. Before testing, the sample tubes were thawed in a 37°C water bath for 15 minutes, and then the serum sample was used to determine the Aβ oligomer concentration using an enzyme-linked immunosorbent assay (ELISA).

[0151] APOE ε4 carrier status determination (using anticoagulant tubes): During the same blood collection session, collect approximately 3-5 mL of venous blood and inject it into an EDTA anticoagulant tube (usually a purple-tipped tube). Immediately after blood collection, gently invert and mix 8-10 times to prevent clotting. Centrifuge the EDTA anticoagulant tube at room temperature at a relative centrifugation force of approximately 1500 xg for 10-15 minutes. After centrifugation, the sample separates into three layers: an upper plasma layer, a thin middle layer of leukocytes, and a lower layer of erythrocytes. Carefully aspirate the upper plasma layer and collect the leukocyte layer from which genomic DNA is extracted. Use the extracted DNA for APOE genotyping to determine whether the individual is an APOE ε4 carrier.

[0152] Routine clinical and demographic data (10 features in total): Obtained through inquiry: Demographics (4 items): age, sex, years of education, body mass index.

[0153] Clinical and lifestyle factors (6 items): history of alcohol consumption, family history of dementia, hypertension, diabetes, hyperlipidemia, and depressive state (based on the Geriatric Depression Scale-15, a score greater than 4 indicates a depressive state).

[0154] Inputting data: Enter the collected subject data on the online webpage https: / / ad-screening-app-az0909.streamlit.app / .

[0155] Data Preprocessing: Input the data of “Subject X” into the data preprocessing module. Missing Value Imputation (if missing values ​​exist during data collection): If a variable is missing, the system will start IterativeImputer and use the regression model of the remaining 10 features to estimate the value of the missing variable; Standardization: The system starts StandardScaler and applies the mean and standard deviation saved during training to the four continuous features of age (65), body mass index (BMI) (22), serum Aβ oligomer: 50.0, and years of education: 12, converting them into standardized values.

[0156] Risk prediction result: The 12 preprocessed feature vectors are input into the risk prediction module (a trained random forest model). After calculation, the model outputs a probability value. Output risk probability: 8.78%.

[0157] Decision Support and Explanation 1: Dual-threshold classification decision: The system presets the first threshold to be 0.3969 and the second threshold to be 0.6. The risk probability of subject X (8.78%) is compared with the preset dual thresholds: 8.78% < 0.3969, the system determines that "subject X" is classified as "low risk of AD".

[0158] Decision Support and Explanation 2: The system presets a first threshold of 0.38 and a second threshold of 0.55. The risk probability of subject X (8.78%) is compared with the preset thresholds: 8.78% < 0.38, therefore the system determines that subject X's classification information is "low risk of AD".

[0159] Decision Support and Explanation 3: The system presets a first threshold of 0.45 and a second threshold of 0.61. The risk probability of subject X (8.78%) is compared with the preset thresholds: 8.78% < 0.45, therefore the system determines that subject X's classification information is "low risk of AD".

[0160] Decision Support and Explanation 4: The system presets a first threshold of 0.43 and a second threshold of 0.61. The risk probability of subject X (8.78%) is compared with the preset thresholds: 8.78% < 0.43, therefore the system determines that subject X's classification information is "low risk of AD". Example 3

[0161] like Figure 15 As shown, this embodiment includes the following steps: Data Collection: During a community health check, subject Z (72 years old) was assessed, and 12 characteristic data required for this invention were collected: Age: 72; Gender: Female; Body Mass Index (BMI): 21; Years of Education: 6 (years); Serum Aβ oligomer: 90.0 (pg / mL); APOEε4 carrier status: No; History of alcohol consumption: None; Family history of dementia: None; Hypertension: No; Diabetes: No; Hyperlipidemia: No; Depression status: No.

[0162] The 12 core features incorporated into this system model were collected through the following two main approaches: Blood Collection and Analysis (2 core features): A blood sample is collected from a routine venous blood draw and used for the following two key tests: Serum Aβ oligomer assay: Approximately 5 mL of venous blood was collected from the subject and injected into a serum coagulation tube. After blood collection, the tube was left to stand at room temperature (approximately 20-25°C) for 30-60 minutes to allow the blood to coagulate fully. The coagulated sample tube was then placed in a centrifuge and centrifuged at a relative centrifugal force of 1500-1800 xg for 10-15 minutes at room temperature. The supernatant after centrifugation is the serum. The serum was carefully aspirated, avoiding contact with the lower clot. The separated serum was aliquoted into polypropylene cryovials and immediately frozen at -80°C until analysis. Before testing, the sample tubes were thawed in a 37°C water bath for 15 minutes, and then the serum sample was used to determine the Aβ oligomer concentration using an enzyme-linked immunosorbent assay (ELISA).

