Chronic disease early warning intervention system based on artificial intelligence

Abstract

The invention relates to the field of chronic disease early warning, and discloses a chronic disease early warning intervention system based on artificial intelligence, and the system comprises a data collection module which is used for collecting multi-modal data, and the multi-modal data comprises gene detection data, proteomics data, microbiomics data, environment data and social economic data; the pre-processing module is used for pre-processing the multi-modal data to obtain basic data, and the pre-processing comprises noise removal, abnormal value and missing value processing and targeted data cleaning and feature extraction operation on the data of different modals; and a conversion fusion module. Through cooperative operation of all the modules, a complete and automatic chronic disease early warning intervention process is formed, an organic whole is formed from data acquisition, risk assessment and intervention measure formulation and implementation, and the chronic disease risk whole process management and control are achieved, so that the chronic disease prevention and treatment efficiency and effect are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of chronic disease early warning, and specifically to an artificial intelligence-based chronic disease early warning intervention system. Background Art

[0002] With the rapid development of modern medical technology and the increasing demand for health management, the prevention and treatment of chronic diseases has become an important research topic in the global medical field. Traditional early warning of chronic diseases mainly relies on clinical symptoms, limited physiological index tests and patient self-reports. It usually judges the risk of chronic diseases based on routine physical examination indicators such as blood pressure, blood sugar, blood lipids, etc., combined with family medical history and self-reported symptoms of patients.

[0003] However, it is difficult to comprehensively and accurately assess an individual's chronic disease risk based on limited indicators, because many chronic diseases may not have obvious clinical symptoms in the early stages or only show slight changes in physiological indicators, which can easily be overlooked. Therefore, an artificial intelligence-based chronic disease early warning intervention system is proposed to solve the above problems. Summary of the invention

[0004] In response to the shortcomings of the existing technology, the present invention provides an artificial intelligence-based chronic disease early warning intervention system to solve the problem that traditional chronic disease early warning systems are difficult to comprehensively and accurately assess an individual's chronic disease risk based on only limited indicators, because many chronic diseases may not have obvious clinical symptoms in the early stages or only show slight changes in physiological indicators and are easily overlooked.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: an artificial intelligence-based chronic disease early warning intervention system, comprising:

[0006] Data collection module: used to collect multimodal data, including genetic testing data, proteomics data, microbiome data, environmental data and socioeconomic data;

[0007] Preprocessing module: used to preprocess the multimodal data to obtain basic data, wherein the preprocessing includes removing noise, outliers and missing values, and performing targeted data cleaning and feature extraction operations on data of different modalities;

[0008] Conversion and fusion module: used to convert and fuse basic data to obtain fused data;

[0009] Building training modules: used to build and train early warning models by integrating data;

[0010] Evaluation and optimization module: used to evaluate and optimize the early warning model to obtain an early warning model for chronic diseases;

[0011] Early warning and intervention module: used to provide early warning and intervention based on the chronic disease early warning model, including formulating intervention plans and sending early warning notifications and suggestions.

[0012] Preferably, the data acquisition module includes a gene detection acquisition unit, a proteomics acquisition unit, a microbiome acquisition unit, an environment acquisition unit, and a socioeconomic acquisition unit. The gene detection acquisition unit is used to collect whole genome sequencing data or gene chip data, and the sequencing depth of the whole genome sequencing data is not less than 30x. The proteomics acquisition unit is used to collect proteomics data, and the proteomics data includes the identification results and quantitative information of proteins obtained from the patient's biological samples by liquid chromatography-mass spectrometry. The microbiome acquisition unit is used to collect microbiome data, and the microbiome data includes microbial community composition and diversity information obtained by high-throughput sequencing analysis of the patient's feces, oral swab or skin swab samples. The environment acquisition unit is used to collect environmental data, and the environmental data includes air quality, water quality, noise, temperature and humidity. The socioeconomic acquisition unit is used to collect socioeconomic data, and the socioeconomic data includes age, gender, occupation, education level, income level and family population information.

[0013] Preferably, the preprocessing module includes a preprocessing unit, a communication unit, and a database unit. The preprocessing unit is used to perform quality assessment, data cleaning, data conversion and normalization operations on the collected multimodal data to obtain basic data. The communication unit is used for data transmission between units and information interaction between modules. The database unit is used to store the multimodal data before and after preprocessing, providing data storage and retrieval services for subsequent data conversion, fusion and model construction, while ensuring the security and integrity of the data.

[0014] Preferably, the conversion and fusion module includes a data conversion unit and a data fusion unit. The data conversion unit is used to perform format conversion and feature reshaping on basic data. The data fusion unit is used to fuse the basic data after format conversion and feature reshaping at the feature level and the decision level. At the feature level, feature matrices of different modalities are weighted and spliced, and dimension reduction is performed through principal component analysis or linear discriminant analysis. At the decision level, the results of independent prediction models based on different modal data are fused to obtain fused data.

