Hypertension prediction and diagnosis and treatment management system based on artificial intelligence and big data analysis

Abstract

The invention discloses a hypertension prediction and diagnosis and treatment management system based on artificial intelligence and big data analysis, and the system comprises a hierarchical diagnosis and treatment decision model module which constructs a patient grouping and hierarchical diagnosis and treatment decision model based on historical data; the big data diagnosis and treatment scheme module is used for matching the model and generating a big data diagnosis and treatment scheme; the machine learning diagnosis and treatment scheme module generates a machine learning diagnosis and treatment scheme through a machine learning model by using the time sequence data; the diagnosis and treatment scheme fusion module fuses the two schemes to generate a final diagnosis and treatment scheme; and the diagnosis and treatment suggestion output module outputs a complete diagnosis and treatment report containing decision basis. According to the method, abnormal output is effectively filtered by fusing a big data statistical rule and machine learning individual recognition, the accuracy and reliability of diagnosis and treatment suggestions are improved, and meanwhile, the interpretability of the scheme and the doctor-patient trust and adaptability are enhanced by providing a decision basis.

Description

Technical Field

[0001] This invention belongs to the field of health management technology, and in particular relates to a hypertension prediction and treatment management system based on artificial intelligence and big data analysis. Background Technology

[0002] Hypertension, a global chronic disease, has become a major risk factor for cardiovascular and cerebrovascular diseases, posing a serious challenge to global public health. Long-term management of hypertension involves multiple aspects, including medication, lifestyle modifications, and regular monitoring, requiring continuous and precise interaction and intervention between doctors and patients. Therefore, establishing an efficient and personalized hypertension prediction and management system is of great significance for improving cardiovascular and cerebrovascular health and enhancing patients' quality of life.

[0003] Currently, the common approach to hypertension management is based primarily on clinical guidelines and physician experience. Doctors develop corresponding treatment plans based on patients' blood pressure measurements, basic health conditions, and past medical history. With the development of artificial intelligence technology, some medical institutions have begun to introduce machine learning models to assist in hypertension diagnosis and treatment decisions, such as using deep learning algorithms to analyze patient data, predict blood pressure trends, or recommend treatment plans. While these methods have improved efficiency to some extent, they still have significant limitations.

[0004] While existing machine learning models can capture the temporal changes in individual patients' characteristics and provide personalized treatment recommendations, their excessive focus on short-term data fluctuations and lack of sufficient medical knowledge constraints can lead to aberrant results that deviate from clinical consensus. This phenomenon typically stems from the high sensitivity of deep learning algorithms to minute changes in input features and their tendency to overfit noisy data. Furthermore, mainstream deep learning models generally suffer from the "black box" problem, meaning they cannot provide transparent decision-making basis for their generated treatment recommendations. This is because the complex parameter interactions within neural networks make the reasoning process from input to output difficult to explain using medical logic. In practice, when an AI system recommends a treatment plan, doctors cannot know the medical logic behind the recommendation, making it difficult to fully understand the reasons behind the system's recommendation and thus unable to make effective professional judgments or necessary adjustments. For patients, the lack of understanding of the decision-making basis behind the recommendation leads to a lack of understanding and trust in the treatment plan, reducing patient cooperation and potentially affecting treatment outcomes. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a hypertension prediction and treatment management system based on artificial intelligence and big data analysis, thereby resolving the issues present in the prior art.

[0006] To achieve the above objectives, the present invention provides a hypertension prediction and treatment management system based on artificial intelligence and big data analysis, comprising:

[0007] The stratified diagnosis and treatment decision model module is used to group patients into several patient groups based on historical patient characteristic data, and to build a corresponding stratified diagnosis and treatment decision model based on the historical diagnosis and treatment records of each patient group.

[0008] The big data diagnosis and treatment solution module is used to match the corresponding hierarchical diagnosis and treatment decision model based on the characteristic data of the patient to be diagnosed, and generate a big data diagnosis and treatment solution based on the matched hierarchical diagnosis and treatment decision model.

[0009] The machine learning diagnosis and treatment plan module is used to generate a machine learning diagnosis and treatment plan based on multiple sets of time-series feature data of the patient to be diagnosed, using a preset machine learning model.

[0010] The diagnosis and treatment plan fusion module is used to fuse the big data diagnosis and treatment plan and the machine learning diagnosis and treatment plan to generate a final diagnosis and treatment plan.

[0011] The diagnosis and treatment suggestion output module is used to extract the corresponding decision basis from the hierarchical diagnosis and treatment decision model according to the final diagnosis and treatment plan, and generate a complete diagnosis and treatment report containing the decision basis and diagnosis and treatment suggestions.

[0012] Optionally, the hierarchical diagnosis and treatment decision model module is used to set weights for each feature data to construct a weighted space, construct a patient feature vector based on the feature data and the weighted space, cluster the patient feature vectors to obtain several clusters, each cluster as a patient group, and extract the center vector of each cluster; wherein, the feature data includes blood pressure data, heart rate data, physical data, exercise data, sleep data, dietary data, and lifestyle data.