[0163] APOE ε4 carrier status determination (using anticoagulant tubes): During the same blood collection session, collect approximately 3-5 mL of venous blood and inject it into an EDTA anticoagulant tube (usually a purple-tipped tube). Immediately after blood collection, gently invert and mix 8-10 times to prevent clotting. Centrifuge the EDTA anticoagulant tube at room temperature at a relative centrifugation force of approximately 1500 xg for 10-15 minutes. After centrifugation, the sample separates into three layers: an upper plasma layer, a thin middle layer of leukocytes, and a lower layer of erythrocytes. Carefully aspirate the upper plasma layer and collect the leukocyte layer from which genomic DNA is extracted. Use the extracted DNA for APOE genotyping to determine whether the individual is an APOE ε4 carrier.

[0164] Routine clinical and demographic data (10 features in total): Obtained through inquiry: Demographics (4 items): age, sex, years of education, body mass index.

[0165] Clinical and lifestyle factors (6 items): history of alcohol consumption, family history of dementia, hypertension, diabetes, hyperlipidemia, and depressive state (based on the Geriatric Depression Scale-15, a score greater than 4 indicates a depressive state).

[0166] Inputting data: Enter the collected subject data on the online webpage https: / / ad-screening-app-az0909.streamlit.app / .

[0167] Data Preprocessing: Input the data of “Subject X” into the data preprocessing module. Missing Value Imputation (if missing values ​​exist during data collection): If a variable is missing, the system will start IterativeImputer and use the regression model of the remaining 10 features to estimate the value of the missing variable; Standardization: The system starts StandardScaler and applies the mean and standard deviation saved during training to the four continuous features of age (72), body mass index (BMI) (21), serum Aβ oligomer: 90.0, and years of education: 6, converting them into standardized values.

[0168] Risk prediction result: The 12 preprocessed feature vectors are input into the risk prediction module (a trained random forest model). After calculation, the model outputs a probability value. Output risk probability: 44.13%.

[0169] Decision Support and Explanation 1: Dual-threshold classification decision: The system presets the first threshold to be 0.3969 and the second threshold to be 0.6. The risk probability of subject Z (44.13%) is compared with the preset dual thresholds: 44.13% > 0.3969 and < 0.6. The system determines that the classification information for "subject Z" is "medium risk of AD, or early risk of AD".

[0170] Decision Support and Explanation 2: The system presets the first threshold to be 0.38 and the second threshold to be 0.55. The risk probability of subject Z (44.13%) is compared with the preset two thresholds: 44.13% > 0.38 and < 0.55. The system determines that the classification information for "subject Z" is "moderate risk of AD" or "early risk of AD".

[0171] Decision Support and Explanation 3: The system presets the first threshold to be 0.44 and the second threshold to be 0.61. The risk probability of subject Z (44.13%) is compared with the preset two thresholds: 44.13% < 0.44, so the system determines that the classification information for "subject Z" is "low risk of AD".

[0172] Decision Support and Explanation 4: The system presets the first threshold to be 0.43 and the second threshold to be 0.61. The risk probability of subject Z (44.13%) is compared with the preset two thresholds: 44.13% > 0.43 and < 0.61. The system determines that the classification information for "subject Z" is "moderate risk of AD" or "early risk of AD".

[0173] The specific explanation of the preset threshold is as follows: Phase 1: Low-risk zone (probability value < 0.43) Definition: Cognitive normality and high confidence zone.

[0174] Explanation: The predicted probability of the subjects was lower than the low cutoff value for AD (0.43). Within this range, the model excludes the risk of AD with high specificity.

[0175] Clinical recommendation: No intervention is needed; routine regular check-ups are recommended.

[0176] Phase 2: Medium risk, or early risk zone of AD (0.43 to 0.6000) Definition: A clearly defined area of ​​cognitive impairment.

[0177] explain: Lower limit: The score has exceeded the high cutoff value of AD (0.43), indicating that the model is confident that the subject is no longer cognitively normal.

[0178] Upper limit: The score has not yet reached the high cutoff value for Alzheimer's disease (0.600), indicating that it has not yet developed into the typical Alzheimer's dementia stage. It is very likely to be prodromal Alzheimer's disease.