[0015] Preferably, the construction training module includes a fusion model unit, a hybrid model unit, a data set division unit, and a model training unit. The fusion model unit is used to construct an early warning model combining a convolutional neural network and a recurrent neural network according to the characteristics of the fusion data and the requirements of the chronic disease prediction task, wherein the convolution kernel size of the convolutional neural network is 3x3, the step size is 1, the activation function is ReLU, the recurrent neural network adopts a long short-term memory network or a gated recurrent unit, the number of hidden units is 128, and the number of layers is 2. The characteristics of the fusion data include the dimension of the data, the relationship between different characteristics, and the distribution of the data. The requirements of the chronic disease prediction task include binary classification prediction, multi-classification prediction, and chronic disease characteristics. The hybrid model unit is used according to A decision tree model combining a multilayer perceptron and a decision tree is constructed based on the characteristics of the fused data and clinical experience, wherein the number of nodes in the input layer of the multilayer perceptron is determined according to the dimension of the fused feature vector, the hidden layer is 2-3 layers, the number of nodes is 256 and 128 respectively, the activation function is ReLU, the depth of the decision tree is 5-10 layers, the characteristics of the fused data include linear or nonlinear relationship, data volume, the data set partitioning unit is used to divide the fused data into a training set, a validation set and a test set in a ratio of 70%:15%:15%, and the model training unit is used to train the model using stochastic gradient descent or Adam optimization algorithm, the training cycle is 100-200, and the corresponding loss function is selected according to the prediction task.

[0016] Preferably, when the model training unit performs binary classification prediction, the loss function used is ,in is the number of training samples, is the true label of the sample, is the model prediction probability; when performing multi-classification prediction, the loss function used is ,in is the number of categories, For sample Belongs to category The real label, Predict samples for the model Belongs to category probability.

[0017] Preferably, the evaluation and optimization module includes a model evaluation unit and a model optimization unit. The model evaluation unit is used to calculate the indicators of the early warning model using the test set data and analyze the confusion matrix. The indicators include accuracy, recall rate, F1 value, and area under the receiver operating characteristic curve. The model optimization unit is used to optimize the early warning model to obtain an early warning model for chronic diseases. The optimization includes adding L1 or L2 regularization terms, Dropout technology or early stopping method when overfitting, increasing the number of hidden layers or neurons in the model, adjusting the activation function or using a more complex model when underfitting, and using Bayesian optimization, Hyperband optimization or expanding the search range of hyperparameters when optimizing hyperparameters.

[0018] Preferably, the model optimization unit, when using L1 regularization, adds ,in are the parameters of the model, is the regularization strength coefficient. When using L2 regularization, add .

[0019] Preferably, the early warning intervention module includes an early warning unit, an intervention plan generating unit, and a feedback interaction unit. The early warning unit is used to send early warning information based on the prediction results of the chronic disease early warning model when the predicted risk exceeds a preset threshold, and the warning information includes the disease type, risk level and possible onset time. The intervention plan generating unit is used to formulate a personalized intervention plan for the patient based on the patient's individual data and prediction results, combined with clinical experience and medical guidelines, and the intervention plan includes medication adjustment, diet plan formulation, exercise plan formulation and lifestyle change recommendations. The feedback interaction unit is used to receive feedback information from the patient on the intervention plan, and to perform human-computer interaction, and the feedback information includes changes in health indicators and execution status.

[0020] An artificial intelligence-based early warning intervention method for chronic diseases, comprising the following steps:

[0021] S1. Data collection: The whole genome sequencing or gene chip data is obtained through the gene detection collection unit. The proteomics collection unit uses liquid chromatography-mass spectrometry to collect protein identification and quantitative information. The microbiome collection unit uses high-throughput sequencing of patient-specific samples to analyze microbial community data. The environmental collection unit collects air quality, water quality, noise, temperature and humidity data. The socioeconomic collection unit collects patient socioeconomic information and summarizes them into multimodal data.

[0022] S2, data preprocessing: the preprocessing unit then performs quality assessment, noise outlier and missing value processing, specific mode cleaning, data conversion and normalization on the multimodal data to obtain basic data, the communication unit circulates the data and interacts with the modules, and the database unit stores the data;

[0023] S3, data conversion and fusion: the data conversion unit then processes the format and features of the basic data, the data fusion unit performs weighted splicing and dimension reduction at the feature level, and fuses the results of independent models at the decision level to generate fused data;

[0024] S4, model construction and training: the fusion model unit then builds a CNN-RNN early warning model based on the fusion data characteristics and chronic disease prediction tasks, the hybrid model unit builds an MLP-decision tree model based on the fusion data characteristics and clinical experience, and the data set division unit divides the fusion data into 70%:15%:15%; the model training unit uses SGD or Adam algorithm to select the loss function to train the model according to the prediction task;

[0025] S5. Model evaluation and optimization: The model evaluation unit then uses the test set to calculate the model's indicators and analyze the confusion matrix. The model optimization unit uses corresponding strategies to optimize the model for overfitting, underfitting, and hyperparameter problems, and obtains an early warning model for chronic diseases. The corresponding strategies include L1 / L2 regularization, Dropout, early stopping method, adjusting the model structure, and Bayesian optimization.

[0026] S6. Early warning intervention: The early warning unit then sends early warning information when the threshold is exceeded based on the results of the chronic disease early warning model. The intervention plan generation unit formulates an intervention plan. The feedback interaction unit receives patient feedback and human-computer interaction, adjusts the plan accordingly and stores data.

