Cerebral stroke patient risk prediction method and system based on large model technology

Abstract

The invention relates to the technical field of medical care, in particular to a cerebral apoplexy patient risk prediction method and system based on a large model technology, and the method comprises the following steps: S1, data collection and preprocessing: collecting health data of a cerebral apoplexy patient; s2, microbiome-driven health risk assessment: collecting microbiome data of the stroke patient, constructing a microbiome-driven health risk assessment model, assessing the relationship between the microbiome change and the stroke occurrence risk, and generating a microbiome health score of the patient; s3, multi-modal data fusion and risk prediction: performing risk prediction on the stroke patient; s4, personalized intervention suggestions are generated, wherein personalized health intervention suggestions are generated according to the risk prediction result of the stroke patient. According to the method, rapid and efficient cerebral apoplexy risk prediction is realized, and the applicability and practicability of the prediction method are enhanced.

Description

Technical Field

[0001] The present invention relates to the field of medical care technology, and in particular to a method and system for predicting the risk of stroke patients based on large model technology. Background Art

[0002] Stroke is one of the leading causes of death and disability worldwide, with high incidence and recurrence rates, posing a huge challenge to patients' life, health and medical resources. The occurrence of stroke is usually related to a variety of complex factors, including hypertension, diabetes, dyslipidemia, lifestyle, etc. It is also closely related to the imbalance of microbiota such as intestinal flora. In recent years, with the development of precision medicine, stroke risk prediction based on multi-source data analysis and large model technology has gradually become a research hotspot, aiming to reduce the incidence and mortality of stroke by identifying high-risk individuals at an early stage and providing personalized intervention recommendations.

[0003] Most current stroke risk prediction methods rely on single-modality data or simple statistical models, and it is difficult to effectively integrate multi-source data, resulting in insufficient prediction accuracy and adaptability. In addition, existing technologies have limitations in data preprocessing, feature extraction, and multimodal fusion, such as insufficient processing of noise data, failure to effectively fill in missing data, and insufficient utilization of associations between modal features. At the same time, the nonlinear associations and complex patterns of microbiome characteristics are difficult to model in traditional statistical methods, limiting the potential of the microbiome in stroke risk assessment. In addition, the lack of personalized health intervention programs has also led to limited health management effects for patients.

[0004] In response to the above problems, the present invention proposes a risk prediction method and system for stroke patients based on large model technology, which not only improves the accuracy and scientificity of the prediction, but also significantly enhances the pertinence and practicality of health management, providing a new technical approach and scientific basis for stroke risk assessment and personalized intervention. Summary of the invention

[0005] The present invention provides a method and system for predicting the risk of stroke patients based on large model technology.

[0006] A method for predicting the risk of stroke patients based on a large model technology comprises the following steps:

[0007] S1, data collection and preprocessing: collect health data of stroke patients from multiple sources, including clinical data, imaging data, and lifestyle data, and preprocess the collected health data, including denoising, filling missing values, and standardization;

[0008] S2, Microbiome-driven health risk assessment: Collect microbiome data of stroke patients, build a microbiome-driven health risk assessment model, evaluate the relationship between microbiome changes and stroke risk, and generate patients' microbiome health scores, including:

[0009] S21, microbiome data collection and feature extraction: collect microbiome data of stroke patients, including intestinal flora and oral flora, and extract microbial features related to stroke, including flora diversity index, species abundance, and concentration of microbial metabolites;

[0010] S22, model construction and association analysis: Based on the extracted microbial features, a microbiome-driven health risk assessment model was constructed to evaluate the relationship between microbiome changes and the risk of stroke;

[0011] S23, health score generation: Based on the evaluation results of the relationship between microbiome changes and stroke risk, the patient's microbiome health score is calculated and risk is graded into low risk, medium risk, and high risk;

[0012] S3, multimodal data fusion and risk prediction: The pre-processed health data is fused into comprehensive data through multimodal data, and combined with the results of microbiome-driven health risk assessment to predict the risk of stroke patients;

[0013] S4, Generation of personalized intervention recommendations: Generate personalized health intervention recommendations based on the risk prediction results of stroke patients, including lifestyle adjustments, drug treatment recommendations, and microbiome regulation.

[0014] Optionally, the data collection and preprocessing in S1 includes:

[0015] S11, health data collection: real-time collection of health data of stroke patients, including:

[0016] Clinical data: Extract patients’ clinical data through the electronic health record (EHR) platform, including diagnostic records, blood test results, blood pressure parameters, and blood sugar parameters;

[0017] Imaging data: brain structure images are obtained from computed tomography (CT) and cerebrovascular imaging data are obtained from magnetic resonance angiography (MRA);

[0018] Lifestyle data: Collect the patient’s eating habits, exercise frequency, smoking and drinking status through smart devices;

[0019] S12, data denoising: Gaussian filtering is used to eliminate random noise in health data;

[0020] S13, missing value filling: missing value filling is performed using the nearest neighbor interpolation method (KNN);

[0021] S14, data standardization processing: Z-score standardization method is used to eliminate data magnitude differences.