[0013] Optionally, the stratified diagnosis and treatment decision model module constructs a corresponding stratified diagnosis and treatment decision model based on the diagnosis and treatment records of the patient group, including:

[0014] Obtain a set of diagnostic and treatment intervention categories, and create a corresponding set of specific keywords for each category;

[0015] Based on the specific keyword set created for each treatment intervention category, all historical treatment record texts within each patient group are read, each treatment record text is traversed, and a keyword matching algorithm is used to identify text fragments containing the corresponding category keywords;

[0016] Based on the identified text fragments, evidence-recommendation pairs are obtained; the evidence-recommendation pairs are organized according to the treatment intervention category, and a hierarchical structure is constructed for each patient group, wherein each layer corresponds to a treatment intervention category, and all evidence-recommendation pairs belonging to that category extracted from the patient group's historical treatment records are stored in each layer;

[0017] Statistically analyze the frequency of each basis or recommendation within its respective patient group, and calculate its frequency distribution based on the frequency.

[0018] Based on the frequency distribution, all the evidence-suggestion pairs stored in each layer are sorted in descending order to complete the construction of the hierarchical diagnosis and treatment decision model.

[0019] Optionally, the process of obtaining evidence-suggestion pairs based on the identified text fragments includes:

[0020] Natural language processing (NLP) techniques are used to analyze the grammatical structure of text fragments, identify sentence components representing state descriptions as decision basis candidates, and identify sentence components representing behavior instructions as suggestion candidates. Based on syntactic dependency relations, a correspondence is established between the decision basis candidates and the suggestion candidates. Medical text standardization processing is performed on the decision basis candidates and suggestion candidates to obtain standardized decision basis and standardized suggestions, and basis-suggestion pairs are formed based on the correspondence.

[0021] Optionally, the medical text standardization process includes:

[0022] A standard expression library is constructed, which stores standard terminology texts and their corresponding semantic vector representations; text vectorization technology is applied to the text to be standardized to generate a semantic vector to be standardized; the similarity between the semantic vector to be standardized and the semantic vector representations corresponding to each standard term in the standard expression library is calculated; and the text to be standardized is converted into the standard terminology text with the highest similarity.

[0023] Optionally, the big data diagnosis and treatment solution module is used to construct the feature vector of the patient to be diagnosed based on the weighted space; calculate the distance between the feature vector of the patient to be diagnosed and the central vector of each patient group; assign the patient to be diagnosed to the patient group with the smallest distance and obtain the hierarchical diagnosis and treatment decision model corresponding to the patient group; traverse each diagnosis and treatment intervention category layer of the obtained hierarchical diagnosis and treatment decision model; in each diagnosis and treatment intervention category layer, select the basis-suggestion pair with the highest frequency distribution and extract the standardized suggestions therein; combine the standardized suggestions extracted from each diagnosis and treatment intervention category layer to form the big data diagnosis and treatment solution.

[0024] Optionally, the preset machine learning model in the machine learning diagnosis and treatment solution module is trained through the following process:

[0025] Collect the patient's characteristic data from multiple consecutive historical data points and the corresponding medical records for the last characteristic data point;

[0026] Extracting feature vectors from consecutive feature data;

[0027] The recommended content is extracted from the medical records, and the extracted recommended content is processed by medical text standardization to obtain standardized recommendations. The standardized recommendations are then organized according to the categories of medical interventions.

[0028] The feature vector is used as the input sample, and the standardized recommendations under each corresponding diagnosis and treatment intervention category are used as the output label to train a deep learning model as the machine learning model.

[0029] The deep learning model's input layer receives feature vectors, its intermediate layer uses a long short-term memory network, and its output layer is divided into multiple sub-networks. Each sub-network corresponds to a diagnostic intervention category and is used to predict standardized recommendations under that category.

[0030] Optionally, the machine learning diagnosis and treatment plan module generates a machine learning diagnosis and treatment plan based on multiple sets of time-series feature data of the patient to be diagnosed, including:

[0031] Feature vectors are extracted from the multiple sets of time-series feature data; the feature vectors are input into the machine learning model; standardized suggestions under each diagnostic and treatment intervention category are obtained from the output of the machine learning model; and the standardized suggestions under each diagnostic and treatment intervention category are combined to form the machine learning diagnostic and treatment plan.

[0032] Optionally, the treatment plan fusion module is used to obtain standardized recommendations under each treatment intervention category in the big data treatment plan; obtain standardized recommendations under each treatment intervention category in the machine learning treatment plan; for each treatment intervention category, compare whether the standardized recommendations in the big data treatment plan and the machine learning treatment plan are consistent; if consistent, use the standardized recommendation as the recommendation for that category in the final treatment plan; if inconsistent, determine whether the standardized recommendation in the machine learning treatment plan exists in the corresponding category layer of the hierarchical treatment decision model of the matched patient group; if it exists, use the standardized recommendation of the machine learning treatment plan; if it does not exist, use the standardized recommendation of the big data treatment plan; combine the final selected recommendations of each treatment intervention category to form the final treatment plan.

[0033] Optionally, the treatment suggestion output module is used to obtain standardized suggestions under each treatment intervention category in the final treatment plan; in the hierarchical treatment decision model of the matched patient group, for each treatment intervention category, retrieve the basis-suggestion pair containing the corresponding standardized suggestions; and extract standardized decision basis from the retrieved basis-suggestion pair.

