Diabetic nephropathy risk assessment and prediction system based on platelet indexes

Abstract

The invention discloses a diabetic nephropathy risk assessment and prediction system based on platelet indexes, and particularly relates to the technical field of disease risk prediction, which comprises a platelet index acquisition module, a drug use data acquisition module, a drug intervention assessment module, an index correction module, a clustering analysis module and a diabetic nephropathy risk assessment module, the method comprises the following steps: acquiring platelet related detection indexes and historical medication data of a patient, calculating a drug intervention influence index and generating a correction factor, realizing platelet index reverse reduction correction by adopting a convolutional neural network model, determining a similar group in combination with clustering analysis, and dynamically evaluating the risk of diabetic nephropathy based on the average morbidity of the group. Personalized accurate prediction is realized; the platelet index correction accuracy is improved, individualized drug intervention elimination is achieved, the similar population clustering analysis precision is enhanced, diabetic nephropathy risk prediction is dynamically adjusted in combination with comprehensive scores, and the early screening and individualized intervention effects are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of disease risk prediction, and more specifically, to a risk assessment and prediction system for diabetic nephropathy based on platelet indicators. Background Art

[0002] Diabetic nephropathy is one of the most common chronic complications of diabetes and is also the main cause of end-stage renal disease. With the continuous increase in the prevalence of diabetes, the incidence of diabetic nephropathy has also been increasing year by year, seriously threatening the life and health of patients. At present, the clinical screening of diabetic nephropathy mainly relies on urine protein detection, glomerular filtration rate assessment, and imaging examinations. However, existing detection methods usually can only detect lesions after significant kidney damage has occurred in patients, making it difficult to identify the early risks of diabetic nephropathy in a timely manner and delaying the best intervention opportunity.

[0003] In recent years, studies have shown that abnormal platelet function plays an important role in the occurrence and development of diabetic nephropathy. Platelet-related detection indicators, such as platelet count, mean platelet volume, platelet distribution width, and large platelet ratio, can reflect the inflammatory state, vascular endothelial function, and degree of microvascular damage of patients, and have the potential to be used as biomarkers for predicting the early risks of diabetic nephropathy. However, platelet indicators are easily interfered by the drugs taken by patients, and there are significant differences among different individuals. The existing technology lacks a systematic method for effectively correcting drug intervention factors and dynamically analyzing the individual differences of patients.

[0004] Most existing disease risk prediction systems rely on single indicators or traditional statistical models, making it difficult to handle the complex interaction relationships among multiple dimensions and multiple variables, with limited prediction accuracy and difficult to meet the clinical demand for individualized and accurate prediction. Therefore, there is an urgent need for a system based on platelet-related detection indicators, integrating patient medication information and individual characteristics, which can dynamically correct the impact of drug intervention and improve the accuracy of diabetic nephropathy risk prediction. Summary of the Invention

[0005] To achieve the above object, the present invention provides the following technical solutions: A risk assessment and prediction system for diabetic nephropathy based on platelet indicators, comprising: A platelet indicator acquisition module for obtaining a set of platelet-related detection indicators of a patient; A drug use data acquisition module for collecting the historical medication data of a patient; A drug intervention evaluation module for calculating a drug intervention impact index MDI based on the historical medication data of a patient and generating a correction factor RF according to the drug intervention impact index MDI; An index correction module, which is used to reversely restore and correct platelet indexes through a convolutional neural network model combined with a correction factor RF when the Medication Intervention Impact Index (MDI) exceeds a preset standard threshold, and output a set of corrected platelet-related detection indexes; when the MDI does not exceed the preset standard threshold, the average value of multiple measurements is used to determine a set of normal platelet indexes; A clustering analysis module, which is used to input historical medication data and the set of platelet-related detection indexes into a clustering model to determine the similar group to which the patient belongs; A diabetic nephropathy risk assessment module, which is used to evaluate and output the patient's diabetic nephropathy risk based on the average incidence of diabetic nephropathy in the similar group.

[0006] In a preferred embodiment, the platelet-related detection index acquisition step includes: using an automated hematology analyzer for multi-timepoint continuous sampling, and eliminating the detection error caused by circadian rhythm at different sampling time points to ensure the temporal consistency and data stability of platelet indexes.

[0007] In a preferred embodiment, the Medication Intervention Impact Index (MDI) adopts a weight matrix calculation method, dynamically sets weights for different categories of drugs respectively, and uses multi-dimensional feature cross-modeling according to the impact mechanism of drugs on different platelet indexes to improve the individual precision of MDI.

[0008] In a preferred embodiment, the non-linear function of the correction factor RF is modeled by a piecewise function based on the MDI comprehensive value interval. Among them, the low-intervention interval adopts an exponential decay model, the moderate-intervention interval adopts a logarithmic growth model, and the high-intervention interval adopts a saturation exponential model, so as to achieve hierarchical non-linear correction of the drug intervention intensity.

[0009] In a preferred embodiment, the convolutional neural network model (CNN) adopts a multi-channel input structure. Among them, the first input channel is used to input the set of platelet-related detection indexes, the second input channel is used to input the set of drug usage data, and the third input channel is used to input the correction factor RF. Cross-channel convolution operations are used to extract the interaction features of different input sources to achieve more accurate reverse restoration and correction.

[0010] In a preferred embodiment, the clustering analysis step of the similar group is implemented by using the self-organizing mapping neural network algorithm. When this algorithm performs spatial mapping of the patient's platelet indexes, based on the topology-preserving mapping method, it maps high-dimensional input data to a two-dimensional feature space, improving the visualization effect of the clustering result and the accuracy of similarity analysis.