[0022] Optionally, the microbiome data collection and feature extraction in S21 includes:

[0023] S211, microbiome data collection: high-throughput sequencing technology (16S rRNA gene sequencing) was used to collect intestinal and oral flora data of stroke patients to generate microbiome sequence data;

[0024] S212, bacterial community diversity index calculation: Shannon diversity index was used to calculate the diversity of bacterial communities;

[0025] S213, species abundance extraction: based on the comparison results of the generated microbiome sequence data with the reference database (Silva database), the number of sequences of each bacterial community in the sample was counted, and the relative abundance of each bacterial community was calculated;

[0026] S214, calculation of microbial metabolite concentration: combined with the microbiome function prediction tool (PICRUSt), the types and concentrations of bacterial metabolites were inferred;

[0027] S215, feature extraction and integration: The bacterial diversity index, species abundance and metabolite concentration are used as features to construct a multidimensional feature matrix X.

[0028] Optionally, the microbiome-driven health risk assessment model in S22 adopts a convolutional neural network (CNN) model, and the convolutional neural network (CNN) model includes:

[0029] S221, multi-channel representation of input data: the constructed multi-dimensional feature matrix X is used as input data and represented as a multi-channel structure, where each channel represents bacterial diversity, species abundance and metabolite characteristics;

[0030] S222, adaptive weight convolution layer: uses adaptive weight convolution kernel to dynamically adjust the weight of each feature channel;

[0031] S223, feature fusion pooling layer: a cross-channel weighted global pooling layer is used to integrate local features into a global representation;

[0032] S224, microbiome feature interaction layer: adding a feature interaction layer to achieve feature fusion through cross-channel convolution;

[0033] S225, Output Layer and Risk Assessment: The output layer uses the Softmax activation function to map the final features to the classification probabilities of stroke risks.

[0034] Optionally, the health score generation in S23 includes:

[0035] S231, Health Score Calculation: Based on the evaluation result of the relationship between the microbiome change and the stroke occurrence risk, calculate the microbiome health score S of the patient;

[0036] S232, Risk Grading: Based on the calculated microbiome health score S, compare it with the preset lower risk threshold T1 and upper risk threshold T2, and grade the patient's risk, including low risk, medium risk, and high risk. Specifically, it includes:

[0037] Low Risk: When S ≤ T1, it is indicated as low risk;

[0038] Medium Risk: When T1 < S ≤ T2, it is indicated as medium risk;

[0039] High Risk: When S > T2, it is indicated as high risk.

[0040] Optionally, the multimodal data fusion and risk prediction in S3 include:

[0041] S31, Multimodal Data Fusion: Fuse the preprocessed health data to generate comprehensive data;

[0042] S32, Comprehensive Data Judgment and Risk Prediction: Judge the fused comprehensive data and make conditional judgments in combination with the grading result of the microbiome health score to predict the stroke risk result of the patient, including at risk and not at risk.

[0043] Optionally, the multimodal data fusion in S31 includes:

[0044] S311, Feature Concatenation and Comprehensive Representation: Concatenate the feature matrices of the preprocessed health data of different modalities into a comprehensive feature matrix;

[0045] S312, Comprehensive Data Generation: Based on the concatenated comprehensive feature matrix, generate comprehensive data X through the weighted summation algorithm 综合数据 。

[0046] Optionally, the comprehensive data judgment and risk prediction in S32 includes:

[0047] S321, Threshold Judgment of Comprehensive Data: Based on the generated comprehensive data X 综合数据 , by comparing it with the set comprehensive risk threshold T 综合 When X 综合数据 > T综合 When X 综合数据 ≤T 综合 , it means there is no potential risk;

[0048] S322, Risk Prediction: Combine the results of risk grading to predict the final risk, including:

[0049] At risk: When the risk classification is medium risk or high risk, and the threshold of the comprehensive data is judged as potential risk, it means that the stroke patient is at risk;

[0050] No risk: When the risk classification is low risk and the threshold of the comprehensive data is judged as no potential risk, it means that the stroke patient is at no risk.