[0027] The present invention provides an artificial intelligence-based early warning intervention system for chronic diseases. It has the following beneficial effects:

[0028] 1. The present invention forms a complete and automated chronic disease early warning intervention process through the coordinated operation of various modules, from data acquisition to risk assessment to the formulation and implementation of intervention measures, forming an organic whole, and managing and controlling the whole process of chronic disease risks, improving the efficiency and effectiveness of chronic disease prevention and treatment, thereby solving the problem that traditional chronic disease early warning is difficult to comprehensively and accurately assess an individual's chronic disease risk based on only limited indicators, because many chronic diseases may not have obvious clinical symptoms in the early stages or only show slight changes in physiological indicators, which are easily overlooked.

[0029] 2. The present invention guides patients to improve their lifestyles and use drugs rationally through the formulation and implementation of personalized intervention plans, thereby effectively reducing the risk of chronic diseases or delaying disease progression. At the same time, it achieves the purpose of identifying the health risks of chronic disease patients in advance and conducting effective intervention.

[0030] 3. Through comprehensive collection and deep fusion analysis of multi-source data, the present invention can dig out more complex information and potential rules related to chronic diseases hidden behind the data, capture early signs of chronic disease risks more comprehensively and accurately, improve the accuracy of early warning of chronic diseases, reduce the occurrence of misjudgments and missed judgments, and gain more time for early intervention for patients.

[0031] 4. The present invention has significant advantages in formulating intervention plans based on the rich multimodal data collected and the accurate early warning model constructed and trained. It takes into account multiple factors such as individual differences at the genetic level, the uniqueness of protein expression and microbial communities in the body, the particularity of the environment, and the lifestyle determined by the socioeconomic background. It can tailor personalized intervention plans for each patient, better fit the actual situation of the patient, and improve the patient's compliance with the intervention plan, thereby more effectively preventing and controlling the occurrence and development of chronic diseases.

[0032] 5. The present invention constructs an early warning model by combining fusion model units and hybrid model units, making full use of the advantages of convolutional neural networks in feature extraction and the ability of recurrent neural networks to process sequence data, combining the powerful nonlinear mapping capabilities of multi-layer perceptrons and the interpretability advantages of decision trees based on rule judgments, and the model evaluation and optimization module continuously optimizes the model performance, so that it has better generalization ability, can adapt to the early warning task requirements of different populations, different types of chronic diseases and different data environments, and ensure the reliability and stability of the model in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a system architecture diagram of an artificial intelligence-based chronic disease early warning intervention system proposed by the present invention;

[0034] Figure 2 This is a data acquisition module architecture diagram of an artificial intelligence-based chronic disease early warning intervention system proposed by the present invention;

[0035] Figure 3 This is a preprocessing module architecture diagram of an artificial intelligence-based chronic disease early warning intervention system proposed by the present invention;

[0036] Figure 4 This is a conversion fusion module architecture diagram of an artificial intelligence-based chronic disease early warning intervention system proposed by the present invention;

[0037] Figure 5 This is a diagram of the training module architecture of an artificial intelligence-based chronic disease early warning intervention system proposed by the present invention;

[0038] Figure 6 This is an architecture diagram of an evaluation and optimization module of an artificial intelligence-based chronic disease early warning intervention system proposed by the present invention;

[0039] Figure 7 This is a diagram of the architecture of the early warning intervention module of the chronic disease early warning intervention system based on artificial intelligence proposed by the present invention;

[0040] Figure 8 This is a method flow chart of an artificial intelligence-based early warning intervention method for chronic diseases proposed by the present invention. DETAILED DESCRIPTION

[0041] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] Example:

[0043] Please refer to the attached Figure 1 , an embodiment of the present invention provides an artificial intelligence-based chronic disease early warning intervention system, comprising:

[0044] Data collection module: used to collect multimodal data, including genetic testing data, proteomics data, microbiome data, environmental data and socioeconomic data;

[0045] Preprocessing module: used to preprocess multimodal data to obtain basic data. Preprocessing includes removing noise, outliers and missing values, as well as performing targeted data cleaning and feature extraction operations on data of different modalities.

[0046] Conversion and fusion module: used to convert and fuse basic data to obtain fused data;

[0047] Building training modules: used to build and train early warning models by integrating data;

[0048] Evaluation and optimization module: used to evaluate and optimize the early warning model to obtain an early warning model for chronic diseases;

[0049] Early warning and intervention module: used for early warning and intervention based on the chronic disease early warning model. Intervention includes formulating intervention plans and sending early warning notifications and suggestions.

[0050] Specifically, by collecting multimodal data, we can comprehensively collect various potential information sources related to chronic diseases, covering data from genetic factors, microenvironment in the body, external environmental exposure to individual lifestyle and socioeconomic background, providing rich materials for accurate assessment of chronic disease risks.

[0051] By preprocessing multimodal data, basic data is obtained, which effectively improves data quality, reduces interference factors and erroneous information in the data, enables subsequent processing to be based on more reliable data, and improves the accuracy of analysis results.

[0052] By converting and fusing the basic data, we obtain fused data, thereby integrating the characteristics and information of data from different modalities, exploring the hidden associations and synergies between data from different modalities, and enhancing the expressiveness and usability of the data.

[0053] By integrating data to build and train early warning models, we can use advanced artificial intelligence algorithms to learn the rules and patterns in the data, build a model that can accurately identify the risk characteristics of chronic diseases, and achieve intelligent prediction of chronic disease risks.

[0054] By evaluating and optimizing the early warning model, an early warning model for chronic diseases is obtained, thereby ensuring that the model has good generalization ability, accuracy and stability, and can reliably play a warning role in actual application scenarios.

