A primary hypertension typing system and method based on multi-modal data fusion

Abstract

The application discloses a primary hypertension typing system and method based on multi-modal data fusion, and relates to the technical field of health management.The application traverses multi-modal hypertension diagnosis records through various traversal pointers, obtains multi-modal cause nodes from the multi-modal hypertension diagnosis records according to the traversal results, connects each multi-modal cause node and a disease root node to each other according to the corresponding relationship between the multi-modal cause nodes and the disease root node in the multi-modal hypertension diagnosis records, obtains a multi-stage disease diagnosis network, sets an association weight collection target between each multi-modal cause node in the multi-stage disease diagnosis network, collects real-time physical state data of a target patient in multiple time periods, matches the real-time physical state data with the multi-modal cause nodes contained in the multi-stage disease diagnosis network, matches the matched multi-modal cause nodes with each cause sequence chain, and accumulates risk values of each disease root node according to the matching results.

Description

Technical Field

[0001] This invention relates to the field of health management technology, specifically to a primary hypertension classification system and method based on multimodal data fusion. Background Technology

[0002] Essential hypertension, one of the most common cardiovascular diseases worldwide, has a complex pathogenesis and extremely high clinical heterogeneity. Current diagnosis primarily relies on blood pressure measurements, and treatment largely follows general guidelines for stepwise drug therapy. However, this "one-size-fits-all" management model faces significant challenges in practice: Firstly, the patient population with essential hypertension exhibits vast differences in etiology, pathophysiological characteristics, complication risks, and treatment responses. Blood pressure values ​​alone cannot reflect these intrinsic differences, leading to poor treatment outcomes or unnecessary side effects in some patients. Secondly, clinical practice lacks systematic and operational tools for refined etiological or phenotypic classification of hypertension patients. Physicians often rely on fragmented clinical indicators and experience for general judgments, which are highly subjective, difficult to standardize, and unable to dynamically integrate multi-dimensional health data (such as ambulatory blood pressure, heart rate variability, blood biochemistry, and vascular function) that accumulates over time.

[0003] While existing research has proposed several theories for hypertension classification (such as those based on hemodynamics, salt sensitivity, and neuroendocrine activation), these classification methods generally have the following limitations: First, the data dimensions are singular, usually based only on one type of examination result (such as blood tests or blood pressure monitoring), failing to effectively integrate multimodal and multi-time-point clinical data, resulting in a one-sided patient profile; second, the classification criteria are static, mostly based on fixed thresholds from expert consensus or small sample studies, unable to utilize massive amounts of real-world clinical data for dynamic learning and weight adjustment, resulting in poor adaptability; third, the lack of a structured and automated reasoning network from data to classification makes it difficult to apply the classification process quickly and reproducibly in clinical settings. Therefore, this paper proposes a primary hypertension classification system and method based on multimodal data fusion. Summary of the Invention

[0004] The purpose of this invention is to provide a primary hypertension classification system and method based on multimodal data fusion to address the shortcomings in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for classifying primary hypertension based on multimodal data fusion includes the following steps: Step S1: Obtain multimodal hypertension diagnosis records under different age groups, set traversal pointers, and then traverse the multimodal hypertension diagnosis records through various traversal pointers. Based on the traversal results, obtain multimodal etiology nodes from the multimodal hypertension diagnosis records, and set disease root nodes according to the disease types corresponding to the multimodal hypertension diagnosis records. Step S2: Based on the correspondence between each multimodal etiology node and the disease root node in the multimodal hypertension diagnosis record, connect each multimodal etiology node and the disease root node to obtain a multi-stage disease diagnosis network, and set the association weight between each multimodal etiology node in the multi-stage disease diagnosis network. Step S3: Collect real-time physical status data of the target patient at multiple time periods, match the real-time physical status data with the multimodal etiology nodes contained in the multi-stage disease diagnosis network, match the matched multimodal etiology nodes with each etiology sequence chain, and accumulate the risk value of each disease root node according to the matching results. Step S4: Preset risk thresholds and output a risk assessment report based on the relationship between the risk values ​​of each disease root node and the risk thresholds.