[0034] Compared with the prior art, the present invention has the following advantages and technical effects:

[0035] This invention constructs a hierarchical diagnosis and treatment decision-making model through big data analysis, enabling comprehensive capture of the diagnosis and treatment characteristics of different patient groups. Simultaneously, it extracts standardized evidence-recommendation pairs from historical medical records, establishing a mapping relationship between patient characteristics and clinical decisions. This big data-based analysis method allows the system to accurately group new patients and generate standardized diagnosis and treatment recommendations. By integrating big data-driven diagnosis and treatment plans with machine learning-based plans, the two approaches are compared, and the existence of machine learning recommendations is verified within the hierarchical diagnosis and treatment decision-making model. This fusion mechanism effectively filters out potential anomalous outputs from the machine learning model, preserving the advantage of machine learning in identifying individual characteristics while providing reliable medical practice references through big data statistical patterns, significantly improving the accuracy and reliability of diagnosis and treatment recommendations.

[0036] By acquiring standardized recommendations under each intervention category in the final treatment plan, and retrieving the basis-recommendation pairs containing the corresponding recommendations from the matched patient group stratified treatment decision-making model, the decision-making basis portion is extracted. This enables physicians to clearly understand the reasons behind the system's recommendations, facilitating professional judgment and necessary adjustments, significantly improving the acceptance of AI recommendations in the clinical setting. Simultaneously, it allows patients to understand the decision-making basis behind the recommendations, enhancing their understanding and trust in the treatment plan, increasing cooperation, and ultimately improving treatment outcomes. Attached Figure Description

[0037] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0038] Figure 1 This is a schematic diagram of the modules of the hypertension prediction and treatment management system based on artificial intelligence and big data analysis according to an embodiment of the present invention. Detailed Implementation

[0039] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0040] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0041] Example 1

[0042] like Figure 1 As shown, this embodiment provides a hypertension prediction and treatment management system based on artificial intelligence and big data analysis, including:

[0043] The stratified diagnosis and treatment decision model module is used to acquire historical patient characteristic data and medical records, group patients into several patient groups based on the characteristic data, and construct corresponding stratified diagnosis and treatment decision models based on the medical records of the patient groups.

[0044] Based on the characteristic data, patients were grouped into several patient groups, including:

[0045] Feature data includes blood pressure data, heart rate data, physical data, exercise data, sleep data, dietary data, and lifestyle data;

[0046] A weighted space is constructed by assigning weights to each feature data, and a patient feature vector is constructed based on the feature data and the weighted space.

[0047] A clustering algorithm is performed based on the patient's feature vector to obtain several clusters, with each cluster representing a patient group. The center vector of the corresponding cluster is also obtained.

[0048] The steps involved in constructing a stratified diagnosis and treatment decision model based on the medical records of patient groups include:

[0049] Obtain a set of treatment intervention categories, including medication management, diet management, exercise management, and rest management, and create a specific set of keywords for each treatment intervention category;

[0050] Read all historical medical records within patient groups with similar characteristics;

[0051] Iterate through each medical record text and use a keyword matching algorithm to identify text fragments containing keywords of each category;

[0052] For the identified text fragments, natural language processing techniques are applied to analyze the grammatical structure, identify sentence components that represent state descriptions as candidate decision-making criteria, identify sentence components that represent behavior instructions as candidate suggestions, and establish the correspondence between decision-making criteria and suggestions based on syntactic dependency relations.

[0053] Medical text standardization processing is applied to the decision-making basis and recommendations to obtain corresponding standardized decision-making basis and standardized recommendations, and basis-recommendation pairs are established according to the corresponding relationship.

[0054] Organize the basis-recommendation pairs according to the category of diagnosis and treatment intervention. Construct a hierarchical structure for each patient group. Each layer corresponds to a category of diagnosis and treatment intervention. Within each layer, store all basis-recommendation pairs extracted under that category.

[0055] Calculate the frequency distribution of each criterion / recommendation for its occurrence within the corresponding patient group;

[0056] The criteria and recommendations in each layer are arranged in descending order of frequency distribution to complete the construction of the stratified diagnosis and treatment decision model.

[0057] Medical text standardization includes:

[0058] Construct a standard expression library, where each standard expression stores its corresponding standard terminology text and its semantic vector representation;

[0059] For standardized text, text vectorization technology is applied to generate semantic vectors to be standardized;

[0060] Calculate the similarity between the semantic vector to be standardized and each standard expression vector in the standard expression library;

[0061] The text to be standardized is standardized into the standard terminology text with the highest similarity.

[0062] In this embodiment, the system can obtain historical characteristic data and medical records of hypertensive patients through the hospital's electronic medical record system, wearable device data platform, and patient follow-up system. The characteristic data collection covers multiple dimensions of physiological and lifestyle indicators, specifically including: blood pressure data (systolic blood pressure, diastolic blood pressure), heart rate data (resting heart rate, heart rate variability index), physical data (weight, BMI), exercise data (average daily steps, exercise duration), sleep data (sleep duration, sleep quality score), dietary data (sodium intake, water intake, protein intake), and lifestyle data (smoking status, frequency of alcohol consumption).