[0055] By conducting early warning and intervention based on the chronic disease early warning model, we can timely identify individuals with potential chronic disease risks, provide them with personalized early warning notifications and targeted intervention plans, and guide individuals to take preventive measures or adjust their lifestyles in advance, thereby achieving early and accurate warning and effective intervention of chronic diseases.

[0056] Through the coordinated operation of various modules, a complete and automated chronic disease early warning intervention process is formed. All links are closely connected and work together, from data acquisition to risk assessment to the formulation and implementation of intervention measures, forming an organic whole, thereby realizing full-process management and control of chronic disease risks, greatly improving the efficiency and effectiveness of chronic disease prevention and treatment, and solving the problem that traditional chronic disease early warning is based on limited indicators and is difficult to comprehensively and accurately assess an individual's chronic disease risk, because many chronic diseases may not have obvious clinical symptoms in the early stages or only show slight changes in physiological indicators, which are easily overlooked.

[0057] Please see attached Figure 2The data collection module includes a gene detection collection unit, a proteomics collection unit, a microbiome collection unit, an environmental collection unit, and a socioeconomic collection unit. The gene detection collection unit is used to collect whole genome sequencing data or gene chip data. The sequencing depth of the whole genome sequencing data is not less than 30x. The proteomics collection unit is used to collect proteomics data. The proteomics data includes the identification results and quantitative information of proteins obtained from the patient's biological samples by liquid chromatography-mass spectrometry. The microbiome collection unit is used to collect microbiome data. The microbiome data includes the composition and diversity of the microbial community obtained by high-throughput sequencing analysis of the patient's feces, oral swabs or skin swab samples. The environmental collection unit is used to collect environmental data. The environmental data includes air quality, water quality, noise, temperature and humidity. The socioeconomic collection unit is used to collect socioeconomic data. The socioeconomic data includes age, gender, occupation, education level, income level and family population information.

[0058] Specifically, by collecting whole genome sequencing data or gene chip data, we can obtain the individual's genetic code information, explore gene mutations and potential genetic susceptibility factors related to chronic diseases, and provide key genetic basis for subsequent accurate judgment of chronic disease risks, so as to analyze the potential tendency of chronic disease from the genetic root level.

[0059] By collecting proteomic data, it is possible to monitor the dynamic changes in the types, quantities and modification states of proteins in the body. These changes are often closely related to abnormalities in cell functions and metabolic pathways, which in turn helps to discover subtle abnormalities in protein expression in the early stages of chronic diseases, providing clues at the biomolecular level for early detection of disease signs, and accurately capturing changes in biological processes related to chronic diseases.

[0060] By collecting microbiome data, we can gain insight into the composition and diversity characteristics of microbial communities in the human intestines, oral cavity, skin and other parts. Because the imbalance of microbial communities is closely related to a variety of chronic diseases such as inflammatory bowel disease and diabetes, we can determine the intrinsic connection between microbial ecology and the occurrence and development of chronic diseases, and explore the inducing factors and potential risks of chronic diseases from the perspective of microbial ecology.

[0061] By collecting environmental data, we can understand the long-term external environmental conditions of individuals, such as the concentration of pollutants in the air, the quality of water, the intensity of noise, and changes in temperature and humidity. These environmental factors may directly or indirectly affect human health, and provide data support for evaluating the contribution of environmental factors to the risk of chronic diseases, thereby comprehensively considering the role of environmental exposure in the formation of chronic diseases.

[0062] By collecting socioeconomic data, we can learn about individuals' lifestyle characteristics, social status, and economic status. Basic information such as age and gender can help determine differences in disease risk among different populations. Factors such as occupation, education level, and income level can reflect living habits, stress levels, and the ability to acquire health resources, providing a social behavioral basis for formulating personalized and practical chronic disease prevention and intervention strategies, and improving the chronic disease risk assessment system from multiple dimensions.

[0063] Please see attached Figure 3 ,The preprocessing module includes a preprocessing unit, a ,communication unit, and a database unit. The preprocessing unit is used to perform quality ,assessment, data cleaning, data conversion and normalization operations on the ,collected multimodal data to obtain basic data. The communication unit is used for data ,transmission between units and information interaction between modules. The database unit is used to store ,multimodal data before and after preprocessing, providing data ,storage and retrieval services for subsequent data conversion, fusion, ,and model construction, while ensuring data security and integrity.

[0064] Specifically, basic data is obtained by performing quality assessment, data cleaning, data conversion and normalization operations on the collected multimodal data, thereby effectively filtering out erroneous information, outliers and noise interference in the data, and standardizing data of different formats and magnitudes, laying a solid foundation for subsequent accurate data analysis and model training, and improving the availability and reliability of data.

[0065] Through data transmission between units and information interaction between modules, data can be smoothly transferred throughout the system, so that each module can work together to avoid system failures caused by data transmission delays or errors, improve the efficiency and stability of system operation, and ensure the normal functioning of system functions.

[0066] By storing multimodal data before and after preprocessing, data storage and retrieval services are provided for subsequent data conversion, fusion and model building, while ensuring the security and integrity of the data, thereby facilitating retrospective analysis and comparative research of the data. Historical data can be used for reference verification during model optimization and system upgrades, and compliance guarantees are provided for data management to prevent data leakage and damage, ensuring that the data resources of the entire chronic disease early warning intervention system are properly managed and effectively utilized.