[0051] Optionally, the generation of personalized intervention suggestions in S4 includes:

[0052] S41, Lifestyle adjustment: For patients predicted to be at risk, lifestyle adjustment measures are recommended, including:

[0053] Dietary adjustment: Patients are advised to reduce salt intake and increase foods rich in dietary fiber and antioxidants (such as fruits and vegetables);

[0054] Exercise planning: Aerobic exercise (such as walking and swimming) is recommended, and an exercise plan should be developed after consulting a doctor;

[0055] S42, Drug treatment recommendations: Provide drug treatment recommendations for patients predicted to be at risk, including:

[0056] Antiplatelet drugs: Antiplatelet drugs (such as aspirin) are recommended to prevent thrombosis;

[0057] Antihypertensive drugs: Antihypertensive drugs (such as calcium channel blockers or angiotensin receptor blockers) are recommended to lower blood pressure;

[0058] Lipid-lowering drugs: Lipid-lowering drugs (such as statins) are recommended to control blood lipid levels;

[0059] S43, Microbiome Modulation: For patients at risk, probiotics or dietary supplementation are recommended to regulate the microbiome balance, including:

[0060] Probiotic supplementation: Probiotic preparations rich in lactic acid bacteria or bifidobacteria are recommended;

[0061] Dietary supplementation: Patients are advised to increase their dietary fiber intake, including whole grains and legumes.

[0062] A stroke patient risk prediction system based on large model technology is used to implement the above-mentioned stroke patient risk prediction method based on large model technology, and includes the following modules:

[0063] Data collection and preprocessing module: collects health data of stroke patients from multiple sources, including clinical data, imaging data, and lifestyle data, and preprocesses the collected health data, including denoising, filling missing values, and standardization;

[0064] Microbiome health risk assessment module: collects microbiome data of stroke patients, builds a microbiome-driven health risk assessment model, evaluates the relationship between microbiome changes and stroke risk, and generates microbiome health scores for patients;

[0065] Multimodal data fusion and risk prediction module: The pre-processed health data is fused into comprehensive data through multimodal data, and combined with the microbiome health score to predict the risk of stroke patients;

[0066] Personalized intervention recommendation generation module: Generate personalized health intervention recommendations based on the risk prediction results of stroke patients, including lifestyle adjustments, drug treatment recommendations, and microbiome regulation.

[0067] Beneficial effects of the present invention:

[0068] The present invention constructs a multimodal and dynamic stroke risk prediction method by fusing multi-source health data. It uses a microbiome-driven health risk assessment model and a multimodal data fusion mechanism to comprehensively capture the key factors of stroke risk. The data collection and preprocessing module improves the quality and consistency of the data through Gaussian filtering, missing value filling and standardization methods, providing a highly reliable input basis for the model, thereby significantly improving the accuracy and robustness of risk assessment.

[0069] The present invention, by adopting a convolutional neural network model, combined with multi-channel input, adaptive weight convolution layer, feature interaction layer and dynamic attention mechanism, realizes the deep expression of microbiome characteristics and the modeling ability of complex associations. The model can accurately evaluate the relationship between microbiome changes and stroke risk, generate health scores and perform scientific low, medium and high risk classification. In addition, the comprehensive data generated based on multimodal fusion is combined with the grading results of the health score, and through flexible conditional judgment, fast and efficient stroke risk prediction is achieved, without the need for additional complex model training, which enhances the applicability and practicality of the prediction method.

[0070] The present invention, through personalized health management needs and according to the patient's risk prediction results, generates comprehensive health intervention recommendations covering lifestyle adjustments, drug treatment recommendations and microbiome regulation. In particular, the introduction of microbiome regulation provides a new scientific basis for precise intervention in stroke risk. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0072] Figure 1 A schematic diagram of a prediction method flow chart of an embodiment of the present invention;

[0073] Figure 2 Schematic diagram of system function modules according to an embodiment of the present invention. DETAILED DESCRIPTION

[0074] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0075] like Figure 1 As shown, a method for predicting the risk of stroke patients based on large model technology includes the following steps:

[0076] S1, data collection and preprocessing: collect health data of stroke patients from multiple sources, including clinical data, imaging data, and lifestyle data, and preprocess the collected health data, including denoising, filling missing values, and standardization;

[0077] S2, Microbiome-driven health risk assessment: Collect microbiome data of stroke patients, build a microbiome-driven health risk assessment model, evaluate the relationship between microbiome changes and stroke risk, and generate patients' microbiome health scores, including:

[0078] S21, microbiome data collection and feature extraction: collect microbiome data of stroke patients, including intestinal flora and oral flora, and extract microbial features related to stroke, including flora diversity index, species abundance, and concentration of microbial metabolites;

[0079] S22, model construction and association analysis: Based on the extracted microbial features, a microbiome-driven health risk assessment model was constructed to evaluate the relationship between microbiome changes and the risk of stroke;

[0080] S23, health score generation: Based on the evaluation results of the relationship between microbiome changes and stroke risk, the patient's microbiome health score is calculated and risk is graded into low risk, medium risk, and high risk;

[0081] S3, multimodal data fusion and risk prediction: The pre-processed health data is fused into comprehensive data through multimodal data, and combined with the results of microbiome-driven health risk assessment to predict the risk of stroke patients;

[0082] S4, generation of personalized intervention recommendations: Generate personalized health intervention recommendations based on risk prediction results for stroke patients, including lifestyle adjustments, drug treatment recommendations, and microbiome regulation;

[0083] Through the above content, full use is made of multi-source health data and microbiome characteristics, which enhances the model's understanding of complex physiological and pathological relationships, while providing dynamic and personalized intervention recommendations, which helps to improve the accuracy and practicality of risk prediction and provides a scientific basis for precision medicine and personalized health management.