[0006] Furthermore, the process of acquiring multimodal hypertension diagnosis records across different age groups and setting up traversal pointers includes: We obtain multimodal hypertension diagnosis records covering different age groups such as young adults, middle-aged, and elderly. Each multimodal hypertension diagnosis record is marked with the corresponding primary hypertension subtype, which serves as the disease label for that multimodal hypertension diagnosis record. The system is configured to retrieve multimodal hypertension diagnosis records for different age groups, and to set up traversal pointers to iterate through the multimodal hypertension diagnosis records. Based on the traversal results, multimodal etiology nodes are retrieved from the multimodal hypertension diagnosis records, and disease root nodes are set according to the disease types corresponding to the multimodal hypertension diagnosis records.

[0007] Furthermore, the process of traversing the multimodal hypertension diagnosis records using various traversal pointers, obtaining the multimodal etiology nodes from the multimodal hypertension diagnosis records based on the traversal results, and setting the disease root node includes: The blood pressure traversal pointer, pulse traversal pointer, and blood viscosity traversal pointer are used to traverse the corresponding body data in all multimodal hypertension diagnosis records in turn. During the interval traversal, when a certain body data value in a multimodal hypertension diagnosis record falls into a data interval associated with any traversal pointer, the corresponding interval label is marked on the body data. Based on the interval traversal results, identify and record all combinations of blood pressure level interval labels, arterial elasticity index interval labels, and blood viscosity interval labels that appear in the multimodal hypertension diagnosis records and have significant distinguishing significance. Each unique combination, as long as it appears repeatedly in the record group and has a directional relationship with certain subtypes, is defined as an independent multimodal etiological node. The multimodal etiology node includes a range of body data values, which is a set of precise numerical boundaries of the interval labels obtained by the three traversal pointers mentioned above. Each multimodal etiology node contains a different range or type of body data values. That is, the range of body data values ​​is the same, but the types are necessarily different. If the range of body data values ​​is partially the same or completely different, then the corresponding types are the same or different. At the same time, based on the disease label accompanying each multimodal hypertension diagnosis record, a corresponding disease root node is set, with each disease root node corresponding to a hypertension type feature.

[0008] Furthermore, based on the correspondence between each multimodal etiology node and the disease root node in the multimodal hypertension diagnosis record, each multimodal etiology node and the disease root node are interconnected to obtain a multi-stage disease diagnosis network. The specific process includes: All disease root nodes and all multimodal etiology nodes are grouped into a network node set. This network is a heterogeneous network containing two levels: an etiology layer and a disease layer. For each multimodal hypertension diagnosis record, the order in which various abnormal physical data appear is traced along the entire timeline from early abnormal physiological indicators to final diagnosis. Based on the order of appearance, a disease root node corresponding to the record is directionally connected to a series of multimodal etiology nodes to form an etiology → disease propagation link. For any disease root node, analyze all multimodal hypertension diagnosis records pointing to the disease root node, extract the propagation path of the evolution of multimodal etiology nodes in each multimodal hypertension diagnosis record, and define it as etiology sequence chain; Each etiological sequence chain consists of multiple multimodal etiological nodes connected sequentially in chronological order, and a directed connection line is used to represent the progressive relationship between the nodes. Therefore, each disease root node is connected to multiple etiological sequence chains at the same time. The connection order of the multimodal etiological nodes contained in each etiological sequence chain must be partially different, which is used to indicate that there may be multiple different pathophysiological evolution paths for the same hypertension subtype.

[0009] Furthermore, the process of setting the association weights between various multimodal etiology nodes in a multi-stage disease diagnosis network includes: Within each etiology sequence chain, it is divided into multiple temporal levels according to the time window of the first appearance of the multimodal etiology nodes. The first temporal level contains one or more of the first multimodal etiology nodes, the second temporal level contains all the multimodal etiology nodes that appear in the next time window after the first level, and so on. The multimodal etiological nodes contained in the same temporal hierarchy are all different, and there are connecting lines between adjacent temporal hierarchy levels to represent the stage crossing of the pathophysiological process. Each etiological sequence chain has an associated weight on the connecting line.