[0011] The calculation of the diabetic nephropathy risk assessment of the clustering similar group adopts the following steps and formulas: Step 1: In the historical database, assume that there are already K clustering groups, and the target patient is classified into the j-th cluster, numbered C_j; Step 2: Count the total number of patients N_j in cluster C_j and the number of patients D_j with diagnosed diabetic nephropathy among them; Step 3: Calculate the average incidence rate R_j of diabetic nephropathy in cluster C_j. The formula is: R_j = D_j / N_j; Step 4: Determine the basic risk score P_base of the target patient according to R_j. The formula is: P_base = γ × R_j, where γ is a preset model adjustment coefficient; Step 5: Combine the individual characteristic differences of the target patient to correct the basic risk score P_base and calculate the final diabetic nephropathy risk score P_final. The formula is: P_final = P_base + δ × (S_patient - S_cluster_avg); δ is a preset risk adjustment coefficient, S_patient is the comprehensive score of the patient's individual platelet indicators, and S_cluster_avg is the average comprehensive score of the platelet indicators of the clustering group.

[0012] In a preferred embodiment, the acquisition logic of the comprehensive score of the patient's individual platelet indicators is as follows: Use Z-Score for data standardization processing, then convert the corrected set of platelet-related detection indicators into a patient vector. There is a reference standard vector, and calculate the Mahalanobis distance between the patient vector and the reference standard vector as the comprehensive score of the patient's individual platelet indicators.

[0013] In a preferred embodiment, the logic of the average comprehensive score of the platelet indicators of the clustering group is as follows: Obtain the Mahalanobis distance between the individual vector corresponding to each individual in the clustering group and the reference standard vector, and then calculate the average value as the average comprehensive score of the platelet indicators of the clustering group.

[0014] The technical effects and advantages of the present invention: By introducing a drug intervention impact index calculation and correction factor generation mechanism, the present invention realizes the dynamic correction of drug interference factors in platelet-related detection indicators. Compared with the prior art that cannot effectively eliminate the influence of drugs on platelet indicators, the present invention combines the patient's individual medication information, metabolic characteristics, and combined medication effects to construct a dynamic weight model and a non-linear correction function, which can accurately restore the true platelet function state of the patient. Through the convolutional neural network model, the reverse reduction and correction process is completed, improving the reliability of platelet indicator data, providing an accurate data basis for subsequent diabetic nephropathy risk assessment, and significantly improving the scientificity and clinical applicability of the prediction results.

[0015] The present invention uses the self-organizing mapping neural network algorithm to perform similarity clustering analysis on patient groups. It can map the platelet index characteristics of patients to a two-dimensional space while maintaining the topological structure of high-dimensional input data, clearly revealing the similarities and distribution laws of different patient groups. Through the application of this algorithm, not only the accuracy and stability of clustering analysis are improved, but also the clustering results have good visualization effects, facilitating clinicians to intuitively judge the risk stratification of patients. By using the average incidence rate evaluation method based on similar groups, the problem of insufficient adaptability of traditional models to individual differences is effectively overcome, realizing the stratified prediction and personalized intervention of the risk of diabetic nephropathy.

[0016] The present invention establishes a risk scoring model for diabetic nephropathy based on the comprehensive score of platelet-related detection indicators. By combining the incidence rate of similar groups and the comprehensive score of individual patient platelet indicators, the basic risk score is dynamically adjusted to achieve the accurate assessment of the final risk of diabetic nephropathy. The Mahalanobis distance is used to calculate the comprehensive score of individual platelet indicators, fully considering the correlation between different indicators, and enhancing the scientificity and robustness of the scoring model. Through the dynamic correction of individual risk scores, the present invention can more accurately reflect the true disease risk level of patients, assist clinicians in formulating early screening, early intervention, and personalized treatment plans, and improve the early detection rate and intervention effect of diabetic nephropathy. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] For the convenience of those skilled in the art to understand, the present invention will be further described below in conjunction with the accompanying drawings; Figure 1 FIG. is a schematic diagram of a risk assessment and prediction system for diabetic nephropathy based on platelet indicators in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0019] Referring to Figure 1 the following embodiments are obtained: Embodiment 1: A risk assessment and prediction system for diabetic nephropathy based on platelet indicators, comprising: Platelet index collection module, used to obtain the set of platelet-related detection indexes of patients; specifically: used to collect the set of platelet-related detection indexes in the blood of patients, including but not limited to key parameters such as platelet count, mean platelet volume, platelet distribution width, and large platelet ratio. Through an automated blood analysis device, continuous sampling is carried out at multiple time points, and the influence of circadian rhythm on the test results is eliminated, so as to ensure the temporal consistency of data and the stability of indexes. Platelet indexes, as early biological signals of diabetic patients developing kidney complications, reflect the coagulation function, inflammatory state, and microvascular health status of the patient's blood system. This module ensures that the data source is objective and accurate, which is the basis for subsequent analysis and risk assessment.

[0020] Drug use data collection module, used to collect the historical drug use data of patients; specifically: used to systematically collect the historical drug use information of patients, including the types of drugs used by patients, drug doses, usage duration, and whether there is combination drug use, etc. The data source can include the electronic medical record system, the drug management system, or the manual entry system. Drugs have direct or indirect effects on platelet indexes, especially antiplatelet drugs, anticoagulant drugs, and certain therapeutic drugs that affect platelet generation and function. Collecting the complete drug use history of patients is a prerequisite for evaluating the degree of interference of drugs on platelet indexes.

