Method for detecting abnormal data of serum bile acid in PD cognitive function evaluation

Abstract

The invention discloses a method for detecting serum bile acid abnormal data in PD cognitive function evaluation, and relates to the technical field of biological data detection. The method comprises the following steps: on the basis of a serum bile acid detection data set, determining first N bile acids with the highest correlation with data abnormality by combining a multivariate analysis method to form a key bile acid set; the method comprises the following steps: by taking normal serum bile acid detection data as training data, constructing corresponding N univariate prediction channels for N bile acids in a key bile acid set; performing data reconstruction to obtain an anomaly analysis data set; establishing a distribution relationship between the anomaly analysis data set and N single-variable prediction channels, inputting the N anomaly analysis data clusters into the N single-variable prediction channels for iterative prediction, and obtaining N single-variable prediction results; and calculating a variable deviation coefficient, and if the variable deviation coefficient exceeds a preset deviation threshold, marking the data as abnormal data. According to the method, the accuracy of marking the abnormal data of the serum bile acid is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of biological data detection, specifically to a method for detecting abnormal serum bile acid data in PD cognitive function assessment. Background Technology

[0002] Parkinson's disease (PD) is a common neurodegenerative disease. Cognitive impairment (CI) is a common non-motor symptom of PD, often appearing in the early stages of the disease. Early identification and assessment of cognitive risk are crucial for clinical intervention and disease management. Changes in serum bile acid profiles in PD patients are closely related to cognitive impairment, and bile acids are expected to serve as an early biomarker for PD-related cognitive decline.

[0003] Currently, modern mass spectrometry technology can detect dozens of bile acids simultaneously, but the massive amount of data generated is difficult to utilize effectively. Traditional methods are susceptible to information redundancy and noise interference, and there is a risk of dimensionality curse and overfitting, making it extremely difficult to accurately extract key biomarkers from complex data. These limitations together restrict the precise application and promotion of this technology in clinical practice. Summary of the Invention

[0004] This application provides a method for detecting abnormal serum bile acid data in PD cognitive function assessment, aiming to solve the technical problem of low accuracy in the detection and labeling of abnormal serum bile acid data in the prior art.

[0005] In view of the above problems, this application provides a method for detecting abnormal serum bile acid data in PD cognitive function assessment, including: Based on the serum bile acid detection dataset, multivariate analysis was used to identify the top N bile acids that were most correlated with data anomalies, forming a key bile acid set, where N is a positive integer greater than 1. Using normal serum bile acid detection data as training data, N univariate prediction channels are constructed for each of the N bile acids in the key bile acid set, wherein each univariate prediction channel is constructed based on an ensemble learning method. Acquire serum bile acid detection data and reconstruct the data to obtain an anomaly analysis dataset, wherein the anomaly analysis dataset includes N anomaly analysis data clusters; Establish the allocation relationship between the anomaly analysis dataset and the N univariate prediction channels, and input the N anomaly analysis data clusters into the N univariate prediction channels for iterative prediction to obtain N univariate prediction results; The serum bile acid detection data is compared with the N univariate prediction results, and the corresponding variable deviation coefficient is calculated. If the variable deviation coefficient exceeds a preset deviation threshold, the serum bile acid detection data is marked as abnormal data.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: The key bile acid set screened through multivariate analysis focuses on biomarkers most strongly associated with cognitive function in PD, excluding irrelevant or confounding indicators. This makes the objectives of subsequent analyses clearer, and the results more biologically interpretable and clinically specific. The univariate prediction channel constructed using ensemble learning can learn the complex patterns of each bile acid under normal conditions, making it more sensitive to subtle but persistent abnormal changes. A multi-channel, collaborative intelligent detection channel is constructed, with each channel responsible for monitoring a key indicator. Finally, the decision is made by summarizing the variables based on their deviation coefficients, achieving an upgrade from point to surface, and providing a more comprehensive characterization of complex disease states. The entire process, from feature selection and model construction to anomaly detection, is largely automated, reducing over-reliance on expert experience. The high degree of standardization facilitates large-scale application and standardized assessment in clinical settings. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a flowchart illustrating the method for detecting abnormal serum bile acid data in PD cognitive function assessment provided in this application embodiment. Detailed Implementation

[0009] This application provides a method for detecting abnormal serum bile acid data in PD cognitive function assessment, which addresses the technical problem of low accuracy in labeling abnormal serum bile acid data in the prior art.

[0010] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0011] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0012] Examples, such as Figure 1 As shown, this application provides a method for detecting abnormal serum bile acid data in PD cognitive function assessment, the method comprising: S100: Based on the serum bile acid detection dataset, the top N bile acids with the highest correlation to data anomalies are determined by multivariate analysis, forming the key bile acid set, where N is a positive integer greater than 1.