[0067] Please see attached Figure 4The conversion and fusion module includes a data conversion unit and a data fusion unit. The data conversion unit is used to perform format conversion and feature reshaping on the basic data. The data fusion unit is used to fuse the basic data after format conversion and feature reshaping at the feature level and decision level. At the feature level, the feature matrices of different modes are weighted and spliced, and dimension reduction is performed through principal component analysis or linear discriminant analysis. At the decision level, the results of independent prediction models based on different modal data are fused to obtain fused data.

[0068] Specifically, by converting the format and reshaping the features of basic data, data from different sources and of different types can have a unified form and feature expression that is suitable for subsequent fusion analysis, thereby solving the problem of data heterogeneity, and excavating deeper and more representative feature information of each modal data, thereby improving data quality and mineability, and creating good conditions for accurate data fusion.

[0069] By fusing the basic data after format conversion and feature reshaping at the feature level and the decision-making level, the key features of multimodal data are integrated at the feature level, the influence of important modes is highlighted through weighted splicing, and the dimension reduction technology is used to reduce data redundancy and complexity, extract core information, and integrate the results of independent prediction models of different modalities at the decision-making level. The advantages of each modality in chronic disease risk judgment are fully utilized, and they complement and verify each other to obtain more comprehensive, accurate and stable fused data, providing strong support for building a highly reliable early warning model for chronic diseases.

[0070] Please refer to the attached Figure 5, the training module includes a fusion model unit, a hybrid model unit, a data set division unit, and a model training unit. The fusion model unit is used to construct an early warning model combining a convolutional neural network and a recurrent neural network according to the characteristics of the fusion data and the needs of the chronic disease prediction task. The convolution kernel size of the convolutional neural network is 3x3, the step size is 1, and the activation function is ReLU. The recurrent neural network uses a long short-term memory network or a gated recurrent unit, the number of hidden units is 128, and the number of layers is 2. The characteristics of the fusion data include the dimension of the data, the relationship between different features, and the distribution of the data. The needs of the chronic disease prediction task include binary classification prediction, multi-classification prediction, and chronic disease characteristics. The hybrid model unit is used to A decision tree model combining a multilayer perceptron and a decision tree was constructed based on the characteristics of the data and clinical experience. The number of nodes in the input layer of the multilayer perceptron was determined according to the dimension of the fused feature vector, the hidden layer was 2-3 layers, the number of nodes was 256 and 128 respectively, the activation function was ReLU, the depth of the decision tree was 5-10 layers, and the characteristics of the fused data included linear or nonlinear relationships and data volume. The data set partitioning unit was used to divide the fused data into training set, validation set and test set in a ratio of 70%:15%:15%. The model training unit was used to train the model using stochastic gradient descent or Adam optimization algorithm. The training cycle was 100-200, and the corresponding loss function was selected according to the prediction task.

[0071] When the model training unit performs binary classification prediction, the loss function used is ,in is the number of training samples, is the true label of the sample, is the model prediction probability; when performing multi-classification prediction, the loss function used is ,in is the number of categories, For sample Belongs to category The real label, Predict samples for the model Belongs to category probability.

[0072] Specifically, by constructing an early warning model combining convolutional neural network and recurrent neural network according to the characteristics of fused data and the requirements of chronic disease prediction tasks, the convolutional neural network is used to extract local and global features in the fused data, and the recurrent neural network is used to process the sequence information of the data (such as time series related features, etc.). The advantages of both are fully combined to adapt to the characteristics of chronic disease prediction tasks, accurately capture the complex patterns and laws in the data, and improve the accuracy and generalization ability of chronic disease risk prediction.

[0073] By constructing a decision tree model that combines a multi-layer perceptron and a decision tree based on the characteristics of the fused data and clinical experience, the fused features are preliminarily processed with the help of the powerful nonlinear fitting ability of the multi-layer perceptron, and then the decision tree is used to further refine and optimize the results based on clinical experience and the inherent logic of the data, giving full play to the complementary advantages of the two, and effectively dealing with the linear or nonlinear relationship characteristics of the fused data, making the model more in line with the actual data situation, and improving the reliability of chronic disease risk judgment.

[0074] By dividing the fused data into training set, validation set and test set in a ratio of 70%:15%:15%, reasonable data support is provided for model training, parameter tuning and performance evaluation, ensuring that the model learns rules in a large amount of training data. At the same time, the validation set is used to avoid overfitting, and the generalization ability of the model on unseen data is objectively and accurately measured based on the test set, so that the performance of the final constructed model is stable and can be effectively applied to actual chronic disease warning scenarios.

[0075] By using stochastic gradient descent or Adam optimization algorithm to train the model, the model parameters are dynamically adjusted according to the set optimization algorithm, and the model is gradually optimized in the direction of reducing the loss function, so that it can converge to better parameter values ​​quickly and efficiently, allowing the model to better fit the chronic disease risk-related characteristics and laws contained in the training data, thereby improving the accuracy and performance of the model in predicting chronic disease risks.