[0021] Drug intervention evaluation module, used to calculate the drug intervention impact index MDI based on the historical drug use data of patients, and generate a correction factor RF according to the drug intervention impact index MDI; specifically: according to the drug use data of patients, considering factors such as drug type, dosage, medication cycle, and individual metabolic characteristics, calculate the drug intervention impact index. At the same time, according to the drug intervention impact index, a correction factor is generated through a non-linear function. The correction factor is used to reflect the possible interference intensity of drugs on platelet detection indexes. The effects of different drugs on platelet indexes are complex and individually different. The calculation of this module can dynamically quantify the drug intervention effect, provide a data basis for subsequent index correction and model accuracy, and avoid misjudgment of indexes caused by drug intervention.

[0022] An index correction module, which is used to inversely restore and correct platelet indices through a convolutional neural network model combined with a correction factor RF when the Medication Intervention Impact Index (MDI) exceeds a preset standard threshold, and output a set of corrected platelet-related detection indices; when the MDI does not exceed the preset standard threshold, the average value of multiple measurements is used to determine a set of normal platelet indices; specifically: when the MDI exceeds the threshold set by the system, the correction process is initiated, and the collected platelet index data is inversely restored and corrected through a convolutional neural network model combined with a correction factor to obtain the true platelet indices after removing the medication intervention factors. When the MDI does not exceed the threshold, the average value of multi-time point detection data is used to ensure the reliability and stability of platelet indices. Ensure the true effectiveness of platelet detection data, eliminate the bias caused by medication interference, and provide accurate data input for subsequent similar group division and risk prediction. Effectively avoid the problem of inaccurate risk assessment caused by abnormal detection indices.

[0023] A clustering analysis module, which is used to input historical medication data and a set of platelet-related detection indices into a clustering model to determine the similar group to which the patient belongs; specifically: input the patient's platelet detection indices and historical medication information into the clustering model, and use an unsupervised learning algorithm to automatically analyze and divide the patient's similar groups.

[0024] The clustering algorithm maps high-dimensional data space based on data similarity, classifies patients according to platelet index characteristics and medication influence status, and forms different risk groups. Different patients have significant differences due to different disease states, platelet functions, and medication influences. Through similar group clustering analysis, patient groups with the same characteristics can be mined, providing data support and a scientific basis for disease risk prediction and individualized treatment strategies.

[0025] A diabetic nephropathy risk assessment module, which is used to evaluate and output the diabetic nephropathy risk of a patient based on the average incidence of diabetic nephropathy in the similar group; specifically: based on the average incidence of diabetic nephropathy in the similar group to which the patient belongs, combined with the comprehensive score of individual platelet indices and the group average score, evaluate the risk of diabetic nephropathy occurrence in the target patient. The system outputs the patient's risk level and prediction value according to the statistical analysis results, providing intervention suggestions and treatment references for clinicians. Through a risk prediction method based on a data model, the probability of diabetic nephropathy occurrence can be predicted in advance before the patient shows obvious clinical symptoms. This module can improve the scientificity and accuracy of clinical risk assessment, assist doctors in achieving early screening, early intervention, and individualized treatment, and reduce the incidence and progression rate of diabetic nephropathy.

[0026] The steps for collecting platelet-related detection indicators include: using an automated hematology analyzer for multi-timepoint continuous sampling, and eliminating the detection errors caused by circadian rhythm at different sampling time points to ensure the temporal consistency and data stability of platelet indicators.

[0027] Specific implementation methods for the steps of collecting platelet-related detection indicators: I. Sampling preparation: Before collecting blood from the patient, inform them to maintain normal work and rest, avoid strenuous exercise and mental stress to ensure the stability of the blood state. Ensure that the sampling environment has a constant temperature and humidity to avoid the influence of the external environment on the quality of blood samples.

[0028] II. Sampling time node design: According to the influence law of circadian rhythm on platelet physiological changes, set multiple sampling time points. Usually divided into early morning fasting, morning, afternoon, and night periods. Each sampling time point is separated by a certain time to avoid overly concentrated sampling times.

[0029] III. Continuous sampling operation: At the specified time points, collect venous blood samples through an automated hematology analyzer. The blood collection tubes are uniformly treated with anticoagulants to avoid platelet aggregation. After each collection is completed, immediately send it to the analysis equipment for platelet indicator detection to avoid data errors caused by the sample being placed for too long.

[0030] IV. Data collection and recording: After the detection is completed at each time point, record the key indicator values such as platelet count, mean platelet volume, platelet distribution width, and large platelet ratio. All sampling data is automatically uploaded to the information system for time marking and patient identity identification.

[0031] V. Elimination of circadian rhythm errors: Using the reference model of circadian rhythm changes established from historical large-scale clinical sample data, compare and analyze the platelet indicator values collected from the patient at different time points with the reference model to identify and eliminate the fluctuations caused by physiological rhythm changes. Use interpolation algorithms to standardize the platelet indicator values to a unified time reference system to ensure the comparability and consistency of data in different periods.

[0032] VI. Temporal data consistency test: Perform outlier detection on the platelet indicator data collected at all time points. If there are data points with deviations exceeding the preset range, the system will automatically prompt for recheck or elimination of outliers. Form a complete platelet indicator sequence for the continuous sampling data that passes the test in chronological order and calculate the data fluctuation range to ensure the stability of the indicators.