[0013] In this embodiment, based on a serum bile acid detection dataset, multivariate analysis is used to determine the top N bile acids most highly correlated with data anomalies, forming a key bile acid set. By introducing a data anomaly index, cognitive impairment is treated as a continuous spectrum, enabling correlation analysis to capture more subtle biological changes related to severity, resulting in more clinically significant bile acids. By traversing all variables and quantifying their correlation ranking, the process is entirely driven by the data itself, avoiding subjective bias and increasing the likelihood of discovering new and unexpected key biomarkers. By selecting the top N most critical features, the complexity and computational cost of subsequent modeling are reduced.

[0014] Step S100 in the method provided in this application embodiment includes: Based on the serum bile acid detection dataset, abnormal serum bile acid detection data are extracted, and the corresponding data abnormality index is evaluated and labeled. Combining the abnormal serum bile acid detection data with the data abnormality index, a correlation analysis is performed on all M types of bile acids to obtain M correlation measures for the M types of bile acids, where M is a positive integer greater than or equal to N. M bile acids are sorted in descending order according to the M correlation measures, and the M correlation measures are cumulatively analyzed. The top N bile acids whose cumulative results satisfy the preset correlation measure threshold are selected and stored as the key bile acid set.

[0015] In this embodiment, firstly, abnormal serum bile acid detection data are extracted from the serum bile acid detection dataset, and corresponding data anomaly indices are evaluated and labeled. The data anomaly index refers to the degree of deviation of serum bile acid detection concentration from typical detection data. For example, all detection data from patients clinically diagnosed with PD cognitive impairment are identified from the dataset and labeled as abnormal serum bile acid detection data. For each abnormal data point, a specific data anomaly index is assigned based on the severity of cognitive impairment. For example, patient 1 has moderate cognitive impairment, and their data anomaly index is set to 85; patient 2 has mild cognitive impairment, and their data anomaly index is set to 60, and so on, scoring all abnormal samples.

[0016] Secondly, combining the abnormal serum bile acid detection data with the data anomaly index, a correlation analysis was performed on all M types of bile acids to obtain M correlation measures for the M types of bile acids, where M is a positive integer greater than or equal to N. The analysis determined which bile acid concentration change was most strongly correlated with the data anomaly index among all abnormal samples. For example, the Pearson correlation coefficient was used to calculate the correlation between the concentration of each bile acid and the data anomaly index. Assuming M=5, the following five correlation measures were obtained: correlation between bile acid A and the anomaly index: 0.75; correlation between bile acid B and the anomaly index: 0.10; correlation between bile acid C and the anomaly index: 0.82; correlation between bile acid D and the anomaly index: 0.08; correlation between bile acid E and the anomaly index: 0.71.

[0017] Finally, the M bile acids are sorted in descending order according to the M correlation measures, and a cumulative analysis is performed on the M correlation measures. The top N bile acids whose cumulative results meet the preset correlation measure threshold are selected and stored as the key bile acid set. Cumulative analysis is a screening strategy that accumulates the sorted correlation measures to ensure that the selected variable set represents most of the overall information. Continuing the previous example, the 5 bile acids are sorted from highest to lowest correlation: 0.82 > 0.75 > 0.71 > 0.10 > 0.08. Cumulative analysis: Calculate the cumulative correlation contribution. A correlation measure threshold is set, for example, the cumulative contribution rate reaches 80% of the total correlation. The sum of the absolute values ​​of the total correlation ≈ 0.82 + 0.75 + 0.71 + 0.10 + 0.08 = 2.46. The cumulative contribution of the first type is 0.82, accounting for approximately 33.3% (0.82 / 2.46) < 80%; the cumulative contribution of the first two types is 0.82 + 0.75 = 1.57, accounting for approximately 63.8% (1.57 / 2.46) < 80%; the cumulative contribution of the first three types is 1.57 + 0.71 = 2.28, accounting for approximately 92.7% (2.28 / 2.46) > 80%. Since the cumulative contribution rate of the first three bile acids has exceeded the 80% threshold, N=2 is set, and the final set of key bile acids selected is: {bile acid C, bile acid A}.

[0018] In this embodiment, the key bile acid set selected is the set that has been proven by data to be most relevant to disease severity, which can improve the clinical relevance of biomarkers and lay a solid foundation for subsequent accurate modeling. By focusing on the most critical variables, the whole process is less affected by irrelevant variables and noise, and the results are more stable. Doctors and researchers can clearly determine which bile acids play a major role, which facilitates biological interpretation and clinical decision-making.

[0019] S200: Using normal serum bile acid detection data as training data, N univariate prediction channels are constructed for each of the N bile acids in the key bile acid set, wherein each univariate prediction channel is constructed based on an ensemble learning method.