[0063] The system can group patients based on collected feature data. First, based on expert advice and relevant research literature, different weights are assigned to each feature indicator to reflect its importance in hypertension management. Specifically, blood pressure-related indicators (systolic blood pressure, diastolic blood pressure, and blood pressure variability) are given higher weights (0.3-0.4), followed by lifestyle indicators (0.2-0.25), physical indicators (0.15-0.2), and other physiological indicators (0.1-0.15). Weight assignment can also be achieved using machine learning methods, such as automatic determination based on feature importance assessment of hypertension treatment effectiveness. The system constructs a weighted space using these weights, converting each feature into a unified, standardized vector.

[0064] After constructing the weighted feature vectors of all patients, the system uses an improved K-means++ clustering algorithm to group patients. The improvements are mainly reflected in the initialization strategy for cluster centers and the automatic determination of the K value. For determining the K value, the system uses the silhouette coefficient method, trying multiple K values ​​(e.g., 3 to 10), and selecting the K value with the largest silhouette coefficient as the final number of clusters. After executing the clustering algorithm, the system obtains several patient groups, where patients within each group are similar in the feature space. Simultaneously, the system calculates the center vector of each cluster, which represents the overall characteristic of that patient group and is used for subsequent matching of new patients.

[0065] For each patient group, the system further constructs a stratified treatment decision-making model. First, it defines a set of treatment intervention categories, including four main aspects: medication management, diet management, exercise management, and lifestyle management. The system creates a specific set of keywords for each category. For example, the keyword set for the diet management category includes: "diet," "intake," "sodium," "fruits and vegetables," "protein," and "water intake." These keywords are determined based on medical literature, clinical guidelines, and expert knowledge, and can also be automatically extracted from a large number of medical records using text mining technology. The medication management, exercise management, and lifestyle management categories use a similar method to determine their respective keyword sets.

[0066] The system reads all historical medical records from each patient group, sourced from doctors' records, electronic medical records, follow-up records, etc. For each record, the system uses a keyword matching algorithm to identify text fragments containing keywords of various categories. Specifically, the system employs regular expression-based pattern matching and sliding window techniques to extract sentences or paragraphs containing keywords and their context.

[0067] The system employs natural language processing (NLP) techniques to deeply analyze the identified text fragments. Specifically, it uses a BERT-based text analysis model, fine-tuned with medical corpora to better understand the semantic structure of medical texts. The system uses dependency parsing to identify state descriptions (such as "persistently high blood pressure" and "excessive sodium intake") as candidate decision-making criteria and behavioral indicators (such as "suggest increasing medication frequency" and "reducing sodium intake") as candidate suggestions. By analyzing syntactic dependency relations (such as cause, effect, suggestion, and purpose relationships), the system establishes a correspondence between decision-making criteria and suggestions.

[0068] The extracted decision-making basis and recommendation texts often suffer from expression diversity issues. The system addresses this problem by applying medical text standardization processing. First, the system constructs a standard expression library covering common hypertension-related medical terms and their standard expressions. This library is built through medical expert annotation and large-scale medical literature analysis. Each standard expression stores its corresponding standard term text and a pre-computed semantic vector representation.

[0069] For text to be standardized, the system uses BERT-based text vectorization technology to generate its semantic vector. Then, it calculates the cosine similarity between this vector and each vector in the standard expression library, and standardizes the text to the standard expression with the highest similarity. For example, "eat too much salt" and "sodium intake is too high" may be standardized to the same standard expression "sodium intake exceeds the standard", while "eat less salt" and "reduce sodium intake" may be standardized to "limit sodium intake".

[0070] After standardization, the system establishes basis-recommendation pairs between standardized decision-making criteria and standardized recommendations based on the original correspondence. These basis-recommendation pairs are organized according to treatment intervention categories, constructing a hierarchical structure for each patient group. Each layer corresponds to a treatment intervention category, storing all basis-recommendation pairs extracted under that category.

[0071] The system further analyzes the frequency of each basis-recommendation pair within the corresponding patient group, calculating its frequency distribution. This step uses counting statistics to record the number of times each standardized basis-recommendation pair appears in all treatment records for that patient group, dividing this number by the total number of basis-recommendation pairs in that group to obtain the frequency distribution. Finally, the system sorts the basis-recommendation pairs within each stratum in descending order of frequency distribution. The resulting stratified treatment decision-making model reflects physicians' common treatment strategies and preferences for specific types of patient groups.

[0072] This decision-making model, based on patient grouping and stratification, can capture the characteristic differences of different patient groups and their corresponding personalized treatment strategies, providing a structured knowledge base for the generation of subsequent treatment plans. The basis-recommendation pairs stored in the model retain the original logic of clinical decision-making, ensuring that the generated treatment recommendations are both targeted and interpretable.

[0073] The big data diagnosis and treatment solution module is used to match the corresponding hierarchical diagnosis and treatment decision model based on the characteristic data of the patient to be diagnosed, and generate a big data diagnosis and treatment solution based on the matched hierarchical diagnosis and treatment decision model.