[0084] Data collection and preprocessing in S1 include:

[0085] S11, health data collection: real-time collection of health data of stroke patients, including:

[0086] Clinical data: Extract patients’ clinical data through the electronic health record (EHR) platform, including diagnostic records, blood test results, blood pressure parameters, and blood sugar parameters;

[0087] Imaging data: brain structure images are obtained from computed tomography (CT) and cerebrovascular imaging data are obtained from magnetic resonance angiography (MRA);

[0088] Lifestyle data: Collect the patient’s eating habits, exercise frequency, smoking and drinking status through smart devices;

[0089] S12, data denoising: Gaussian filtering is used to eliminate random noise in health data and improve data reliability, which is expressed as:

[0090]

[0091] Among them, y(t) is the denoised data, x(t ′ ) is the original data, σ is the smoothing parameter, t is the time point of the filtered signal, t ′is the input signal x(t ′ ) time point;

[0092] S13, missing value filling: The missing value filling method based on the nearest neighbor interpolation method (KNN) is used to make up for the problem of missing data, which is expressed as:

[0093]

[0094] Among them, x i For missing values, KNN(x i ) represents the k nearest neighbor data, x j is the value of the neighbor data point;

[0095] S14, data standardization: Z-score standardization method is used to eliminate data magnitude differences and ensure that different data sources contribute the same amount to the model, expressed as:

[0096]

[0097] Among them, x is the original value, μ is the mean of the feature, σ ′ is the standard deviation, z is the standardized value;

[0098] Through the above content, multidimensional information such as clinical, imaging and lifestyle is fully integrated. At the same time, Gaussian filtering, missing value filling and standardization are used to improve the quality and consistency of the data. It can effectively remove noise, make up for data missing, and ensure the availability and complementarity of different types of data in the risk prediction model, providing an accurate and reliable input basis for stroke risk assessment, thereby improving the accuracy and robustness of model predictions.

[0099] Microbiome data collection and feature extraction in S21 include:

[0100] S211, microbiome data collection: high-throughput sequencing technology (16S rRNA gene sequencing) was used to collect intestinal and oral flora data of stroke patients to generate microbiome sequence data;

[0101] S212, Calculation of bacterial diversity index: The Shannon diversity index is used to calculate the diversity of the bacterial community, expressed as:

[0102]

[0103] Among them, H is the diversity index, S is the total number of species detected, and p i is the relative abundance of the i-th species;

[0104] S213, species abundance extraction: According to the comparison results of the generated microbiome sequence data with the reference database (Silva database), the number of sequences of each bacterial community in the sample was counted, and the relative abundance of each bacterial community was calculated, expressed as:

[0105]

[0106] Among them, p i is the relative abundance of bacterial group i, A i is the number of sequences of bacterial community i, and B is the total number of all sequences in the sample;

[0107] S214, calculation of microbial metabolite concentration: combined with the microbiome function prediction tool (PICRUSt), the types and concentrations of bacterial metabolites were inferred and expressed as:

[0108]

[0109] Among them, C j is the concentration of metabolite j, p i is the relative abundance of bacterial group i, f ij is the functional coefficient of bacterial metabolite j (derived from the database);

[0110] S215, feature extraction and integration: The bacterial diversity index, species abundance and metabolite concentration are used as features to construct a multidimensional feature matrix X, which is expressed as:

[0111] X={H,p1,p2,...p S ,C1,C2,...,C M};

[0112] Among them, H is the diversity index, p1, p2, ... p S are the relative abundances of species 1, 2, ..., S, C1, C2, ..., C M are the concentrations of metabolites 1, 2, ..., M, respectively, and S and M are the total number of species and metabolites, respectively;

[0113] Through the above content, the stroke-related bacterial flora can be accurately identified, and key characteristics such as diversity index, species abundance and metabolite concentration can be quantified. This can comprehensively capture the multidimensional changes of the microbiome and reveal the potential association between the bacterial flora and stroke risk, providing high-quality basic data for subsequent risk assessment and personalized intervention, significantly improving the scientificity and accuracy of the prediction.