[0020] In this embodiment, normal serum bile acid detection data is used as training data. For each of the N bile acids in the key bile acid set, N corresponding univariate prediction channels are constructed, each constructed based on an ensemble learning method. Since the key bile acid set contains only N types, the number of individual normal samples may be limited, making direct modeling prone to overfitting. The meta-learning model utilizes information from all m types of bile acids to learn the general rules of bile acid prediction, providing prior knowledge for the base predictor and enabling it to maintain good generalization ability even with small samples. By generating replicas and randomly mutating them, model diversity is introduced while preserving the capabilities of the base predictor, providing a complementary foundation for subsequent ensemble learning. Valid replicas are selected using preset validation criteria, ensuring that all models within the base predictor cluster have satisfactory performance and avoiding interference from invalid models.

[0021] Step S200 in the method provided in this application embodiment includes: Based on the serum bile acid detection dataset, all M types of bile acids are extracted, and corresponding meta-learning paradigms are defined to construct a meta-learning model for variable prediction. Traverse the N bile acids in the set of key bile acids, and construct N base predictor clusters according to the meta-learning model; Based on the ensemble learning method, N base predictor clusters are integrated to construct corresponding N univariate prediction channels, wherein the prediction output of each univariate prediction channel includes the prediction values ​​of several base predictors in the base predictor clusters.

[0022] First, based on the serum bile acid detection dataset, all M types of bile acids are extracted, and a meta-learning paradigm is defined accordingly to construct a meta-learning model for variable prediction. For example, a control group is set up, and serum bile acid detection data from normal individuals are collected, which includes the concentrations of the aforementioned M=5 bile acids. The meta-learning paradigm is a learning-to-learn framework that enables the model to infer the normal level of a particular member based on the overall state of the bile acid metabolism network. Constructing the meta-learning model: A general model framework is designed, such as a multilayer perceptron. The input is the concentration of the other four bile acids besides the target bile acid, and the output is the predicted concentration of the target bile acid. For example, for a given normal sample, such as the blood sample of a healthy volunteer, the target bile acid C to be predicted is excluded, and the concentration values ​​of the remaining four bile acids are used as input features. For example, given the input: [Bile acid A = 1.2 μM, Bile acid B = 0.8 μM, Bile acid D = 0.1 μM, Bile acid E = 2.5 μM], the meta-learning model predicts the theoretical normal concentration of the target bile acid C under healthy metabolic conditions based on the concentrations of these four bile acids. The output of the meta-learning model is a specific, predicted concentration value. For example, the output might be: predicted bile acid C = 0.5 μM. This model is trained using data from tens of thousands of healthy individuals. For each healthy individual, the input includes the concentrations of bile acids A, B, D, and E, along with the actual measured concentration of bile acid C. Through extensive data training, the meta-learning model gradually learns the stable, intrinsic balance among bile acids A, B, D, E, and C in a healthy individual.

[0023] Secondly, the N bile acids in the set of key bile acids are traversed, and N base predictor clusters are constructed according to the meta-learning model.

[0024] The method provided in this application embodiment, which involves "traversing the N bile acids in the key bile acid set and constructing N base predictor clusters according to the meta-learning model," includes: Based on the key bile acid set and the meta-learning model, N base predictors corresponding to N bile acids are constructed. Several replica predictors are established for each of the N base predictors, and the replica predictors are randomly selected for random mutation. The predictive performance of the random mutation results is verified, and multiple replica predictors whose predictive performance verification results are passed are selected from the random mutation results. These replica predictors are then merged with the corresponding base predictors to output the base predictor clusters, and N base predictor clusters are obtained.

[0025] First, based on the key bile acid set and the meta-learning model, N base predictors corresponding to the N bile acids are constructed. For example, if the target bile acid is bile acid C, using all normal data, the other four bile acids besides bile acid C are used as input features, and the true concentration of bile acid C is used as the label to train the model, resulting in a trained model specifically for predicting the concentration of bile acid C, thus obtaining the bile acid C base predictor. Similarly, if the target bile acid is bile acid A, using all normal data, the other four bile acids besides bile acid A are used as input features, and the true concentration of bile acid A is used as the label to train the model, resulting in the bile acid A base predictor. Thus, N=2 base predictors are obtained.

[0026] Secondly, several replica predictors are established for each of the N base predictors, and these replica predictors are randomly selected for random mutation. Taking the bile acid C base predictor as an example, this illustrates how to construct a cluster of base predictors. First, five replicas of the bile acid C base predictor are created, with the same model structure and initial weights. Then, random mutation is performed, randomly selecting three replicas for mutation. The mutation method involves fine-tuning, i.e., training these replicas with a subset of normal data for a few extra rounds to slightly differentiate their parameters, or using Dropout, randomly shutting down some neurons in the network during prediction to introduce randomness.