[0074] The steps involved in matching the characteristic data of the patient to be diagnosed with the corresponding stratified diagnosis and treatment decision model, and generating a big data diagnosis and treatment plan based on the matched stratified diagnosis and treatment decision model, include:

[0075] Obtain the feature data of the patient to be diagnosed, and construct the feature vector of the patient to be diagnosed based on the weighted space;

[0076] Calculate the distance between the feature vector of the patient to be diagnosed and the center vector of each patient group;

[0077] Patients to be diagnosed are assigned to the patient group with the smallest distance, and the stratified diagnosis and treatment decision model corresponding to that patient group is obtained;

[0078] Iterate through each treatment intervention category layer of the matched hierarchical diagnosis and treatment decision model;

[0079] Within each category of diagnostic and treatment interventions, the most frequently occurring basis-recommendation pairs are selected, and the recommendation portion is extracted.

[0080] The standardized recommendations extracted from each category of diagnosis and treatment intervention are combined to form a big data diagnosis and treatment plan.

[0081] The machine learning diagnosis and treatment plan module is used to input the most recent N sets of feature data of the patient to be diagnosed into a pre-trained machine learning model to obtain a machine learning diagnosis and treatment plan, where N is a preset value and N is greater than or equal to 2;

[0082] The training process for a machine learning model includes:

[0083] Collect characteristic data of a patient from N consecutive historical data points, as well as the medical records corresponding to the last characteristic data point;

[0084] Feature vectors are extracted from N consecutive feature data sets; text content is obtained from the last medical record; suggestions are extracted from the text content; the extracted suggestions are processed by medical text standardization to obtain standardized suggestions; and the standardized suggestions are organized according to the category of medical intervention.

[0085] The feature vectors extracted from N consecutive feature data will be used as input samples, and the standardized recommendations under each corresponding diagnosis and treatment intervention category will be used as output labels.

[0086] A deep learning model is constructed, in which the input layer receives feature vectors, the intermediate layer uses a long short-term memory network, and the output layer is divided into multiple sub-networks. Each sub-network corresponds to a diagnosis and treatment intervention category and is used to predict standardized recommendations under that category.

[0087] The prediction accuracy of comprehensive diagnosis and treatment intervention categories is used to establish a loss function, and the model parameters are trained using an optimization algorithm with the goal of minimizing the loss function;

[0088] The model performance is evaluated on the validation set. When the performance reaches a preset threshold or there is no improvement after multiple rounds of training, the training of the machine learning model is completed. The machine learning model is used to predict standardized recommendations under each diagnostic and treatment intervention category based on N consecutive feature data, and generate machine learning diagnostic and treatment plans.

[0089] In this embodiment, the machine learning diagnosis and treatment plan module is responsible for inputting the most recent N sets of feature data of the patient to be diagnosed into a pre-trained machine learning model to generate a machine learning diagnosis and treatment plan. The system first needs to train the machine learning model to capture the correlation between the temporal changes of the patient's feature data and the diagnosis and treatment recommendations.

[0090] The model training process begins with data collection. The system selects hypertensive patients with complete follow-up records from the electronic medical record database, and obtains the feature data of each patient for N consecutive times (N is set to 4 in this embodiment) and the medical record of the last follow-up visit.

[0091] Extracting feature vectors from N consecutive measurements is a crucial step in model training. The system not only extracts the original feature values ​​from each measurement but also calculates the trend characteristics of each indicator. For example, for blood pressure data, in addition to extracting the systolic and diastolic blood pressure values ​​from each measurement, it also calculates derived features such as the first-order difference (the change between two adjacent measurements), rate of change, fluctuation amplitude, and trend slope. These temporal features reflect the dynamic changes in the patient's condition, providing important evidence for predicting treatment recommendations. For all N measurements, the system organizes them in chronological order to form a complete feature vector containing both original and derived features.

[0092] Simultaneously, the system processes the medical record text from the last follow-up visit, extracting the doctor's treatment recommendations. This process employs a method similar to that in the hierarchical treatment decision-making model module, using natural language processing technology to identify the recommendations within the text. The identified recommendations undergo medical text standardization processing, transforming them into standardized recommendation expressions. The system then organizes these standardized recommendations according to treatment intervention categories (medication management, diet management, exercise management, and rest management), forming output labels for the training data.

[0093] After data preparation, the system constructs a deep learning model. This model employs a multi-input, multi-output architecture, specifically: the input layer receives the aforementioned feature vectors; the intermediate layer uses a Long Short-Term Memory (LSTM) network to process time-series data; and the output layer is divided into multiple sub-networks, each corresponding to a diagnostic intervention category. The LSTM layer design is particularly suitable for capturing the long-term dependencies and short-term fluctuations of physiological indicators such as blood pressure, and the model can effectively learn the complex correlation between changes in patient condition and physician recommendations.

[0094] The model employs an LSTM architecture suitable for time-series data processing, supplemented by an attention mechanism to enhance its ability to identify key time points and features. The output layer features multiple sub-networks designed for different diagnostic and treatment intervention categories, using appropriate activation functions to generate proposed probability distributions for each category. The system incorporates a comprehensive loss function, assigning different weights to categories of varying clinical importance, and is trained using suitable optimizers and regularization techniques.