[0114] The microbiome-driven health risk assessment model in S22 uses a convolutional neural network (CNN) model, which includes:

[0115] S221, multi-channel representation of input data: the constructed multi-dimensional feature matrix X is used as input data and represented as a multi-channel structure, where each channel represents bacterial diversity, species abundance and metabolite characteristics;

[0116] S222, adaptive weight convolution layer: To address the differences in the importance of microbiome features, an adaptive weight convolution kernel is used to dynamically adjust the weights of each feature channel, expressed as:

[0117]

[0118] in, is the activation value of position (i, j) after the l-th layer convolution, is the weight of the adaptive convolution kernel, dynamically learned for input channel k, X k is the feature matrix of input channel k, is the bias term, σ is the ReLU activation function, and C is the number of channels of the input data;

[0119] S223, feature fusion pooling layer: In order to combine the global information of bacterial flora features, a cross-channel weighted global pooling layer is used to integrate local features into a global representation while retaining the key information of each channel, expressed as:

[0120]

[0121] Among them, P k is the global pooling result of channel k, H ij,k is the convolution eigenvalue of channel k, α ij,k is the weight coefficient of position (i, j), which is dynamically generated by the attention mechanism;

[0122]

[0123]

[0124] Among them, α ij,k is the weighting coefficient of position (i, j) and channel k, e ij,k is the attention score of position (i,j), ∑ i′,j′ exp(e i′j′,k ) is a normalization term used to ensure that the sum of the weights of all positions is 1, H ij,k is the input feature value, representing the convolution feature of channel k at position (i, j), f(H ij,k ) is the tanh feature mapping function, W k is a weight vector, indicating the feature importance of channel k, b k is the bias term;

[0125] S224, microbiome feature interaction layer: In order to model the nonlinear interaction between microbial diversity, species abundance and metabolites, a feature interaction layer is added to achieve feature fusion through cross-channel convolution, which is expressed as:

[0126]

[0127] in, is the feature generated after interaction, is a cross-channel convolution kernel used to model feature P m and P n The interaction, P m and P n are the global pooling results of the mth and nth channels, respectively, and b (l) is the bias term;

[0128] S225, output layer and risk assessment: The output layer uses the Softmax activation function to map the final features into the classification probability of stroke risk, expressed as:

[0129]

[0130] Among them, P(y=c|X) is the probability of belonging to category c, W c and b c are the weight and bias of the classifier, F is the output of the feature interaction layer, c′ is the category index of the classification, and W c′ and b c′ are the weight and bias of category c′ respectively;

[0131] Through the above content, the model's ability to express microbiome characteristics and model nonlinear associations has been comprehensively enhanced. It can accurately capture the complex patterns of bacterial diversity, species abundance and metabolite concentrations, and at the same time highlight the contribution of key features through a dynamic weighting mechanism. By combining local and global information, a more accurate prediction of stroke risk can be achieved. In addition, the model has stronger interpretability and can reveal the potential relationship between microbiome changes and stroke risk, providing a scientific basis for precision medicine and personalized health management.

[0132] The health score generation in S23 includes:

[0133] S231, health score calculation: Based on the evaluation results of the relationship between microbiome changes and the risk of stroke, the patient's microbiome health score S is calculated, expressed as:

[0134]

[0135] Where S is the patient's microbiome health score, P(y=c|X) is the probability distribution of the output category c, indicating the predicted probability that the patient belongs to category c, and wc The weight for category c;

[0136] S232, Risk grading: Based on the calculated microbiome health score S, compare it with the preset lower risk threshold T1 and upper risk threshold T2 to grade the patient's risk, including low risk, medium risk, and high risk, specifically including:

[0137] Low risk: When S ≤ T1, it is represented as low risk;

[0138] Medium risk: When T1 < S ≤ T2, it is represented as medium risk;

[0139] High risk: When S > T2, it is represented as high risk;

[0140] The lower risk threshold T1 and upper risk threshold T2 are set based on the statistical distribution of historical patient data, specifically including:

[0141] Collect the health score data of historical patients: Calculate the microbiome health score S based on historical patient data;

[0142] Calculate the mean and standard deviation of the health score: Based on the calculated microbiome health score S, calculate the mean and standard deviation of the health score, expressed as:

[0143]

[0144]

[0145] where, μ S is the mean of the health score S, S i is the health score of the i-th patient, N is the total number of patients, and σ S is the standard deviation of the health score S;

[0146] Set the risk grading thresholds: Based on the calculated mean and standard deviation of the health score, set the risk thresholds, expressed as:

[0147] T1 = μ S - σ S ;

[0148] T2 = μ S + σ S ;

[0149] Through the above content, calculate the patient's microbiome health score and achieve scientific grading of low, medium, and high risks, ensure the accuracy of score generation and risk grading, be able to clarify the impact of key microbial characteristics on the score and risk, and provide reliable support for precision medicine and personalized health management.

[0150] The multi-modal data fusion and risk prediction in S3 include:

[0151] S31, multimodal data fusion: fusion of pre-processed health data to generate comprehensive data;

[0152] S32, comprehensive data judgment and risk prediction: by judging the integrated data after fusion and combining the grading results of the microbiome health score to make conditional judgments, the patient's stroke risk results, including risk and no risk, are predicted;

[0153] Through the above content, it is possible to fully integrate information from multi-source data, improve the comprehensiveness and accuracy of risk assessment, and achieve accurate stroke risk classification (at risk or not) without complex model training. At the same time, it has good adaptability and can dynamically adjust the fusion mechanism and judgment rules. It is suitable for actual scenarios of different patient data and provides a scientific basis for personalized health management and intervention.