[0027] Furthermore, the predictive performance of the random mutation results is validated, and multiple replica predictors that pass the validation are selected and merged with their corresponding base predictors to form the base predictor clusters, resulting in N base predictor clusters. The performance of these replicas is validated on a preserved normal data validation set, evaluating the predictive performance of all replicas, including both unmutated and mutated ones. The performance metric can be the mean squared error (MSE). Finally, filtering and merging are performed, retaining only those replicas whose predictive performance passes validation; for example, an MSE below a certain threshold indicates accurate prediction. Assuming 3 out of 5 replicas pass validation, these 3 high-performing replicas are merged with the original bile acid C base predictor to form the bile acid C base predictor cluster. The above process is repeated for bile acid A to obtain the bile acid A base predictor cluster. Ultimately, 2 base predictor clusters are obtained.

[0028] Finally, based on the ensemble learning method, N base predictor clusters are integrated to construct corresponding N univariate prediction channels. The prediction output of each univariate prediction channel includes the prediction values ​​of several base predictors within the base predictor clusters. Final prediction channels are constructed for bile acid C and bile acid A respectively. For bile acid C: the decision mechanism of its bile acid C univariate prediction channel is as follows: when a new bile acid C value needs to be predicted, the concentrations of the other four bile acids in the new sample are input into all four predictors in the bile acid C base predictor cluster, including one base predictor and three high-quality replicas, and the output values ​​of these four predictors are finally output. For example, the measured concentrations of the four bile acids other than bile acid C in the new sample are [1.5, 1.0, 0.2, 3.0]. This input vector is fed into each predictor in the bile acid C-based predictor cluster. Each of the four predictors performs its own calculation and outputs its independent prediction result, assumed to be [0.52, 0.55, 0.58, 0.55]. Similarly, for bile acid A, a univariate prediction channel for bile acid A is constructed, whose output is the integrated result of the outputs of all predictors in the bile acid A-based predictor cluster, assumed to be [1.48, 1.50, 1.52, 1.50]. Thus, N=2 univariate prediction channels have been successfully constructed.

[0029] In this embodiment, by establishing a personalized prediction benchmark, each univariate prediction channel no longer provides a fixed value for a population range, but rather dynamically predicts the individualized, theoretically normal value for a specific sample based on its unique bile acid background. This provides a precise reference for subsequent anomaly detection. Since each channel is an integrated system, its prediction result is the consensus of multiple high-quality models, reducing the impact of irrelevant fluctuations on the predicted value. When the true value of the target bile acid deviates significantly from the precise prediction value derived from the consensus, anomalies can be sensitively detected. This provides reliable input for subsequent calculations and effectively improves the accuracy of the final anomaly labeling.

[0030] S300: Acquire serum bile acid detection data and reconstruct the data to obtain an anomaly analysis dataset, wherein the anomaly analysis dataset includes N anomaly analysis data clusters.

[0031] In this embodiment, serum bile acid detection data is acquired and reconstructed to obtain an anomaly analysis dataset, which includes N anomaly analysis data clusters. Through data reconstruction, data in the required format is accurately input to each channel, ensuring that when analyzing any key bile acid, only information from other key bile acids is used, avoiding data leakage and making anomaly detection more rigorous and reliable. By creating independent data clusters for each key variable, isolated and parallel analysis can be achieved, which helps to further pinpoint which one or more key bile acids are abnormal.

[0032] Step S300 in the method provided in this application embodiment includes: Numerical extraction is performed based on the serum bile acid detection data, using the key bile acid set as an index. Establish N replica sets of numerical extraction results, and establish N anomaly analysis data clusters with N bile acids in the key bile acid set as single variables respectively; In the anomaly analysis data cluster, any one type of bile acid is used as the supervisor, and the corresponding N-1 types of bile acids are used as the sample input.

[0033] In this embodiment, firstly, numerical extraction is performed based on the serum bile acid detection data, using the key bile acid set as an index. Assuming a complete test report for a newly obtained serum sample is available, containing the measured concentrations of M=5 bile acids, the concentration values ​​of these two key bile acids are extracted from the report using the key bile acid set {bile acid C, bile acid A} as an index, resulting in a simplified data vector: [bile acid C = 0.80 μM, bile acid A = 1.60 μM]. This vector is the numerical extraction result.