[0095] Training data is divided into training and validation sets proportionally. During training, the system continuously evaluates the model's performance metrics on the validation set, including accuracy, precision, and recall. For clinically sensitive categories such as medication, the system pays particular attention to the false positive rate. The training process ends when the model's performance reaches a preset threshold or when there is no significant improvement after multiple training runs.

[0096] The trained machine learning model is deployed into the machine learning-based treatment plan module. When a new patient enters the system, the model uses the patient's most recent N (4 in this example) feature data as input to predict standardized recommendations for each treatment intervention category, forming a complete machine learning-based treatment plan. These recommendations reflect personalized treatment strategies based on dynamic changes in the patient's condition, capturing treatment opportunities and adjustment needs that are difficult to identify using traditional methods.

[0097] Unlike stratified treatment decision-making models, which rely on statistical analysis of similar patient groups, machine learning models focus on the trajectory of individual patient status changes, providing more dynamic and forward-looking treatment recommendations. The combination of these two approaches will complement each other in the subsequent fusion module, further improving the accuracy and personalization of treatment plans.

[0098] The treatment plan fusion module is used to integrate big data treatment plans and machine learning treatment plans to generate the final treatment plan.

[0099] The process of integrating big data and machine learning approaches to generate a final treatment plan includes:

[0100] Obtain standardized recommendations for each treatment intervention category within the big data-driven treatment plan;

[0101] Obtain standardized recommendations for each intervention category within the machine learning-based diagnostic and treatment plan;

[0102] For each category of diagnostic and treatment intervention, compare whether the standardized recommendations in big data diagnostic and treatment protocols and machine learning diagnostic and treatment protocols are consistent;

[0103] When there is agreement, the standardized recommendation shall be directly adopted as the standardized recommendation for that category in the final treatment plan;

[0104] When inconsistencies are found, check whether the standardized recommendations of the machine learning approach exist in the corresponding category layer of the matched patient group stratified diagnosis and treatment decision model;

[0105] If it exists, the standardized recommendations for machine learning solutions will be adopted; if it does not exist, the standardized recommendations for big data solutions will be adopted.

[0106] The final standardized recommendations for each diagnostic and treatment intervention category are combined to form the final treatment plan.

[0107] In this embodiment, the system first obtains suggestions generated by the big data treatment plan module and the machine learning treatment plan module, respectively. The big data treatment plan comes from statistical analysis of the historical treatment records of patient groups with similar characteristics, and includes standardized suggestions under each treatment intervention category (medication management, diet management, exercise management, and rest management). The machine learning treatment plan comes from time-series analysis based on the patient's most recent N feature data, and also organizes standardized suggestions according to treatment intervention categories.

[0108] For each intervention category, the system first compares the consistency of standardized recommendations in the two protocols. When the recommendations from both protocols are consistent, it indicates that the static population analysis and dynamic individual analysis have reached a consensus, and the system directly adopts this recommendation as the standardized recommendation for that category in the final protocol. For example, if both protocols recommend "increasing daily fruit and vegetable intake," this recommendation will be directly incorporated into the final protocol.

[0109] When two approaches offer different recommendations, the system initiates a discrepancy assessment mechanism. This situation reflects a potential contradiction between population statistical patterns and individual dynamic characteristics, requiring further determination of which recommendation is more suitable for the current patient. The system checks whether the standardized recommendations of the machine learning approach exist in the corresponding category layer of the stratified treatment decision-making model for the matched patient group. This check essentially assesses whether the "non-mainstream" recommendations proposed by the machine learning model still have a certain clinical basis.

[0110] The tiered diagnosis and treatment decision-making model includes all suggestions ranked by frequency, encompassing not only the most frequent mainstream suggestions but also the less frequent suboptimal suggestions. If a machine learning suggestion exists in the decision-making model (even if its frequency is low), it indicates that while the suggestion may not be the most common, it has practical applications for specific patients in clinical practice. In this case, the system prioritizes the machine learning suggestion. The theoretical basis for this decision-making mechanism lies in the fact that machine learning models, by analyzing the temporal changes in patients' characteristics, may capture individualized features and clinical needs that static feature clustering cannot fully reflect.

[0111] For example, in medication management, a big data approach might recommend a "standard dose of calcium channel blocker" based on the patient's overall characteristics, while a machine learning approach might suggest "adding a low dose of diuretics" based on the patient's recent blood pressure fluctuations and heart rate change patterns. If the stratified treatment decision model finds that "adding a low dose of diuretics" does indeed exist in the medication recommendations for the patient's group (albeit with low frequency), the system will adopt the machine learning suggestion, thus achieving a precise response to the patient's blood pressure fluctuation characteristics.

[0112] Conversely, if the machine learning approach's recommendations are completely absent from the stratified diagnosis and treatment decision-making model, it means that the recommendations lack sufficient clinical support, potentially stemming from model overfitting or misinterpretation of certain noisy features. In this case, the system adopts a more conservative big data approach to ensure the safety and evidence-based foundation of the treatment recommendations.