[0076] Please refer to the attached Figure 6 The evaluation and optimization module includes a model evaluation unit and a model optimization unit. The model evaluation unit is used to calculate the indicators of the early warning model using the test set data and analyze the confusion matrix. The indicators include accuracy, recall rate, F1 value, and area under the receiver operating characteristic curve. The model optimization unit is used to optimize the early warning model to obtain an early warning model for chronic diseases. The optimization includes adding L1 or L2 regularization terms, Dropout technology or early stopping method when overfitting occurs, increasing the number of hidden layers or neurons in the model, adjusting the activation function, or using a more complex model when underfitting occurs, and using Bayesian optimization, Hyperband optimization, or expanding the search range of hyperparameters when optimizing hyperparameters.

[0077] Model optimization unit, when using L1 regularization, add in the loss function ,in are the parameters of the model, is the regularization strength coefficient. When using L2 regularization, add .

[0078] Specifically, by using the test set data to calculate the indicators of the early warning model and analyzing the confusion matrix, we can objectively and comprehensively measure the performance of the early warning model on data that was not actually trained, clearly understand the model's ability to accurately identify positive examples, recall positive examples, and comprehensively judge the balance between accuracy and recall, and intuitively display the model's discrimination ability through the area under the receiver operating characteristic curve. With the help of the confusion matrix, we can further gain insight into the correct and incorrect predictions of the model for different categories, and then accurately judge the advantages and disadvantages of the model, providing a clear direction for subsequent optimization and improvement.

[0079] By optimizing the early warning model, we can obtain an early warning model for chronic diseases, so that we can make effective adjustments to the problems of overfitting, underfitting, and unreasonable hyperparameters in the model, improve the generalization, fitting ability, and overall performance of the model, and enable the model to provide early warning of chronic disease risks more stably and accurately, reduce misjudgments and missed judgments, better adapt to complex and changeable actual application scenarios, and provide a reliable risk prediction basis for the effective intervention of chronic diseases.

[0080] Please see attached Figure 7 The early warning intervention module includes an early warning unit, an intervention plan generation unit, and a feedback interaction unit. The early warning unit is used to send early warning information based on the prediction results of the chronic disease early warning model. When the predicted risk exceeds the preset threshold, the warning information includes the disease type, risk level and possible onset time. The intervention plan generation unit is used to formulate personalized intervention plans for patients based on the patient's individual data and prediction results, combined with clinical experience and medical guidelines. The intervention plan includes drug adjustment, diet plan formulation, exercise plan formulation and lifestyle change recommendations. The feedback interaction unit is used to receive feedback from patients on the intervention plan and conduct human-computer interaction. The feedback information includes changes in health indicators and implementation status.

[0081] Specifically, based on the prediction results of the chronic disease early warning model, when the predicted risk exceeds the preset threshold, a warning message is sent so that patients and medical staff can know the potential chronic disease risk status in time, prompting patients to pay attention to their own health in advance and take corresponding measures. Medical staff can also quickly initiate the intervention process based on this, improve the timeliness of chronic disease prevention and control, and reduce the possibility of further development of the disease.

[0082] By developing personalized intervention plans for patients based on their individual data and predicted results, combined with clinical experience and medical guidelines, we can provide precisely adapted prevention and treatment strategies for their unique circumstances, thereby improving the effectiveness and operability of intervention measures. Whether it is the precise selection and dosage adjustment of drugs, or customized planning of diet, exercise and lifestyle, it will help patients better manage their own health and delay or prevent the occurrence and development of chronic diseases.

[0083] By receiving feedback from patients on the intervention plan and conducting human-computer interaction, we can dynamically adjust the intervention plan based on the patient's actual implementation status and changes in health indicators, forming a closed-loop management model. This makes the intervention plan more in line with the patient's real-time needs and physical reactions, enhances patient participation and compliance, continuously optimizes the effectiveness of chronic disease management, and maximizes the patient's health level and quality of life.

[0084] Please see attached Figure 8 , an artificial intelligence-based early warning intervention method for chronic diseases, comprising the following steps:

[0085] S1. Data collection: The whole genome sequencing or gene chip data is obtained through the gene detection collection unit. The proteomics collection unit uses liquid chromatography-mass spectrometry to collect protein identification and quantitative information. The microbiome collection unit uses high-throughput sequencing of patient-specific samples to analyze microbial community data. The environmental collection unit collects air quality, water quality, noise, temperature and humidity data. The socioeconomic collection unit collects patient socioeconomic information and summarizes them into multimodal data.

[0086] S2, data preprocessing: the preprocessing unit then performs quality assessment, noise outlier and missing value processing, specific mode cleaning, data conversion and normalization on the multimodal data to obtain basic data, the communication unit circulates the data and interacts with the modules, and the database unit stores the data;

[0087] S3, data conversion and fusion: the data conversion unit then processes the format and features of the basic data, the data fusion unit performs weighted splicing and dimension reduction at the feature level, and fuses the results of independent models at the decision level to generate fused data;

[0088] S4, model construction and training: the fusion model unit then builds a CNN-RNN early warning model based on the fusion data characteristics and chronic disease prediction tasks, the hybrid model unit builds an MLP-decision tree model based on the fusion data characteristics and clinical experience, and the data set division unit divides the fusion data into 70%:15%:15%; the model training unit uses SGD or Adam algorithm to select the loss function to train the model according to the prediction task;

[0089] S5, model evaluation and optimization: The model evaluation unit then uses the test set to calculate the model's indicators and analyze the confusion matrix. The model optimization unit uses corresponding strategies to optimize the model for overfitting, underfitting, and hyperparameter problems, and obtains an early warning model for chronic diseases. The corresponding strategies include L1 / L2 regularization, Dropout, early stopping method, adjusting the model structure, and Bayesian optimization.