[0033] VII. Confirmation and output of indicator results: Based on the platelet indicator data after error elimination and stability testing, perform weighted averaging or optimal value screening to form the final platelet indicator set. This indicator set serves as the basic data for subsequent drug intervention evaluation, indicator correction, and diabetes nephropathy risk assessment.

[0034] The reference model of circadian rhythm changes is a model structure in the prior art built into the platelet data processing system or device. The specific construction method is described as follows: I. Data collection: Collect platelet test data of a large number of normal people of different ages, genders, and ethnicities. Ensure that each subject completes multiple platelet index tests at different time points within 24 hours. The collected data includes indicators such as platelet count, mean platelet volume, platelet distribution width, and ratio of large platelets.

[0035] II. Data preprocessing: Clean the collected original platelet data, including removing missing values, removing outliers, and standardization processing. Data standardization is used to eliminate the interference caused by individual differences, ensuring that the model learns the law of time change rather than the basic differences of different populations.

[0036] III. Stratification by time dimension: Stratify the platelet indicators according to the test time, divided into early morning period, morning period, afternoon period, evening period, and night period. Further refine it to the hour level to ensure that the rhythm model has time accuracy. In each time layer, summarize and statistically calculate the mean value, standard deviation, and fluctuation range of platelet indicators as the basic statistical features.

[0037] IV. Model training: Based on the platelet indicator data stratified by time, use a regression model based on a tree structure or a neural network model for training. The model input is the collection time point, and the model output is the predicted value of platelet indicators for the corresponding time period. Through a large amount of training data, learn the change trend of platelet indicators at different time points to form a circadian rhythm model.

[0038] V. Model optimization: Use the cross-validation method to evaluate the accuracy of the model and adjust the model parameters to ensure the generalization ability of the model. The model optimization process includes key steps such as adjusting the time window, balancing the sample distribution, and the division ratio of the data set.

[0039] VI. Model verification: Apply the trained model to an independent validation set, compare the difference between the model predicted value and the true platelet test value. Calculate the model prediction error and evaluate its adaptability to different populations and different test time points. If the error is within the acceptable range, it indicates that the model has practical value.

[0040] VII. Model application: Integrate the circadian rhythm change model into the platelet data processing system to correct the real-time collected platelet indicators. Input the collection time point into the model to obtain the correction coefficient or standard value for the corresponding time, and then compare the actual test value with the model predicted value to correct the natural fluctuation caused by the physiological rhythm. Output the platelet indicator data after rhythm correction to ensure the consistency and reliability of the data.

[0041] The Medication Intervention Impact Index (MDI) uses a weighted matrix calculation method. Dynamic weights are set for different categories of medications respectively. Based on the impact mechanisms of medications on different platelet indicators, multi-dimensional feature cross-modeling is adopted to improve the individual precision of MDI.

[0042] The Medication Intervention Impact Index (MDI) is an important value used to quantify the comprehensive impact degree of medications on platelet detection indicators. Different medication categories have different intervention effects on different platelet indicators. Moreover, individual medication metabolism differences and combined medication effects in patients also affect the final platelet indicators. Therefore, precise calculation of MDI needs to be achieved through dynamic weight setting and multi-dimensional feature cross-modeling.

[0043] Source factors of dynamic weights: The intervention effects of each medication vary due to the following factors: medication category (antiplatelet drugs, anticoagulants, hematopoietic drugs, etc.), drug dosage, duration of medication, combined medication, individual metabolism differences (enzyme activity, genetic factors), and the strength of the impact of drug action targets on different platelet indicators.

[0044] Weight setting principle: Different medications have different degrees of impact on different platelet indicators. The impacts of medications on platelet count, mean volume, distribution width, and large cell ratio need to be calculated separately. Weights are established through the analysis of a large amount of clinical data, pharmacodynamic research results, and pharmacokinetic models. The weights are dynamically adjusted and correlated with patient individual characteristics and the strength of drug interactions.

[0045] Calculation steps of the Medication Intervention Impact Index (MDI): Suppose a patient uses multiple medications, and the medication numbers range from serial number one to the total number of serial numbers. Parameters are assigned to different medications and platelet indicators respectively: the drug dosage is the dosage value, the medication time is the time value, the comprehensive impact factor of the medication on different platelet indicators is the preset indicator impact coefficient, the individual metabolism correction factor is the metabolism correction value, and the combined medication correction factor is the combined use correction value; Each medication affects multiple indicators, and there are also interactive effects among the indicators. Cross-modeling is expressed through the cross-product of multi-dimensional matrices: the intervention score of a single medication = dosage value × time value × indicator impact coefficient × metabolism correction value × combined use correction value; For different categories of drugs, a multi-dimensional weight matrix is set. For example: weight matrix = |weight 1, weight 2, weight 3, weight 4|, corresponding to index 1, index 2, index 3, and index 4 respectively. The matrix is automatically adjusted according to drug categories and patient characteristics. For example, antiplatelet drugs mainly affect the large cell ratio and mean volume, so the weight values on these indicators are high. The number of weights corresponds to the number of indicators. For example, n weight numbers correspond to n indicator numbers. The logic for obtaining weights is as follows: The influence degrees of drugs on different platelet indicators at different time periods are expressed in a matrix form by experts based on pharmacokinetic parameters and clinical pharmacological research results, and a multi-dimensional dynamic weight matrix is constructed to reflect the dynamic influence weights of different drugs on platelet indicators under different conditions.