[0179] Clinical recommendation: Clinical intervention is necessary. This is the optimal window for drug trials or lifestyle interventions to delay the progression to Alzheimer's disease (AD).

[0180] Phase 3: High-risk area for Alzheimer's disease (probability value > 0.6000) Definition: AD high confidence region.

[0181] Explanation: The subject's score exceeded the high cutoff value for AD. The model predicted that the subject had AD with high confidence.

[0182] Clinical recommendations: A complete neuropsychological evaluation and biomarker confirmation (such as PET or cerebrospinal fluid) are recommended, and a standard treatment protocol for Alzheimer's disease (AD) should be initiated.

[0183] Clinical Recommendation: Traditional Approach: Subject Y or Subject Z may be classified as normal on the scale due to mild or absent symptoms, leading to screening failure. This Invention: This system provides an objective high-risk indication driven by Aβ oligomers, a core pathological substance in Alzheimer's disease (AD). Physicians should recommend further neuropsychological evaluation for Subject Y or Subject Z and refer them to a specialist clinic. Simultaneously, interpretability analysis is invoked: The system utilizes the Shapley incremental interpretation module to reveal to physicians that high Aβ oligomer levels, advanced age, APOEε4 carrier status, female sex, and a positive alcohol consumption history are the main driving factors increasing their risk probability. Physicians can then intervene in the patient's case based on this information.

[0184] This embodiment detects soluble oligomers (dimers to dodecamers) formed by the polymerization of Aβ monomers. Peripheral serum from AD patients was analyzed using automated digital western blot capillary electrophoresis. The results are as follows: Figure 16 The oligomer component is mainly composed of the Aβ1-42 fragment. Since Aβ1-42 has two more hydrophobic amino acids than Aβ1-40, it is more hydrophobic and easier to aggregate. Therefore, it is considered to be the most toxic subtype of oligomers.

[0185] The advantages of the testing technology include: Using a patented antibody that can specifically recognize Aβ oligomers as the core raw material, combined with a unique repeat epitope sandwich method, high specificity and high sensitivity detection of Aβ oligomers are achieved (authorized patent number: ZL 202510139096.8).

[0186] Validation experiments conducted in AD model mice (APP / PS1) Experimental Methods: 1. Three 6-month-old male APP / PS1 AD model mice and three wild-type control mice were selected. Blood was collected from the heart after anesthesia, and an anticoagulant was added. The blood was centrifuged, and the supernatant plasma was collected. Simultaneously, brain tissue was removed, and 1 mL of RIPA strong lysis agent (containing protease inhibitors and phosphatase inhibitors) was added. The tissue was homogenized using a Tissue Lyser II tissue homogenizer at 30 Hz for 8 min, centrifuged at 14000 rpm for 30 min at 4℃, and the supernatant was collected as brain tissue homogenate. 2. Using the prepared detection kit, Aβ oligomers in the plasma and brain tissue homogenates of AD mice and control mice were detected, respectively. The content of Aβ oligomers in the peripheral blood and brain tissue of AD model mice was significantly higher than that in control mice. The detection results are as follows: Figure 17 As shown, the levels of Aβ oligomers in the blood and cerebral cortex of the mice were significantly higher than those in the control mice, indicating that the detection method based on peripheral blood Aβ oligomers can effectively reflect relevant changes in brain tissue.

[0187] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any form or substance. It should be noted that those skilled in the art can make various improvements and additions without departing from the method of the present invention, and these improvements and additions should also be considered within the scope of protection of the present invention. Any modifications, alterations, and equivalent changes made by those skilled in the art based on the above-disclosed technical content without departing from the spirit and scope of the present invention are equivalent embodiments of the present invention. Furthermore, any modifications, alterations, and evolutions made to the above embodiments based on the essential technology of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. An Alzheimer's disease risk assessment system, characterized in that, The system includes: The data acquisition module is used to acquire the test data of the subjects. The test data includes at least the concentration of Aβ oligomers, APOEε4 gene carrier status data, and multi-domain risk factor data. The multi-domain risk factor data includes demographic indicators and clinical data related to Alzheimer's disease risk. The risk assessment module is communicatively connected to the data acquisition module and is used to take the detection data of the data acquisition module as input, and output the risk probability value of the subject having Alzheimer's disease through a pre-trained machine learning model. The result output module is communicatively connected to the risk assessment module and is used to compare the risk probability value with a preset threshold and output risk classification information based on the comparison result.