[0090] S6. Early warning intervention: The early warning unit then sends early warning information when the threshold is exceeded based on the results of the chronic disease early warning model. The intervention plan generation unit formulates an intervention plan. The feedback interaction unit receives patient feedback and human-computer interaction, adjusts the plan accordingly and stores data.

[0091] Specifically, through the data collection step, multimodal data covering genes, proteins, microorganisms, environment, socio-economics and other aspects are widely collected to provide a rich source of information for a comprehensive understanding of individual health status. Data preprocessing purifies and standardizes these complex data to improve data quality. Data conversion and fusion further integrates data value and explores potential correlations between data to provide a strong data foundation for model construction and training. Model construction and training use advanced algorithms to build a predictive model that fits the characteristics of chronic diseases. Model evaluation and optimization ensure that the model performance is reliable and accurate. The early warning intervention step promptly detects risks and formulates targeted measures based on model results, and can continuously optimize intervention strategies based on patient feedback.

[0092] This has enabled all-round, multi-dimensional and dynamic monitoring and management of chronic disease risks, breaking through the limitations of traditional single indicator early warning, comprehensively considering the interaction and influence of many factors, and evaluating and predicting chronic disease risks in a more scientific, systematic and intelligent way. Through the formulation and implementation of personalized intervention plans, it has guided patients to improve their lifestyles, use medications rationally, etc., effectively reducing the risk of chronic diseases or delaying disease progression. At the same time, it has achieved the goal of identifying the health risks of chronic disease patients in advance and conducting effective interventions, significantly improving the efficiency and effectiveness of chronic disease prevention and control, and providing innovative solutions for the field of chronic disease management. It is expected to improve the health status of a large number of chronic disease patients and high-risk groups and reduce the social medical burden.

[0093] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence-based chronic disease early warning intervention system, characterized in that: include: Data collection module: used to collect multimodal data, including genetic testing data, proteomics data, microbiome data, environmental data and socioeconomic data; Preprocessing module: used to preprocess the multimodal data to obtain basic data, wherein the preprocessing includes removing noise, outliers and missing values, and performing targeted data cleaning and feature extraction operations on data of different modalities; Conversion and fusion module: used to convert and fuse basic data to obtain fused data; Building training modules: used to build and train early warning models by integrating data; Evaluation and optimization module: used to evaluate and optimize the early warning model to obtain an early warning model for chronic diseases; Early warning and intervention module: used to provide early warning and intervention based on the chronic disease early warning model, including formulating intervention plans and sending early warning notifications and suggestions.

2. According to claim 1, an artificial intelligence-based chronic disease early warning intervention system is characterized by: The data acquisition module includes a gene detection acquisition unit, a proteomics acquisition unit, a microbiome acquisition unit, an environment acquisition unit, and a socioeconomic acquisition unit. The gene detection acquisition unit is used to collect whole genome sequencing data or gene chip data, and the sequencing depth of the whole genome sequencing data is not less than 30x. The proteomics acquisition unit is used to collect proteomics data, and the proteomics data includes the identification results and quantitative information of proteins obtained from the patient's biological samples through liquid chromatography-mass spectrometry technology. The microbiome acquisition unit is used to collect microbiome data, and the microbiome data includes microbial community composition and diversity information obtained by high-throughput sequencing analysis of the patient's feces, oral swab or skin swab samples. The environment acquisition unit is used to collect environmental data, and the environmental data includes air quality, water quality, noise, temperature and humidity. The socioeconomic acquisition unit is used to collect socioeconomic data, and the socioeconomic data includes age, gender, occupation, education level, income level and family population information.

3. The artificial intelligence-based chronic disease early warning intervention system according to claim 1, characterized in that: The preprocessing module includes a preprocessing unit, a communication unit, and a database unit. The preprocessing unit is used to perform quality assessment, data cleaning, data conversion and normalization operations on the collected multimodal data to obtain basic data. The communication unit is used for data transmission between units and information interaction between modules. The database unit is used to store multimodal data before and after preprocessing, providing data storage and retrieval services for subsequent data conversion, fusion and model construction, while ensuring the security and integrity of the data.

4. The artificial intelligence-based chronic disease early warning intervention system according to claim 1, characterized in that: The conversion and fusion module includes a data conversion unit and a data fusion unit. The data conversion unit is used to perform format conversion and feature reshaping on basic data. The data fusion unit is used to fuse the basic data after format conversion and feature reshaping at the feature level and the decision level. At the feature level, feature matrices of different modalities are weighted and spliced, and dimension reduction is performed through principal component analysis or linear discriminant analysis. At the decision level, the results of independent prediction models based on different modal data are fused to obtain fused data.