[0046] The intervention score of each drug index = the single drug intervention score × the corresponding index weight. The total intervention influence of multiple drugs is combined. The drug intervention influence index MDI = the sum of all drug intervention scores. The more indicators a drug affects, the more calculations are required. It can also be calculated as the product of the single drug intervention score and the sum of the weights of all affected indicators to obtain the total intervention score of a single drug. In this case, the drug intervention influence index MDI = the sum of the total intervention scores of all single drugs.

[0047] All data are standardized before entering the calculation. Input factors such as drug dosage, time, and metabolic effects are all normalized. The weight values are based on a standard scale or dimensionless indicators to ensure that all parameters have the same dimension when entering the model and avoid distortion caused by inconsistent units. For example: all drug dosages are unified as daily dosages and normalized to zero to one, time is standardized in days and normalized, and the weight values are dimensionless standard percentages.

[0048] Based on the calculation of individualized data, it is ensured that the effects of drugs on platelet indicators can be accurately evaluated for different patients, taking into account the actual situation of the combined use of multiple drugs, not limited to a certain disease or a certain drug, adapting to diverse clinical scenarios, supporting the dynamic optimization and adaptive learning of subsequent models. Standardization and cross-modeling avoid the influence of a single indicator on model decision-making and ensure the reasonable quantification and evaluation of different input dimensions.

[0049] The individual metabolic correction factor (metabolic correction value) reflects the patient's metabolic ability to drugs and affects the in-vivo exposure concentration and action intensity of drugs. Due to differences in genes, age, liver and kidney functions, etc. among different individuals, the in-vivo metabolism rate of drugs is different. For the same dose of drug, the drug concentration in the body of a person with slow metabolism is high and the action is stronger; the opposite is true for a person with fast metabolism. Therefore, this factor is used to dynamically adjust the influence weight of drugs on platelets. Example of the acquisition method: normal function → metabolic correction value is one, mild injury → correction value is one plus twenty percent, severe injury → correction value is one plus fifty percent.

[0050] Combined Medication Correction Factor (Combined Use Correction Value): When using combined medications, there may be synergistic or antagonistic effects between different drugs, which can affect the overall drug effect. Synergistic enhancement → enhanced platelet effect; antagonistic attenuation → weakened platelet effect; therefore, the combined medication correction factor is used to dynamically adjust the actual effects between various drugs in the MDI. Acquisition method: Pharmacopoeia rule method (static assignment), based on the drug interaction information clearly marked in the pharmacopoeia and guidelines, classified according to intensity as: no interaction, mild enhancement, moderate enhancement, strong synergy, etc., and different weights are assigned. For example: no → correction value is one, mild synergy → correction value is one plus twenty percent, moderate synergy → correction value is one plus fifty percent, strong synergy → correction value is one plus double.

[0051] The non - linear function of the correction factor RF is modeled using a piece - wise function based on the MDI comprehensive value interval. Among them, the low - intervention interval uses an exponential decay model, the moderate - intervention interval uses a logarithmic growth model, and the high - intervention interval uses a saturation exponential model, so as to achieve a hierarchical non - linear correction of the drug intervention intensity.

[0052] Piece - wise modeling method of the non - linear function of the correction factor RF: Calculate the drug intervention impact index of the target object, denoted as the intervention intensity value, which is completed according to the established model. According to the value of the intervention intensity, it is divided into a low - intervention interval, a moderate - intervention interval, and a high - intervention interval.

[0053] The low - intervention interval is where the intervention intensity is less than or equal to the first threshold, the moderate - intervention interval is where the intervention intensity is greater than the first threshold and less than or equal to the second threshold, and the high - intervention interval is where the intervention intensity is greater than the second threshold. The specific numerical values of the thresholds are determined based on clinical data. Different non - linear functions are selected according to the interval to calculate the correction factor RF.

[0054] Piece - wise function formula: (1) In the low - intervention interval, when the intervention intensity is less than or equal to the first threshold, the exponential decay model is used to calculate the correction factor. Correction factor = coefficient one × exponential function (negative parameter one × intervention intensity); the correction factor decreases rapidly with the increase of the intervention intensity, reflecting that the drug intervention effect is small and the correction amplitude is limited. Coefficient one and parameter one are constant values determined by the model.

[0055] (2) In the moderate - intervention interval, when the intervention intensity is greater than the first threshold and less than or equal to the second threshold, the logarithmic growth model is used to calculate the correction factor; correction factor = coefficient two × logarithmic function (intervention intensity - constant two); the correction factor gradually increases with the increase of the intervention intensity, indicating that the drug intervention has a significant impact on platelet indicators. Coefficient two and constant two are parameters set after model training to ensure the continuity of the function.

[0056] (3) High intervention range: When the intervention intensity is greater than the second threshold, the saturation index model is used to calculate the correction factor; Correction factor = Coefficient 3 × [1 - exponential function (negative Parameter 2 × intervention intensity)]. The correction factor approaches the maximum value in the high intervention range, forming a gentle curve, reflecting that the drug intervention effect tends to be saturated. Coefficient 3 and Parameter 2 are set values for clinical model training to avoid excessive correction.

[0057] Classification and processing are carried out according to the intervention intensity range, and different correction strategies are adopted in different intensity ranges to ensure the dynamic adaptability and flexibility of the correction factor. After the functions of each range are trained by the model, the parameters are adjusted to ensure the continuity of the functions at the range boundaries and prevent jumps and abnormal fluctuations. The exponential decay model is used in the low intervention range to reflect that the mild drug intervention has limited influence on platelet indicators. The logarithmic growth model is used in the moderate intervention range to reflect the enhanced drug effect, but the growth rate slows down with the change of intensity. The saturation index model is used in the high intervention range to simulate the saturation of the drug effect and avoid the infinite growth of the correction factor. By reasonably distributing the intervention influence weight through piecewise functions, the personalization and scientific nature of the correction factor are improved, and the accuracy and reliability of the system in predicting the risk of diabetic nephropathy are enhanced.