[0034] Secondly, N replica sets of numerical extraction results are established, and N anomaly analysis data clusters are created, each using one of the N bile acids in the key bile acid set as a single variable. In each anomaly analysis data cluster, any one bile acid serves as the supervisor, and the corresponding N-1 bile acids serve as the sample input. For example, N=2 identical replicas are created for the above numerical extraction results [bile acid C=0.80μM, bile acid A=1.60μM], resulting in two identical data sets. N anomaly analysis data clusters are constructed, each using one of the N bile acids as a single variable, and a dedicated data package is prepared for analyzing anomalies in each key bile acid. For example, to construct an anomaly analysis data cluster for bile acid C: the true measured value of bile acid C is 0.80 μM, and the sample input key set contains all other bile acids besides bile acid C, i.e., {bile acid A}, with a value of 1.60 μM. Therefore, the anomaly analysis data cluster for bile acid C is {supervisory [bile acid C = 0.80], input [bile acid A = 1.60]}. Similarly, to construct an anomaly analysis data cluster for bile acid A: {supervisory [bile acid A] = 1.60, input [bile acid C = 0.80]}. Ultimately, N = 2 anomaly analysis data clusters are obtained, collectively forming the anomaly analysis dataset.

[0035] In this embodiment, the raw, chaotic full-spectrum detection data is reconstructed into a regular data stream that corresponds one-to-one with the prediction model. Each anomaly analysis data cluster contains unique information about that specific sample, making subsequent bias calculations highly targeted. Because the analysis is performed variable-by-variable and in isolation, when anomalies are finally labeled, it is possible to clearly locate which data cluster exhibits significant bias, thus providing doctors or researchers with clearer clues.

[0036] S400: Establish the allocation relationship between the anomaly analysis dataset and the N univariate prediction channels, and input the N anomaly analysis data clusters into the N univariate prediction channels for iterative prediction to obtain N univariate prediction results.

[0037] In this embodiment, an allocation relationship is established between the anomaly analysis dataset and N univariate prediction channels. The N anomaly analysis data clusters are then input into the N univariate prediction channels for iterative prediction to obtain N univariate prediction results. Continuing the previous example, the key bile acid set is: {bile acid C, bile acid A}; there are two anomaly analysis data clusters: Data Cluster 1: {supervisory [bile acid C=0.80], input [bile acid A=1.60]}; Data Cluster 2: {supervisory [bile acid A=1.60], input [bile acid C=0.80]}; and two trained univariate prediction channels: bile acid C prediction channel; bile acid A prediction channel. Pairing is performed according to the target variable: Data Cluster 1 is assigned to Channel 1, and Data Cluster 2 is assigned to Channel 2, establishing a clear one-to-one allocation relationship. Then, according to the allocation relationship, the input portion of each data cluster is sent to the corresponding prediction channel for prediction. Iteration 1: Process data cluster 1, extract the input [bile acid A = 1.60] from data cluster 1 and input it into channel 1. Channel 1 calculates and returns the predicted value of bile acid C based on the input value of bile acid A = 1.60. Assume the prediction result is: predicted bile acid C = 0.55. Iteration 2: Process data cluster 2, extract the input [bile acid C = 0.80] from data cluster 2 and input it into channel 2. Channel 2 calculates and returns the predicted value of bile acid A based on the input value of bile acid C = 0.80. Assume the prediction result is: predicted bile acid A = 1.50. Finally, N = 2 univariate prediction results are obtained: predicted bile acid C = 0.55, predicted bile acid A = 1.50.

[0038] In this embodiment, S400 is a crucial step that connects all the preceding preparatory work, initiating the actual analysis process. The obtained prediction results are not fixed population reference ranges, but rather theoretical normal values ​​calculated based on the unique bile acid profile of the specific sample, providing direct input for the final anomaly labeling. By establishing clear allocation relationships and performing iterative predictions, highly personalized theoretical normal values ​​are successfully generated for each key bile acid in the sample, providing a scientific and accurate comparison benchmark for the final anomaly judgment.

[0039] S500: Compare the serum bile acid detection data with the N univariate prediction results, and calculate the variable deviation coefficient accordingly. If the variable deviation coefficient exceeds the preset deviation threshold, mark the serum bile acid detection data as abnormal data.

[0040] In this embodiment, the deviation of each variable is not considered in isolation, but rather the deviations of all key variables are comprehensively considered to avoid misjudgment due to accidental fluctuations of a single variable. Different prediction channels have varying reliability, and different bile acids have different representativeness of abnormalities; therefore, differentiated weighting is used to assign different weights. The deviation threshold is derived from the distribution of the normal population, ensuring the objectivity and statistical significance of the abnormality judgment.

[0041] Step S500 in the method provided in this application embodiment includes: Based on the N univariate prediction results, establish the predicted value distributions corresponding to the N bile acids, and analyze the points with the maximum distribution density as the distribution centers corresponding to each predicted value distribution. The residuals between the predicted values ​​corresponding to the N distribution centers and the detected values ​​of the N bile acids in the serum bile acid detection data are calculated to obtain the original deviation data, wherein the original deviation data includes N residual values ​​corresponding to the N bile acids. The original deviation data are weighted and calculated to obtain the variable deviation coefficient; The variable deviation coefficient is compared with the preset deviation threshold to determine anomalies. If the variable deviation coefficient is greater than or equal to the deviation threshold, the serum bile acid detection data is marked as abnormal data.