5. The artificial intelligence-based chronic disease early warning intervention system according to claim 1, characterized in that: The training module includes a fusion model unit, a hybrid model unit, a data set division unit, and a model training unit. The fusion model unit is used to construct an early warning model combining a convolutional neural network and a recurrent neural network according to the characteristics of the fusion data and the requirements of the chronic disease prediction task, wherein the convolution kernel size of the convolutional neural network is 3x3, the step size is 1, the activation function is ReLU, the recurrent neural network adopts a long short-term memory network or a gated recurrent unit, the number of hidden units is 128, and the number of layers is 2. The characteristics of the fusion data include the dimension of the data, the relationship between different characteristics, and the distribution of the data. The requirements of the chronic disease prediction task include binary classification prediction, multi-classification prediction, and chronic disease characteristics. The hybrid model unit is used to A decision tree model combining a multilayer perceptron and a decision tree is constructed based on the characteristics of the data and clinical experience, wherein the number of nodes in the input layer of the multilayer perceptron is determined according to the dimension of the fused feature vector, the hidden layer is 2-3 layers, the number of nodes is 256 and 128 respectively, the activation function is ReLU, the depth of the decision tree is 5-10 layers, the characteristics of the fused data include linear or nonlinear relationship, data volume, the data set partitioning unit is used to divide the fused data into a training set, a validation set and a test set in a ratio of 70%:15%:15%, the model training unit is used to train the model using stochastic gradient descent or Adam optimization algorithm, the training cycle is 100-200, and the corresponding loss function is selected according to the prediction task.

6. The artificial intelligence-based chronic disease early warning intervention system according to claim 5, characterized in that: When the model training unit performs binary classification prediction, the loss function used is: ,in is the number of training samples, is the true label of the sample, is the model prediction probability; when performing multi-classification prediction, the loss function used is ,in is the number of categories, For sample Belongs to category The real label, Predict samples for the model Belongs to category probability.

7. The artificial intelligence-based chronic disease early warning intervention system according to claim 1, characterized in that: The evaluation and optimization module includes a model evaluation unit and a model optimization unit. The model evaluation unit is used to calculate the indicators of the early warning model using the test set data and analyze the confusion matrix. The indicators include accuracy, recall rate, F1 value, and area under the receiver operating characteristic curve. The model optimization unit is used to optimize the early warning model to obtain an early warning model for chronic diseases. The optimization includes adding L1 or L2 regularization terms, Dropout technology or early stopping method when overfitting, increasing the number of hidden layers or neurons in the model, adjusting the activation function or using a more complex model when underfitting, and using Bayesian optimization, Hyperband optimization or expanding the search range of hyperparameters when optimizing hyperparameters.

8. The artificial intelligence-based chronic disease early warning intervention system according to claim 7, characterized in that: The model optimization unit, when using L1 regularization, adds ,in are the parameters of the model, is the regularization strength coefficient. When using L2 regularization, add .

9. The artificial intelligence-based chronic disease early warning intervention system according to claim 1, characterized in that: The early warning intervention module includes an early warning unit, an intervention plan generation unit, and a feedback interaction unit. The early warning unit is used to send early warning information based on the prediction results of the chronic disease early warning model when the predicted risk exceeds a preset threshold. The warning information includes the disease type, risk level and possible onset time. The intervention plan generation unit is used to formulate a personalized intervention plan for the patient based on the patient's individual data and prediction results, combined with clinical experience and medical guidelines. The intervention plan includes drug adjustment, diet plan formulation, exercise plan formulation and lifestyle change suggestions. The feedback interaction unit is used to receive feedback information from the patient on the intervention plan and perform human-computer interaction. The feedback information includes changes in health indicators and implementation status.

10. An artificial intelligence-based early warning intervention method for chronic diseases, characterized by: An artificial intelligence-based chronic disease early warning intervention system applied to any one of claims 1 to 9, comprising the following steps: S1. Data collection: The whole genome sequencing or gene chip data is obtained through the gene detection collection unit. The proteomics collection unit uses liquid chromatography-mass spectrometry to collect protein identification and quantitative information. The microbiome collection unit uses high-throughput sequencing of patient-specific samples to analyze microbial community data. The environmental collection unit collects air quality, water quality, noise, temperature and humidity data. The socioeconomic collection unit collects patient socioeconomic information and summarizes them into multimodal data. S2, data preprocessing: the preprocessing unit then performs quality assessment, noise outlier and missing value processing, specific mode cleaning, data conversion and normalization on the multimodal data to obtain basic data, the communication unit circulates the data and interacts with the modules, and the database unit stores the data; S3, data conversion and fusion: the data conversion unit then processes the format and features of the basic data, the data fusion unit performs weighted splicing and dimension reduction at the feature level, and fuses the results of independent models at the decision level to generate fused data; S4, model construction and training: the fusion model unit then builds a CNN-RNN early warning model based on the fusion data characteristics and chronic disease prediction tasks, the hybrid model unit builds an MLP-decision tree model based on the fusion data characteristics and clinical experience, and the data set division unit divides the fusion data into 70%:15%:15%; the model training unit uses SGD or Adam algorithm to select the loss function to train the model according to the prediction task; S5. Model evaluation and optimization: The model evaluation unit then uses the test set to calculate the model's indicators and analyze the confusion matrix. The model optimization unit uses corresponding strategies to optimize the model for overfitting, underfitting, and hyperparameter problems, and obtains an early warning model for chronic diseases. The corresponding strategies include L1 / L2 regularization, Dropout, early stopping method, adjusting the model structure, and Bayesian optimization. S6. Early warning intervention: The early warning unit then sends early warning information when the threshold is exceeded based on the results of the chronic disease early warning model. The intervention plan generation unit formulates an intervention plan. The feedback interaction unit receives patient feedback and human-computer interaction, adjusts the plan accordingly and stores data.

Images