[0154] Multimodal data fusion in S31 includes:

[0155] S311, feature concatenation and comprehensive representation: The feature matrices of preprocessed health data of different modalities are concatenated into a comprehensive feature matrix, which is represented as:

[0156] X 综合 =[X 临床 ,X 影像 ,X 生活方式 +;

[0157] Among them, X 综合 is the comprehensive feature matrix, X 临床 , X 影像 , X 生活方式 They are the feature matrices of clinical data, imaging data, and lifestyle data, respectively;

[0158] S312, comprehensive data generation: Based on the spliced ​​comprehensive feature matrix, the comprehensive data X is generated through the weighted summation algorithm 综合数据 , expressed as:

[0159]

[0160] Among them, X 综合数据 is the final comprehensive data, which represents the data result after integrating all modal features. is the characteristic matrix of mode i, M is the total number of modes, w i is the weight of mode i;

[0161]

[0162] in, is the mean of the characteristic matrix of mode i;

[0163] Through the above content, the uniqueness of each modality feature can be retained, and the importance of different modalities can be dynamically adjusted through weights, which improves the scientificity and flexibility of data integration. The generated comprehensive data provides a comprehensive and accurate multi-dimensional input for stroke risk prediction, significantly improving the accuracy and practicality of the prediction.

[0164] The comprehensive data judgment and risk prediction in S32 include:

[0165] S321, threshold judgment of comprehensive data: based on the generated comprehensive data X 综合数据 , through the comprehensive risk threshold T 综合 For comparison, when X 综合数据 >T 综合 When X 综合数据 ≤T 综合 , it means there is no potential risk;

[0166] Comprehensive risk threshold T 综合 Based on historical data settings, it is expressed as:

[0167] Collect historical patient comprehensive data: Extract the generated comprehensive data from the historical database X 综合数据 sample;

[0168] Calculate the mean and standard deviation of the comprehensive data: Based on the generated comprehensive data X 综合数据 The mean and standard deviation of the sample comprehensive data are expressed as:

[0169]

[0170] Among them, μ X For X 综合数据 The mean value of X 综合数据,i is the comprehensive data value of the i-th sample, L is the total number of historical samples, σ X For X 综合数据 The standard deviation of

[0171] Set comprehensive risk threshold: Based on the calculated mean and standard deviation of comprehensive data, set the comprehensive risk threshold T 综合 , expressed as:

[0172] T 综合 =μ X +k·σ X ;

[0173] Where k is an adjustment parameter (set to 1 or 2);

[0174] S322, Risk Prediction: Combine the results of risk grading to predict the final risk, including:

[0175] At risk: When the risk classification is medium risk or high risk, and the threshold of the comprehensive data is judged as potential risk, it means that the stroke patient is at risk;

[0176] No risk: When the risk classification is low risk and the threshold of the comprehensive data is judged as no potential risk, it means that the stroke patient has no risk;

[0177] Through the above content, an accurate assessment of stroke risk can be achieved. First, potential risks are quickly screened through comprehensive data, and then the judgment is further refined based on the risk grading results to improve the accuracy and reliability of the prediction. It can be flexibly optimized according to the needs of different application scenarios, providing a scientific basis for personalized health management.

[0178] The generation of personalized intervention recommendations in S4 includes:

[0179] S41, Lifestyle adjustment: For patients predicted to be at risk, lifestyle adjustment measures are recommended, including:

[0180] Dietary adjustment: Patients are advised to reduce salt intake and increase foods rich in dietary fiber and antioxidants (such as fruits and vegetables);

[0181] Exercise planning: Aerobic exercise (such as walking and swimming) is recommended, and an exercise plan should be developed after consulting a doctor;

[0182] S42, Drug treatment recommendations: Provide drug treatment recommendations for patients predicted to be at risk, including:

[0183] Antiplatelet drugs: Antiplatelet drugs (such as aspirin) are recommended to prevent thrombosis;

[0184] Antihypertensive drugs: Antihypertensive drugs (such as calcium channel blockers or angiotensin receptor blockers) are recommended to lower blood pressure;

[0185] Lipid-lowering drugs: Lipid-lowering drugs (such as statins) are recommended to control blood lipid levels;

[0186] S43, Microbiome Modulation: For patients at risk, probiotics or dietary supplementation are recommended to regulate the microbiome balance, including:

[0187] Probiotic supplementation: Probiotic preparations rich in lactic acid bacteria or bifidobacteria are recommended;

[0188] Dietary supplementation: Patients are advised to increase their dietary fiber intake, including whole grains and legumes;

[0189] Through the above content, a full range of health intervention suggestions are provided for patients at risk, including lifestyle adjustments, drug treatment recommendations and microbiome regulation, which can accurately intervene in patients' health risks from multiple dimensions such as diet, exercise, drugs and microbiome balance. At the same time, the introduction of microbiome regulation provides a new means for stroke prevention, which not only improves the targetedness and effectiveness of intervention, but also is easy to apply, providing a practical solution for personalized health management.