[0058] The convolutional neural network model CNN adopts a multi-channel input structure. The first input channel is used to input the set of platelet-related detection indicators, the second input channel is used to input the set of drug use data, and the third input channel is used to input the correction factor RF. Cross-channel convolution operations are used to extract the interaction features of different input sources to achieve more accurate reverse reduction and correction.

[0059] Input data design: The model has three independent input channels, specifically: The first input channel inputs the set of platelet-related detection indicators, including key indicators such as platelet count, mean volume, distribution width, and large cell ratio, reflecting the platelet function status of the patient. The second input channel inputs the set of drug use data, including the types, dosages, usage times, and combined medication information of the drugs used by the patient in the past and currently, reflecting the possible influence of the drugs on platelet indicators. The third input channel inputs the correction factor data, which is calculated based on the drug intervention influence index and reflects the strength of the comprehensive drug intervention effect.

[0060] Convolution calculation and cross-channel interaction mechanism: After the multi-channel input is completed, the data of each channel enters the corresponding convolutional operation layer respectively for one-dimensional or two-dimensional convolution operations. The convolution operation extracts features from different local regions of the input data through a sliding window, extracting local statistical features and temporal correlation features. The feature maps output by the convolutional layers of each channel are fused through cross-channel interaction operations. The cross-channel interaction operations include the following two core processes: Inter-channel Feature Alignment: Align the feature maps output from different channels along the time dimension or metric dimension, so that the features of each channel have the same scale and structure, facilitating subsequent fusion.

[0061] Inter-channel Convolutional Fusion: Use cross-channel convolutional kernels to perform convolutional calculations between the features extracted from different input channels, extract the correlation patterns between the features of each channel, and construct high-dimensional interaction features among the drug intervention information, platelet metric status, and correction factor.

[0062] Feature Fusion and Fully Connected Processing: After cross-channel convolutional fusion is completed, send the fused features to the feature fusion layer for further compression and high-order feature extraction. After several convolutional layer and pooling layer operations, finally output to the fully connected layer. The fully connected layer performs non-linear mapping and comprehensive operations on the fused feature information to form a high-dimensional feature vector that finally represents the platelet state.

[0063] Output Layer Design: After the fully connected layer outputs a high-dimensional feature vector, enter the output layer. The output layer decodes the feature vector and finally outputs a set of corrected platelet-related detection metrics. This set is the true platelet metric values after removing the influence of drug intervention, specifically including corrected platelet count, mean volume, distribution width, large cell ratio, and other metrics.

[0064] Reverse Reduction and Correction Mechanism: During the training process of this convolutional neural network model, use the platelet metrics before correction, drug usage data, and correction factor as inputs, and use the true platelet metrics not affected by intervention as the supervision signal to construct a reverse reduction and correction model. Through repeated iteration and optimization of the training set, the model automatically learns the true state change rules of platelet metrics under different drug intervention conditions. During actual prediction, the model dynamically predicts and restores the true values of the patient's platelet metrics in the state not affected by drugs based on the input current platelet metrics, drug usage, and correction factor. The output results can be directly applied to subsequent clustering analysis of similar groups and risk assessment of diabetic nephropathy, improving the accuracy and reliability of data processing.

[0065] Through multi-channel input design, effectively fuse multi-source heterogeneous data. Through cross-channel convolutional operations, extract high-order interaction features among drug effects, platelet states, and correction factors, enhancing the model's representation ability for complex intervention factors. Through the reverse reduction and correction mechanism, restore the true platelet function state of patients, eliminate the influence of drugs on test results, and improve the accuracy and scientific nature of subsequent disease risk prediction. This model has strong adaptability and can adapt to differences in drug usage and platelet function states of different individuals, achieving precise correction and personalized prediction.

[0066] The clustering analysis steps for similar groups are implemented using the self-organizing map neural network algorithm. When this algorithm performs spatial mapping of patient platelet indicators, based on the topology-preserving mapping method, it maps high-dimensional input data to a two-dimensional feature space, improving the visualization effect of clustering results and the accuracy of similarity analysis.

[0067] This step conducts clustering analysis of similar groups based on the self-organizing map neural network algorithm. By simulating the self-organizing learning mechanism of neurons, it maps the input patient platelet indicator data from a high-dimensional space to a low-dimensional space to achieve clustering recognition of different patient groups. This algorithm has the property of topology preservation and can maintain the original relative distance and similarity relationships between data during the spatial transformation process, thereby improving the scientificity and accuracy of clustering analysis.

[0068] The input data is a set of corrected or processed patient platelet-related detection indicators, specifically including parameters such as platelet count, mean platelet volume, platelet distribution width, and large platelet ratio. In addition, it can also include patient historical medication and individual metabolism-related data. These data form a high-dimensional input vector, reflecting the patient's comprehensive blood status and medication situation.

[0069] Network structure design: This self-organizing map neural network consists of an input layer, a competitive layer, and an output mapping layer. The input layer is responsible for receiving the patient's high-dimensional indicator vector; the neuron nodes in the competitive layer, driven by the input data, autonomously compete to obtain the node position with the highest response intensity; the output mapping layer performs spatial mapping on the competition results to form an arrangement of nodes in the two-dimensional feature space, representing the aggregation state of different patient data.