[0042] In the method provided in this application embodiment, the deviation threshold is determined by the deviation coefficient distribution on the normal serum bile acid detection data.

[0043] First, based on the N univariate prediction results, a distribution of predicted values ​​for N bile acids is established, and the point of maximum distribution density is obtained through analysis as the distribution center for each predicted value distribution. Since each univariate prediction channel is obtained by integrating multiple base predictor clusters, the prediction result for each bile acid is actually a numerical distribution, not a single value. For example, the predicted value distribution of bile acid C: assuming the outputs of the four predictors in the bile acid C base predictor cluster are [0.52, 0.55, 0.58, 0.55]; the predicted value distribution of bile acid A: assuming the outputs of the four predictors in the bile acid A base predictor cluster are [1.48, 1.50, 1.52, 1.50], through density analysis, the maximum distribution density point for bile acid C is 0.55, i.e., the distribution center is 0.55; the maximum distribution density point for bile acid A is 1.50, i.e., the distribution center is 1.50.

[0044] Next, the residuals between the predicted values ​​corresponding to the N distribution centers and the detected values ​​of the N bile acids in the serum bile acid detection data are calculated to obtain the original deviation data. The original deviation data includes N residual values ​​corresponding to the N bile acids. The difference between the actual measured value of each bile acid and its corresponding distribution center is calculated: for example, if the actual measured value of bile acid C is 0.8 μM and the actual measured value of bile acid A is 1.06 μM, the residual for bile acid C is |0.80 - 0.55| = 0.25, and the residual for bile acid A is |1.60 - 1.50| = 0.10. The original deviation data is: [Bile acid C residual 0.25, bile acid A residual 0.10].

[0045] Furthermore, the original deviation data is weighted and calculated to obtain the variable deviation coefficient.

[0046] The method provided in this application embodiment, which "performs a weighted calculation on the original deviation data to obtain the variable deviation coefficient", includes: Obtain the prediction performance verification results for N univariate prediction channels, and define the prediction confidence level according to the corresponding performance verification results. The prediction confidence level includes an outer confidence level and an inner confidence level, including: Analyze the prediction performance verification results, calculate the probability that the true value falls into the prediction value space of the univariate prediction channel, and output the outer confidence score. Analyze the prediction performance verification results, calculate the probability distribution when the true value falls into the prediction value space of the univariate prediction channel, and define it as the inner confidence level; Using the prediction confidence level as the first weight and the correlation measure of the key bile acid set as the second weight, the N residual values ​​of the original bias data are weighted to obtain the variable bias coefficient.

[0047] First, the prediction performance validation results are analyzed, and the probability that the true value falls into the predicted numerical space of the univariate prediction channel is calculated. The outer confidence score is then output. The outer confidence score measures the overall reliability of the prediction results. For example, for the bile acid C channel: on the validation set, the probability that the true value falls into the predicted numerical space of bile acid C is 92%, so the outer confidence score = 0.92; for the bile acid A channel: on the validation set, the probability that the true value falls into the predicted numerical space of bile acid A is 88%, so the outer confidence score = 0.88.

[0048] Secondly, the prediction performance verification results are analyzed, and the probability distribution when the true value falls into the predicted numerical space of the univariate prediction channel is calculated, defined as the inner confidence level. The inner confidence level measures the degree of centralization of the prediction distribution. For example, for the bile acid C channel: when the true value falls into the prediction space, the probability of it falling near the center of the distribution is very high, and the calculated inner confidence level is 0.95; for the bile acid A channel: the distribution is relatively dispersed, and the inner confidence level is 0.85. Prediction confidence level = outer confidence level × inner confidence level, bile acid C prediction confidence level = 0.92 × 0.95 = 0.874, bile acid A prediction confidence level = 0.88 × 0.85 = 0.748.

[0049] Furthermore, using the predicted confidence level as the first weight and the correlation measure of the key bile acid set as the second weight, the N residual values ​​of the original bias data are weighted to obtain the variable bias coefficient. For example, given that the correlation measure of bile acid C = 0.82 and the correlation measure of bile acid A = 0.75, a dual weighting strategy is adopted: Variable bias coefficient = Σ(predicted confidence level × correlation measure × residual) / Σ(predicted confidence level × correlation measure). Calculation process: Bile acid C contribution = 0.874 × 0.82 × 0.25 = 0.179, bile acid A contribution = 0.748 × 0.75 × 0.10 = 0.056, total weight = (0.874 × 0.82) + (0.748 × 0.75) = 1.278, variable bias coefficient = (0.179 + 0.056) / 1.278 ≈ 0.184.