[0190] like Figure 2 As shown, a stroke patient risk prediction system based on large model technology is used to implement the above-mentioned stroke patient risk prediction method based on large model technology, including the following modules:

[0191] Data collection and preprocessing module: collects health data of stroke patients from multiple sources, including clinical data, imaging data, and lifestyle data, and preprocesses the collected health data, including denoising, filling missing values, and standardization;

[0192] Microbiome health risk assessment module: collects microbiome data of stroke patients, builds a microbiome-driven health risk assessment model, evaluates the relationship between microbiome changes and stroke risk, and generates microbiome health scores for patients;

[0193] Multimodal data fusion and risk prediction module: The pre-processed health data is fused into comprehensive data through multimodal data, and combined with the microbiome health score to predict the risk of stroke patients;

[0194] Personalized intervention recommendation generation module: Generate personalized health intervention recommendations based on the risk prediction results of stroke patients, including lifestyle adjustments, drug treatment recommendations, and microbiome regulation.

[0195] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0196] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for predicting the risk of stroke patients based on large model technology, characterized in that: The following steps are involved: S1, data collection and preprocessing: collect health data of stroke patients from multiple sources, including clinical data, imaging data, and lifestyle data, and preprocess the collected health data, including denoising, filling missing values, and standardization; S2, Microbiome-driven health risk assessment: Collect microbiome data of stroke patients, build a microbiome-driven health risk assessment model, evaluate the relationship between microbiome changes and stroke risk, and generate patients' microbiome health scores, including: S21, microbiome data collection and feature extraction: collect microbiome data of stroke patients, including intestinal flora and oral flora, and extract microbial features related to stroke, including flora diversity index, species abundance, and concentration of microbial metabolites; S22, model construction and association analysis: Based on the extracted microbial features, a microbiome-driven health risk assessment model was constructed to evaluate the relationship between microbiome changes and the risk of stroke; S23, health score generation: Based on the evaluation results of the relationship between microbiome changes and stroke risk, the patient's microbiome health score is calculated and risk is graded into low risk, medium risk, and high risk; S3, multimodal data fusion and risk prediction: The pre-processed health data is fused into comprehensive data through multimodal data, and combined with the results of microbiome-driven health risk assessment to predict the risk of stroke patients; S4, Generation of personalized intervention recommendations: Generate personalized health intervention recommendations based on the risk prediction results of stroke patients, including lifestyle adjustments, drug treatment recommendations, and microbiome regulation.

2. The method for predicting the risk of stroke patients based on large model technology according to claim 1, characterized in that: The data collection and preprocessing in S1 include: S11, health data collection: real-time collection of health data of stroke patients, including: Clinical data: Extract patients’ clinical data through the electronic health record platform, including diagnostic records, blood test results, blood pressure parameters, and blood sugar parameters; Imaging data: brain structural images from computed tomography and cerebrovascular imaging data from magnetic resonance angiography; Lifestyle data: Collect the patient’s eating habits, exercise frequency, smoking and drinking status through smart devices; S12, data denoising: Gaussian filtering is used to eliminate random noise in health data; S13, missing value filling: missing value filling is performed based on the nearest neighbor interpolation method; S14, data standardization processing: Z-score standardization method is used to eliminate data magnitude differences.

3. The method for predicting the risk of stroke patients based on large model technology according to claim 1, characterized in that: The microbiome data collection and feature extraction in S21 include: S211, Microbiome data collection: High-throughput sequencing technology was used to collect intestinal and oral flora data of stroke patients to generate microbiome sequence data; S212, bacterial community diversity index calculation: Shannon diversity index was used to calculate the diversity of bacterial communities; S213, species abundance extraction: based on the comparison results of the generated microbiome sequence data with the reference database, the number of sequences of each bacterial community in the sample is counted, and the relative abundance of each bacterial community is calculated; S214, Calculation of microbial metabolite concentration: Combining microbiome function prediction tools to infer the types and concentrations of microbial metabolites; S215, Feature extraction and integration: Using the microbial diversity index, species abundance, and metabolite concentration as features to construct a multi-dimensional feature matrix X.