[0070] Training and learning mechanism: The self-organizing map neural network adopts an unsupervised learning method and realizes model training and spatial mapping through the following steps: Initialize weight parameters; the connection weight vectors of all neuron nodes are randomly assigned and distributed within the input space as the initial learning state.

[0071] Process input samples one by one: Input each group of patient platelet indicator vectors into the network, perform similarity matching with all nodes in the competitive layer, and calculate the distance value between the node and the input sample.

[0072] Determine the winning node: Through distance calculation, select the node closest to the input sample as the winning node for the current input.

[0073] Neighborhood update mechanism: Centered on the winning node, other nodes within its neighborhood range also adjust their weights according to the similarity strength. The neighborhood range gradually shrinks with the progress of training to achieve local refinement learning.

[0074] Repeated Iteration: Repeat the training of all samples for several rounds until the weights of all nodes are stable, forming an ordered spatial distribution structure.

[0075] Principle of Topology-Preserving Mapping; Through the neighborhood update and global competition mechanism among neurons, the spatial structure is adaptively optimized. During the mapping process, the relative positions of similar patient samples in the input space are kept close in the output space, that is, the spatial topological relationship is consistent. This mechanism ensures that the similarity in the high-dimensional index space is maintained in the low-dimensional mapping space, thus avoiding information distortion or separation of similar samples.

[0076] Mapping and Visualization: After the self-organizing mapping neural network training is completed, the patient platelet index data is mapped to a two-dimensional feature space, and each patient corresponds to a feature point in the output space. Feature points that are adjacent or densely distributed in the space represent patient groups with similar platelet index characteristics. This spatial result can be used for visual display, showing the distribution patterns and quantity differences of different similar groups through a two-dimensional image.

[0077] Similar Group Division: According to the distribution of each feature point in the output space, density analysis or clustering algorithms are used to divide similar groups. Each group represents a subset of patients with similar platelet status and drug influence characteristics, and is subsequently used for diabetes nephropathy risk assessment and clinical intervention guidance.

[0078] By mapping high-dimensional data to a low-dimensional space, the similarity analysis process is simplified, and the system operation efficiency is improved. The topology-preserving mapping mechanism ensures that the similarity characteristics between different patients remain consistent after spatial transformation, improving the accuracy and credibility of clustering analysis. The mapping results have good visualization effects, facilitating doctors and researchers to understand the differences and correlations between patient groups, and assisting in clinical classification and personalized treatment decisions. This step realizes the automatic classification of patient groups based on platelet indices, reveals the abnormal platelet characteristic patterns of different patients, provides a group reference basis for diabetes nephropathy risk prediction, and significantly improves the disease early screening efficiency and clinical intervention effect.

[0079] The risk assessment calculation of diabetic nephropathy for clustered similar groups adopts the following steps and formulas: Step 1: In the historical database, assume that there are already K clustered groups, and the target patient is classified into the j-th cluster, numbered C_j; Step 2: Count the total number of patients N_j within the cluster C_j, and the number of patients D_j who have been diagnosed with diabetic nephropathy among them; Step 3: Calculate the average incidence rate R_j of diabetic nephropathy in the cluster C_j, and the formula is: R_j = D_j / N_j; Step 4. Determine the baseline risk score \(P_{base}\) of the target patient according to \(R_j\). The formula is: \(P_{base}=\gamma\times R_j\), where \(\gamma\) is a preset model adjustment coefficient; Step 5. Combine the individual characteristic differences of the target patient to correct the baseline risk score \(P_{base}\) and calculate the final diabetic nephropathy risk score \(P_{final}\). The formula is: \(P_{final}=P_{base}+\delta\times(S_{patient}-S_{cluster\_avg})\); \(\delta\) is a preset risk adjustment coefficient, \(S_{patient}\) is the comprehensive score of the patient's individual platelet index, and \(S_{cluster\_avg}\) is the average comprehensive score of the platelet index of the clustering group.

[0080] The acquisition logic of the comprehensive score of the patient's individual platelet index is as follows: Use Z - Score for data standardization processing, then convert the corrected set of platelet - related detection indexes into a patient vector. Set a reference standard vector and calculate the Mahalanobis distance between the patient vector and the reference standard vector as the comprehensive score of the patient's individual platelet index.

[0081] The logic of the average comprehensive score of the platelet index of the clustering group is as follows: Obtain the Mahalanobis distance between the individual vector corresponding to each individual in the clustering group and the reference standard vector, and then calculate the average value as the average comprehensive score of the platelet index of the clustering group.

[0082] Through the calculation method of diabetic nephropathy risk assessment based on similar groups, the patient is divided into a clustering group with similar platelet index characteristics and medication information. Use the average incidence of diabetic nephropathy in this group as the baseline risk score, and further combine the difference between the comprehensive score of the patient's individual platelet index and the group average score to dynamically correct the baseline risk value, so as to achieve accurate prediction of individualized diabetic nephropathy risk. This method not only integrates group statistical laws and individual difference analysis, improves the scientific nature of risk assessment and the credibility of prediction results, but also effectively solves the defect of ignoring individual characteristic differences in traditional single - data prediction models.

[0083] By using the Mahalanobis distance to calculate the comprehensive score of the patient's individual platelet index, it is ensured that the correlation between different indexes is considered in the multi - dimensional platelet index data analysis, improving the accuracy and rationality of the comprehensive score. Based on the average comprehensive score of similar groups as the reference standard, it can dynamically reflect the platelet state differences between different patients, ensure the adaptability and stability of the risk score model within different clustering groups, and finally realize the hierarchical management and personalized intervention of diabetic nephropathy risk.