2. The Alzheimer's disease risk assessment system as described in claim 1, characterized in that, It also includes one or more of the following features: a. The demographic and clinical data related to Alzheimer's disease risk include one or more of the following: age, sex, years of education, body mass index, history of alcohol consumption, family history of dementia, hypertension, diabetes, hyperlipidemia, and depressive state. b. The machine learning model is a non-linear machine learning model; c. In the result output module, when the risk probability value is not less than a preset threshold, "non-low risk" classification information is output; When the risk probability value is less than a preset threshold, "low risk" classification information is output; preferably, the preset threshold is 0.38 to 0.45; more preferably, the preset threshold is 0.3969 or 0.

43. d. In the result output module, the preset threshold includes a first preset threshold and a second preset threshold, wherein the first preset threshold is less than the second preset threshold; When the risk probability value is less than the first preset threshold, output "low risk" classification information; When the risk probability value is greater than the second preset threshold, output "high risk" classification information; When the risk probability value is not less than a first preset value and not greater than a second preset threshold, "medium risk" classification information is output; preferably, the value range of the first preset threshold is 0.38 to 0.45, and the value range of the second preset threshold is 0.55 to 0.61; more preferably, the first preset threshold is 0.3969 or 0.43, and the second preset threshold is 0.6; e. The system further includes a data preprocessing module, which is communicatively connected to the data acquisition module, for preprocessing the raw data of the detection data so that the detection data can be used as input to the risk assessment module.

3. The Alzheimer's disease risk assessment system as described in claim 2, characterized in that, In feature b, the machine learning model is one or more of the following: decision tree, support vector machine, random forest, lightweight gradient booster, extreme gradient booster, k-nearest neighbor algorithm, multilayer perceptron classifier, and Gaussian Naive Bayes classifier model; preferably, it is a random forest model.

4. The Alzheimer's disease risk assessment system as described in claim 2, characterized in that, It also includes one or more of the following features: 1) In feature d, the Alzheimer's disease risk assessment is an early risk assessment of Alzheimer's disease; 2) In feature e, the preprocessing includes imputing missing values ​​in the original data of the detection data, and / or standardizing the continuous feature data in the original data to generate a standardized feature vector.

5. The Alzheimer's disease risk assessment system as described in claim 1, characterized in that, The machine learning model is trained on a training dataset, where each training sample's data includes: Aβ oligomer concentration, APOEε4 gene carrier status data, multi-domain risk factor data, and a binary label characterizing whether the individual corresponding to the training sample has Alzheimer's disease; the multi-domain risk factor data includes demographic indicators and clinical data related to Alzheimer's disease risk.

6. A method for assessing the risk of Alzheimer's disease, characterized in that, The method includes the following steps: S1, Obtain the subject's test data, which includes at least the concentration of Aβ oligomers, APOEε4 gene carrier status data, and multi-domain risk factor data; the multi-domain risk factor data includes demographic indicators and clinical data related to Alzheimer's disease risk. S2, taking the detection data described in step S1 as input, and using a pre-trained machine learning model, outputs a risk probability value representing the subject's Alzheimer's disease. S3, compare the risk probability value with a preset threshold, and output risk classification information based on the comparison result.

7. A method for constructing an Alzheimer's disease risk assessment model, comprising at least the following steps: SS1, Obtain training sample data, the training sample data including: Aβ oligomer concentration, APOEε4 gene carrier status data, and multi-domain risk factor data, as well as binary labels characterizing whether the individuals corresponding to the training samples have Alzheimer's disease; the multi-domain risk factor data includes demographic indicators and clinical data related to Alzheimer's disease risk; SS2 uses the training sample data obtained from SS1 to build a machine learning model.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the Alzheimer's disease risk assessment method of claim 6, and / or the method for constructing an Alzheimer's disease risk assessment model of claim 7.

9. A computer processing apparatus, comprising a processor and the computer-readable storage medium of claim 8, characterized in that, The processor executes a computer program on the computer-readable storage medium to implement the Alzheimer's disease risk assessment method of claim 6, and / or the steps of the method for constructing an Alzheimer's disease risk assessment model of claim 7.

10. An electronic terminal, characterized in that, include: Processor, memory, and communication unit; The memory is used to store computer programs, the communicator is used to communicate with external devices, and the processor is used to execute the computer programs stored in the memory, so that the terminal executes the Alzheimer's disease risk assessment method according to claim 6, and / or the method for constructing an Alzheimer's disease risk assessment model according to claim 7.

Images