[0113] This fusion strategy leverages the complementary advantages of the two models: the big data-based treatment plan module provides robustness based on population statistics and clinical practice consensus, ensuring the medical standardization of recommendations; while the machine learning-based treatment plan module provides accuracy based on individual dynamic characteristics, enhancing the personalization of treatment plans. Simultaneously, by using a hierarchical treatment decision-making model as a reference standard, the system establishes an effective verification mechanism, avoiding the risk of unfounded innovation that might arise from machine learning models.

[0114] After completing the fusion decision for all diagnostic and treatment intervention categories, the system combines the final selected standardized recommendations under each category to form a complete final treatment plan.

[0115] The treatment recommendation output module is used to generate and output a complete treatment report containing the decision basis and treatment recommendations based on the final treatment plan and the hierarchical treatment decision model.

[0116] The decision-making criteria matched according to the final treatment plan and the tiered treatment decision-making model include:

[0117] Obtain standardized recommendations for each intervention category in the final treatment plan;

[0118] In the matched patient group stratified diagnosis and treatment decision model, for each diagnosis and treatment intervention category, a basis-recommendation pair containing the corresponding standardized recommendations is retrieved;

[0119] Extract standardized decision-making criteria from the retrieved criteria-recommendation pairs.

[0120] In this embodiment, the treatment suggestion output module combines the final treatment plan generated by fusion with its corresponding decision-making basis to generate a complete treatment report.

[0121] The system obtains standardized recommendations under each treatment intervention category in the final treatment plan, and then searches within the matched patient group stratified treatment decision model. For each standardized recommendation in the treatment intervention category, the system quickly locates the basis-recommendation pair containing that recommendation and extracts the standardized decision basis portion from it.

[0122] For doctors, these treatment reports with supporting decision-making criteria provide the medical logic behind the recommendations, enabling them to more comprehensively understand the reasons for the system's recommendations and facilitate professional judgment and necessary adjustments. For patients, understanding the decision-making criteria behind the recommendations significantly enhances their understanding and trust in the treatment plan, thereby increasing cooperation, which ultimately leads to improved treatment outcomes.

[0123] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only one method, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0124] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An artificial intelligence and big data analysis-based hypertension prediction and diagnosis and treatment management system, characterized in that, Comprise: A hierarchical diagnosis and treatment decision model module for grouping patients according to historical patient characteristic data to obtain a plurality of patient groups, and constructing a corresponding hierarchical diagnosis and treatment decision model based on the historical diagnosis and treatment records of each patient group; A big data diagnosis and treatment scheme module for matching a corresponding hierarchical diagnosis and treatment decision model according to the characteristic data of a patient to be diagnosed, and generating a big data diagnosis and treatment scheme based on the matched hierarchical diagnosis and treatment decision model; A machine learning diagnosis and treatment scheme module for generating a machine learning diagnosis and treatment scheme through a pre-set machine learning model according to a plurality of sets of time series characteristic data of the patient to be diagnosed; A diagnosis and treatment scheme fusion module for fusing the big data diagnosis and treatment scheme and the machine learning diagnosis and treatment scheme to generate a final diagnosis and treatment scheme; A diagnosis and treatment suggestion output module for extracting a corresponding decision basis from the hierarchical diagnosis and treatment decision model according to the final diagnosis and treatment scheme, and generating a complete diagnosis and treatment report containing the decision basis and the diagnosis and treatment suggestion.

2. The hypertension prediction and diagnosis and treatment management system based on artificial intelligence and big data analysis according to claim 1, wherein the hierarchical diagnosis and treatment decision model module is configured to set a weight for each characteristic data to construct a weighted space, construct a patient characteristic vector according to the characteristic data and the weighted space, cluster the patient characteristic vector to obtain a plurality of clusters, each cluster being a patient group, and extract a center vector of each cluster; wherein the characteristic data comprises blood pressure data, heart rate data, physical data, exercise data, sleep data, diet data and lifestyle data.

3. The hypertension prediction and diagnosis and treatment management system based on artificial intelligence and big data analysis according to claim 2, wherein the hierarchical diagnosis and treatment decision model module constructs a corresponding hierarchical diagnosis and treatment decision model according to the diagnosis and treatment records of the patient group, comprising: Obtaining a set of diagnosis and treatment intervention categories, and creating a corresponding specific keyword set for each diagnosis and treatment intervention category; Based on the specific keyword set created for each diagnosis and treatment intervention category, reading all historical diagnosis and treatment record texts in each patient group, traversing each diagnosis and treatment record text and using a keyword matching algorithm to identify text segments containing corresponding category keywords; Based on the identified text segments, obtaining basis-suggestion pairs; organizing the basis-suggestion pairs according to the diagnosis and treatment intervention categories, and constructing a hierarchical structure for each patient group, wherein each layer corresponds to a diagnosis and treatment intervention category, and all basis-suggestion pairs belonging to the category extracted from the historical diagnosis and treatment records of the patient group are stored in each layer; Statistically counting the frequency of each basis-suggestion pair in its corresponding patient group, and calculating its frequency distribution based on the frequency; According to the frequency distribution, arranging all basis-suggestion pairs stored in each layer in descending order to complete the construction of the hierarchical diagnosis and treatment decision model.