[0050] Finally, the variable deviation coefficient is compared with the preset deviation threshold for anomaly detection. If the variable deviation coefficient is greater than or equal to the deviation threshold, the serum bile acid test data is marked as abnormal data. Assuming the deviation threshold calculated based on a large amount of normal serum bile acid test data is 0.15, which is the 95th percentile of the deviation coefficient for the normal population, the variable deviation coefficient 0.184 > 0.15, therefore the serum bile acid test data is marked as abnormal data.

[0051] In the method provided in this application embodiment, the deviation threshold is determined by the distribution of deviation coefficients on the normal serum bile acid detection data. The deviation threshold is determined through the following scientific process: on normal serum bile acid detection data from a large number of healthy individuals, the entire S100-S500 analysis process is run to calculate the variable deviation coefficient for each healthy sample. Then, the statistical distribution of these coefficients is studied, and a threshold is determined based on the distribution, for example, 95%. Samples exceeding this threshold are considered statistically significant anomalies.

[0052] Based on the anomaly detection method provided in the above embodiments of this application, if the detection data is determined to be free of anomalies, i.e., the variable deviation coefficient does not exceed the deviation threshold, the data can be further used to perform a predictive assessment of PD cognitive function, thereby achieving continuous quantification and risk stratification of the patient's cognitive state. For example, the concentration data of key bile acids are used as features and input into a pre-trained predictive model to obtain a quantified PD cognitive prediction assessment value. This predictive model can be constructed using feature fusion combined with machine learning or mathematical regression methods, and can quantify the risk level of cognitive state based on the pattern of bile acid profiles.

[0053] In this embodiment, a dual-weighting strategy is employed to give greater weight to more reliable and important variables in the final decision, thereby improving the accuracy of the judgment. Even if a prediction channel occasionally performs poorly, or a bile acid exhibits random fluctuations, stable performance can still be maintained due to the comprehensive evaluation and weight adjustment. The variable bias coefficient can not only be used for binary classification, but its magnitude can also reflect the degree of abnormality, providing researchers with richer decision-making information. The bias threshold based on normal data distribution can adapt to different detection conditions and population characteristics, maintaining good generalization ability. The method for identifying abnormal bile acid data can be used for subsequent PD cognitive function assessment, ensuring the accuracy of the final assessment.

[0054] One or more technical solutions provided in this application have at least the following technical effects or advantages: By screening the set of key bile acids most relevant to abnormalities through multivariate analysis, the unreliability of relying on a single indicator is avoided, and a more comprehensive reflection of the complex pathological state of PD cognitive impairment is achieved. Ensemble learning is used to construct univariate prediction channels for each key bile acid, establishing personalized theoretical normal values ​​for each sample based on its own metabolic background, rather than using a fixed population reference range. This makes abnormality judgment more refined and can detect smaller and earlier deviations. In the final decision-making process, not only are the prediction biases of each bile acid considered, but prediction confidence and correlation measures are also innovatively introduced as dual weights for comprehensive calculation, making the final variable bias coefficient more statistically significant and clinically interpretable, effectively reducing the misjudgment rate. From data preprocessing, feature selection, model building, data reconstruction to final abnormality discrimination, the entire process is automated and standardized, making it easier to apply and standardize in a large-scale clinical setting.

[0055] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0056] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0057] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for detecting abnormal serum bile acid data in PD cognitive function assessment, characterized in that, include: Based on the serum bile acid detection dataset, multivariate analysis was used to identify the top N bile acids that were most correlated with data anomalies, forming a key bile acid set, where N is a positive integer greater than 1. Using normal serum bile acid detection data as training data, N univariate prediction channels are constructed for each of the N bile acids in the key bile acid set, wherein each univariate prediction channel is constructed based on an ensemble learning method. Acquire serum bile acid detection data and reconstruct the data to obtain an anomaly analysis dataset, wherein the anomaly analysis dataset includes N anomaly analysis data clusters; Establish the allocation relationship between the anomaly analysis dataset and the N univariate prediction channels, and input the N anomaly analysis data clusters into the N univariate prediction channels for iterative prediction to obtain N univariate prediction results; The serum bile acid detection data is compared with the N univariate prediction results, and the corresponding variable deviation coefficient is calculated. If the variable deviation coefficient exceeds a preset deviation threshold, the serum bile acid detection data is marked as abnormal data.