4. The method for predicting the risk of stroke patients based on large model technology according to claim 3, characterized in that: The microbiome-driven health risk assessment model in S22 uses a convolutional neural network model, which includes: S221, Multi-channel representation of input data: Using the constructed multi-dimensional feature matrix X as input data, represented as a multi-channel structure, where each channel represents microbial diversity, species abundance, and metabolite characteristics respectively; S222, Adaptive weight convolutional layer: Using an adaptive weight convolutional kernel to dynamically adjust the weights of each feature channel; S223, Feature fusion pooling layer: Using a cross-channel weighted global pooling layer to integrate local features into a global representation; S224, Microbiome feature interaction layer: Adding a feature interaction layer to achieve feature fusion through cross-channel convolution; S225, Output layer and risk assessment: The output layer uses the Softmax activation function to map the final features to the classification probability of stroke risk.

5. The method for predicting the risk of stroke patients based on large model technology according to claim 4, characterized in that: The health score generation in S23 includes: S231, Health score calculation: Based on the evaluation result of the relationship between microbiome changes and stroke occurrence risk, calculate the microbiome health score S of the patient; S232, Risk grading: Based on the calculated microbiome health score S, compare it with the preset lower risk threshold T1 and upper risk threshold T2 to grade the patient's risk, including low risk, medium risk, and high risk. Specifically, it includes: Low risk: When S ≤ T1, it is represented as low risk; Medium risk: When T1 < S ≤ T2, it is represented as medium risk; High risk: When S > T2, it is represented as high risk.

6. A method for predicting the risk of stroke patients based on large model technology according to claim 5, characterized in that: The multi-modal data fusion and risk prediction in S3 includes: S31, Multi-modal data fusion: Fusing the preprocessed health data to generate comprehensive data; S32, Comprehensive data judgment and risk prediction: Judging the fused comprehensive data and combining the grading results of the microbiome health score for conditional judgment to predict the stroke risk result of the patient, including at risk and not at risk.

7. The method for predicting the risk of stroke patients based on large model technology according to claim 6, characterized in that: The multi-modal data fusion in S31 includes: S311, Feature splicing and comprehensive representation: Splicing the feature matrices of preprocessed health data of different modalities into a comprehensive feature matrix; S312, comprehensive data generation: Based on the spliced ​​comprehensive feature matrix, the comprehensive data X is generated through the weighted summation algorithm 综合数据 .

8. The method for predicting the risk of stroke patients based on large model technology according to claim 7, characterized in that: The comprehensive data judgment and risk prediction in S32 includes: S321, threshold judgment of comprehensive data: based on the generated comprehensive data X 综合数据 , through the comprehensive risk threshold T 综合 For comparison, when X 综合数据 >T 综合 When X 综合数据 ≤T 综合 , it means there is no potential risk; S322, Risk prediction: Combining the results of risk grading to predict the final risk. Specifically, it includes: At risk: When the risk grading is medium risk or high risk, and the threshold judgment of the comprehensive data indicates the existence of potential risk, it means the stroke patient is at risk; Not at risk: When the risk grading is low risk and the threshold judgment of the comprehensive data indicates no potential risk, it means the stroke patient is not at risk.

9. The method for predicting the risk of stroke patients based on large model technology according to claim 8, characterized in that: The generation of personalized intervention suggestions in S4 includes: S41, Lifestyle adjustment: For patients with a predicted result of being at risk, recommend lifestyle adjustment measures. Specifically, it includes: Dietary modification: Patients are advised to reduce salt intake and increase foods rich in dietary fiber and antioxidants; Exercise planning: Aerobic exercise is recommended, and an exercise plan should be developed after consulting a doctor; S42, Drug treatment recommendations: Provide drug treatment recommendations for patients predicted to be at risk, including: Antiplatelet drugs: Antiplatelet drugs are recommended to prevent thrombosis; Antihypertensive drugs: Antihypertensive drugs are recommended to lower blood pressure; Lipid-lowering drugs: Lipid-lowering drugs are recommended to control blood lipid levels; S43, Microbiome Modulation: For patients at risk, probiotics or dietary supplementation are recommended to regulate the microbiome balance, including: Probiotic supplementation: Probiotic preparations rich in lactic acid bacteria or bifidobacteria are recommended; Dietary supplementation: Patients are advised to increase their dietary fiber intake, including whole grains and legumes.

10. A stroke patient risk prediction system based on large model technology, used to implement a stroke patient risk prediction method based on large model technology as described in any one of claims 1 to 9, characterized in that: Includes the following modules: Data collection and preprocessing module: collects health data of stroke patients from multiple sources, including clinical data, imaging data, and lifestyle data, and preprocesses the collected health data, including denoising, filling missing values, and standardization; Microbiome health risk assessment module: collects microbiome data of stroke patients, builds a microbiome-driven health risk assessment model, evaluates the relationship between microbiome changes and stroke risk, and generates microbiome health scores for patients; Multimodal data fusion and risk prediction module: The pre-processed health data is fused into comprehensive data through multimodal data, and combined with the microbiome health score to predict the risk of stroke patients; Personalized intervention recommendation generation module: Generate personalized health intervention recommendations based on the risk prediction results of stroke patients, including lifestyle adjustments, drug treatment recommendations, and microbiome regulation.

Images