[0084] The above formulas are all dimensionless and only take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula that is closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0085] It should be understood that in various embodiments of the present application, the magnitudes of the sequence numbers of the above processes do not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0086] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in this article can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0087] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0088] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, and all should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A diabetes nephropathy risk assessment and prediction system based on platelet indicators, characterized in that, Including: A platelet index acquisition module, which is used to obtain the set of platelet-related detection indexes of a patient; A drug usage data acquisition module, which is used to collect the historical drug usage data of a patient; A drug intervention evaluation module, which is used to calculate the drug intervention impact index MDI based on the historical drug usage data of a patient, and generate a correction factor RF according to the drug intervention impact index MDI; An index correction module, which is used to reversely restore and correct the platelet indexes through a convolutional neural network model combined with the correction factor RF when the drug intervention impact index MDI exceeds the preset standard threshold, and output the corrected set of platelet-related detection indexes; when the drug intervention impact index MDI does not exceed the preset standard threshold, the average value of multiple measurements is used to determine the set of normal platelet indexes; A clustering analysis module, which is used to input the historical drug usage data and the set of platelet-related detection indexes into a clustering model to determine the similar group to which the patient belongs; A diabetic nephropathy risk assessment module, which is used to evaluate and output the diabetic nephropathy risk of a patient based on the average incidence of diabetic nephropathy in the similar group.

2. The diabetes nephropathy risk assessment and prediction system based on platelet indexes according to claim 1, characterized in that The steps for collecting platelet-related detection indexes include: using an automated blood analyzer for multi-timepoint continuous sampling, and eliminating the detection errors caused by circadian rhythm at different sampling time points to ensure the temporal consistency and data stability of platelet indexes.

3. The risk assessment and prediction system for diabetic nephropathy based on platelet indexes according to claim 2, wherein The drug intervention impact index MDI adopts a weighted matrix calculation method, sets dynamic weights for different categories of drugs respectively, and uses multi-dimensional feature cross-modeling according to the influence mechanism of drugs on different platelet indexes to improve the individual accuracy of MDI.

4. The risk assessment and prediction system for diabetic nephropathy based on platelet indices according to claim 3, characterized in that, The non-linear function of the correction factor RF is modeled by a piecewise function based on the MDI comprehensive value interval, where the low intervention interval adopts an exponential decay model, the moderate intervention interval adopts a logarithmic growth model, and the high intervention interval adopts a saturation exponential model, so as to realize the hierarchical non-linear correction of the drug intervention intensity.

5. The risk assessment and prediction system for diabetic nephropathy based on platelet indexes according to claim 4, wherein The convolutional neural network model CNN adopts a multi-channel input structure, where the first input channel is used to input the set of platelet-related detection indexes, the second input channel is used to input the set of drug usage data, and the third input channel is used to input the correction factor RF. Cross-channel convolution operations are used to extract the interaction features of different input sources to achieve more accurate reverse restoration and correction.

6. The risk assessment and prediction system for diabetic nephropathy based on platelet indicators according to claim 5, wherein The clustering analysis steps of the similar group are implemented by using the self-organizing mapping neural network algorithm. When this algorithm maps the patient platelet index space, based on the topology-preserving mapping method, the high-dimensional input data is mapped to a two-dimensional feature space to improve the visualization effect of the clustering result and the accuracy of similarity analysis.

7. The risk assessment and prediction system for diabetic nephropathy based on platelet indexes according to claim 6, characterized in that, The calculation of the diabetic nephropathy risk assessment of the clustering similar group adopts the following steps and formula: Step 1: In the historical database, assume that there are already K clustering groups, and the target patient is classified into the j-th clustering, numbered C_j; Step 2: Count the total number of patients N_j in the clustering C_j and the number of patients D_j with diagnosed diabetic nephropathy among them; Step 3: Calculate the average incidence of diabetic nephropathy R_j of the clustering C_j, and the formula is: R_j = D_j / N_j; Step 4. Determine the basic risk score P_base of the target patient according to R_j. The formula is: P_base = γ × R_j, where γ is a preset model adjustment coefficient; Step 5. Combine the individual characteristic differences of the target patient, correct the basic risk score P_base, and calculate the final diabetic nephropathy risk score P_final. The formula is: P_final = P_base + δ × (S_patient - S_cluster_avg); δ is a preset risk adjustment coefficient, S_patient is the comprehensive score of the patient's individual platelet index, and S_cluster_avg is the comprehensive score of the average platelet index of the clustering group.

8. The risk assessment and prediction system for diabetic nephropathy based on platelet indexes according to claim 7, wherein The acquisition logic of the comprehensive score of the patient's individual platelet index is as follows: Use Z-Score for data standardization processing, then convert the corrected set of platelet-related detection indicators into a patient vector, set a reference standard vector, and calculate the Mahalanobis distance between the patient vector and the reference standard vector as the comprehensive score of the patient's individual platelet index.

9. The diabetes nephropathy risk assessment and prediction system based on platelet indexes according to claim 8, wherein, The logic of the comprehensive score of the average platelet index of the clustering group is as follows: Obtain the Mahalanobis distance between the individual vector corresponding to each individual in the clustering group and the reference standard vector, and then calculate the average value as the comprehensive score of the average platelet index of the clustering group.

Images