4. The hypertension prediction and diagnosis and treatment management system based on artificial intelligence and big data analysis according to claim 3, wherein the process of obtaining basis-suggestion pairs based on the identified text segments comprises: ​ ​ ​ The natural language processing technology is used to analyze the syntax structure of the text segment, identify the sentence components representing the state description as decision basis candidates, identify the sentence components representing the behavior instruction as suggestion candidates, establish a corresponding relationship between the decision basis candidates and the suggestion candidates according to the syntactic dependency relationship, perform medical text standardization processing on the decision basis candidates and the suggestion candidates to obtain standardized decision basis and standardized suggestions, and form a basis-suggestion pair based on the corresponding relationship.

5. The hypertension prediction and diagnosis and treatment management system based on artificial intelligence and big data analysis according to claim 4, characterized in that, The medical text standardization processing includes: A standard expression library is constructed, the standard expression library stores standard term texts and corresponding semantic vector representations, a text vectorization technology is applied to the text to be standardized to generate a standardized semantic vector, the similarity between the standardized semantic vector and the semantic vector representation corresponding to each standard term in the standard expression library is calculated, and the text to be standardized is converted into the standard term text with the highest similarity.

6. The hypertension prediction and diagnosis and treatment management system based on artificial intelligence and big data analysis according to claim 3, characterized in that, The big data diagnosis and treatment scheme module is configured to construct a feature vector of the patient to be diagnosed according to the weighted space, calculate the distance between the feature vector of the patient to be diagnosed and the center vector of each patient group, assign the patient to be diagnosed to the patient group with the smallest distance, and obtain the hierarchical diagnosis and treatment decision model corresponding to the patient group, traverse each diagnosis and treatment intervention category layer of the obtained hierarchical diagnosis and treatment decision model, select the basis-suggestion pair with the highest frequency distribution in each diagnosis and treatment intervention category layer, and extract the standardized suggestions therefrom; The standardized suggestions extracted from each diagnosis and treatment intervention category layer are combined to form the big data diagnosis and treatment scheme.

7. The hypertension prediction and diagnosis and treatment management system based on artificial intelligence and big data analysis according to claim 5, characterized in that, The machine learning model preset in the machine learning diagnosis and treatment scheme module is obtained by training the following process: Collect the diagnosis and treatment records corresponding to the feature data of the last time and the feature data of the patient continuously for multiple times in history; Extract the feature vector from the continuous multiple times of feature data; Extract the suggestion content from the diagnosis and treatment records, perform medical text standardization processing on the extracted suggestion content to obtain standardized suggestions, and organize the standardized suggestions according to the diagnosis and treatment intervention category; The feature vector is used as an input sample, and the standardized suggestions corresponding to each diagnosis and treatment intervention category are used as output labels to train a deep learning model as the machine learning model; The input layer of the deep learning model receives the feature vector, the middle layer adopts a long short-term memory network, and the output layer is divided into multiple sub-networks, each of which corresponds to a diagnosis and treatment intervention category and is used to predict the standardized suggestions in the category.

8. The hypertension prediction and diagnosis and treatment management system based on artificial intelligence and big data analysis according to claim 7, characterized in that, The machine learning diagnosis and treatment scheme module generates a machine learning diagnosis and treatment scheme according to the plurality of sets of time sequence feature data of the patient to be diagnosed, comprising: extracting a feature vector from the plurality of sets of time sequence feature data; inputting the feature vector into the machine learning model; obtaining standardized suggestions under each diagnosis and treatment intervention category output by the machine learning model; and combining the standardized suggestions under each diagnosis and treatment intervention category to form the machine learning diagnosis and treatment scheme.

9. The artificial intelligence and big data analysis-based hypertension prediction and treatment management system according to claim 1, characterized in that: The diagnosis and treatment scheme fusion module is configured to obtain standardized suggestions under each diagnosis and treatment intervention category in the big data diagnosis and treatment scheme; and obtain standardized suggestions under each diagnosis and treatment intervention category in the machine learning diagnosis and treatment scheme. For each diagnosis and treatment intervention category, the consistency of the standardized suggestions in the big data diagnosis and treatment scheme and the machine learning diagnosis and treatment scheme is compared; if consistent, the standardized suggestion is adopted as the suggestion of the category in the final diagnosis and treatment scheme. If inconsistent, it is judged whether the standardized suggestion in the machine learning diagnosis and treatment scheme exists in the corresponding category layer of the stratified diagnosis and treatment decision model of the matched patient group; if existing, the standardized suggestion of the machine learning diagnosis and treatment scheme is adopted. If not existing, the standardized suggestion of the big data diagnosis and treatment scheme is adopted. The final selected suggestions of each diagnosis and treatment intervention category are combined to form the final diagnosis and treatment scheme.

10. The artificial intelligence and big data analysis-based hypertension prediction and treatment management system according to claim 9, characterized in that: The diagnosis and treatment suggestion output module is configured to obtain standardized suggestions under each diagnosis and treatment intervention category in the final diagnosis and treatment scheme; in the stratified diagnosis and treatment decision model of the matched patient group, for each diagnosis and treatment intervention category, retrieve the basis-suggestion pair containing the corresponding standardized suggestion; and extract the standardized decision basis from the retrieved basis-suggestion pair.

Images