2. The method for detecting abnormal serum bile acid data in PD cognitive function assessment as described in claim 1, characterized in that, Based on the serum bile acid detection dataset, multivariate analysis was used to identify the top N bile acids most highly correlated with data anomalies, forming a key bile acid set, where N is a positive integer greater than 1, including: Based on the serum bile acid detection dataset, abnormal serum bile acid detection data are extracted, and the corresponding data abnormality index is evaluated and labeled. Combining the abnormal serum bile acid detection data with the data abnormality index, a correlation analysis is performed on all M types of bile acids to obtain M correlation measures for the M types of bile acids, where M is a positive integer greater than or equal to N. M bile acids are sorted in descending order according to the M correlation measures, and the M correlation measures are cumulatively analyzed. The top N bile acids whose cumulative results satisfy the preset correlation measure threshold are selected and stored as the key bile acid set.

3. The method for detecting abnormal serum bile acid data in PD cognitive function assessment as described in claim 1, characterized in that, Using normal serum bile acid detection data as training data, N univariate prediction channels are constructed for each of the N bile acids in the key bile acid set. Each univariate prediction channel is constructed based on an ensemble learning method and includes: Based on the serum bile acid detection dataset, all M types of bile acids are extracted, and corresponding meta-learning paradigms are defined to construct a meta-learning model for variable prediction. Traverse the N bile acids in the set of key bile acids, and construct N base predictor clusters according to the meta-learning model; Based on the ensemble learning method, N base predictor clusters are integrated to construct corresponding N univariate prediction channels, wherein the prediction output of each univariate prediction channel includes the prediction values ​​of several base predictors in the base predictor clusters.

4. The method for detecting abnormal serum bile acid data in PD cognitive function assessment as described in claim 3, characterized in that, Iterate through the N bile acids in the key bile acid set, and construct N base predictor clusters according to the meta-learning model, including: Based on the key bile acid set and the meta-learning model, N base predictors corresponding to N bile acids are constructed. Several replica predictors are established for each of the N base predictors, and the replica predictors are randomly selected for random mutation. The predictive performance of the random mutation results is verified, and multiple replica predictors whose predictive performance verification results are passed are selected from the random mutation results. These replica predictors are then merged with the corresponding base predictors to output the base predictor clusters, and N base predictor clusters are obtained.

5. The method for detecting abnormal serum bile acid data in PD cognitive function assessment as described in claim 1, characterized in that, Serum bile acid detection data are acquired and reconstructed to obtain an anomaly analysis dataset, wherein the anomaly analysis dataset includes N anomaly analysis data clusters, including: Numerical extraction is performed based on the serum bile acid detection data, using the key bile acid set as an index. Establish N replica sets of numerical extraction results, and establish N anomaly analysis data clusters with N bile acids in the key bile acid set as single variables respectively; In the anomaly analysis data cluster, any one type of bile acid is used as the supervisor, and the corresponding N-1 types of bile acids are used as the sample input.

6. The method for detecting abnormal serum bile acid data in PD cognitive function assessment as described in claim 1, characterized in that, The serum bile acid detection data is compared with the N univariate prediction results, and the corresponding variable deviation coefficient is calculated. If the variable deviation coefficient exceeds a preset deviation threshold, the serum bile acid detection data is marked as abnormal data, including: Based on the N univariate prediction results, establish the predicted value distributions corresponding to the N bile acids, and analyze the points with the maximum distribution density as the distribution centers corresponding to each predicted value distribution. The residuals between the predicted values ​​corresponding to the N distribution centers and the detected values ​​of the N bile acids in the serum bile acid detection data are calculated to obtain the original deviation data, wherein the original deviation data includes N residual values ​​corresponding to the N bile acids. The original deviation data are weighted and calculated to obtain the variable deviation coefficient; The variable deviation coefficient is compared with the preset deviation threshold to determine anomalies. If the variable deviation coefficient is greater than or equal to the deviation threshold, the serum bile acid detection data is marked as abnormal data.

7. The method for detecting abnormal serum bile acid data in PD cognitive function assessment as described in claim 6, characterized in that, The original deviation data are weighted and calculated to obtain the variable deviation coefficients, including: Obtain the prediction performance verification results for N univariate prediction channels, and define the prediction confidence level according to the corresponding performance verification results. The prediction confidence level includes an outer confidence level and an inner confidence level, including: Analyze the prediction performance verification results, calculate the probability that the true value falls into the prediction value space of the univariate prediction channel, and output the outer confidence score. Analyze the prediction performance verification results, calculate the probability distribution when the true value falls into the prediction value space of the univariate prediction channel, and define it as the inner confidence level; Using the prediction confidence level as the first weight and the correlation measure of the key bile acid set as the second weight, the N residual values ​​of the original bias data are weighted to obtain the variable bias coefficient.

8. The method for detecting abnormal serum bile acid data in PD cognitive function assessment as described in claim 6, characterized in that, The deviation threshold is determined by the distribution of deviation coefficients on the normal serum bile acid detection data.

Images