[0010] Furthermore, the process of collecting real-time physical status data of the target patient over multiple time periods and matching this real-time physical status data with the multimodal etiological nodes included in the multi-stage disease diagnosis network includes: For a target patient with essential hypertension to be evaluated, develop a multi-time period data collection plan, with the collection period covering at least three time points: T0 at the first visit, T1 after 24 hours of dynamic monitoring, and T2 at the re-evaluation after lifestyle intervention or drug washout period. Each real-time body status data item collected must be completely consistent with the types of body data on which the multimodal etiology nodes are constructed; All real-time physical status data collected from the target patient during time period T0 are matched one by one with all multimodal etiology nodes in the multi-stage disease diagnosis network. The matching process includes: if a certain real-time physical status data value of the patient falls within the range of physical data values ​​defined by a certain multimodal etiology node, then the data is matched with that multimodal etiology node; otherwise, it is judged as not matching. The data for each time period (T0, T1, T2) are processed sequentially to obtain the set of matched multimodal etiological nodes for each time period.

[0011] Furthermore, the process of matching the multimodal etiology nodes with each etiology sequence chain and accumulating the risk values ​​of each disease root node based on the matching results includes: For all etiological sequence chains associated with a certain disease root node, check each one to see if it contains multimodal etiological nodes matched by the target patient at different time periods; When matching any etiology sequence chain, the temporal constraint must be satisfied: the set of nodes matched by the patient at time T0 must have an intersection with the multimodal etiology nodes in the first temporal level of the etiology sequence chain, and the temporal constraints of the multimodal etiology nodes in subsequent temporal levels are similar. If any of the patient's matched multimodal etiology node sets has no intersection with the corresponding temporal hierarchy, then the patient fails to match this etiology sequence chain. For any successfully matched etiology sequence chain, its cumulative association weight is calculated based on the association weight on the temporal hierarchy connection line where the matched multimodal etiology node is located and the proportion of the number of matches. For each disease root node, the cumulative association weights of the etiology sequence chains that were successfully matched by the target patient in all the etiology sequence chains associated with it are added together to obtain the risk value of each disease root node.

[0012] Furthermore, the process of setting risk thresholds and outputting a risk assessment report based on the relationship between the risk values ​​of each disease root node and the risk thresholds includes: For each type of hypertension, a first risk threshold and a second risk threshold are preset for the disease root node corresponding to the characteristics, wherein the first risk threshold is less than the second risk threshold; For any disease root node, compare its risk value with a preset risk threshold. Based on the comparison results, set a risk level for each disease root node, collect the risk values ​​and corresponding risk levels of all disease root nodes, and generate a risk assessment report.

[0013] A primary hypertension classification system based on multimodal data fusion includes a data collection module, an assessment network construction module, and a grading assessment module; The data collection module is used to acquire multimodal hypertension diagnosis records of different age groups, as well as collect real-time physical status data of target patients at multiple time periods; The evaluation network construction module is used to set traversal pointers, traverse multimodal hypertension diagnosis records through various traversal pointers, obtain multimodal etiology nodes from multimodal hypertension diagnosis records based on traversal results, set disease root nodes based on the disease types corresponding to multimodal hypertension diagnosis records, connect each multimodal etiology node and disease root node to each other based on the correspondence between each multimodal etiology node and disease root node in multimodal hypertension diagnosis records to obtain a multi-stage disease diagnosis network, and set the association weights between each multimodal etiology node in the multi-stage disease diagnosis network. The graded assessment module is used to match real-time physical status data with multimodal etiology nodes included in the multi-stage disease diagnosis network, match the matched multimodal etiology nodes with each etiology sequence chain, accumulate the risk value of each disease root node according to the matching results, preset the risk threshold, and output a risk assessment report based on the relationship between the risk value of each disease root node and the risk threshold.

[0014] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention extracts multimodal etiological nodes from multidimensional body data by setting three types of traversal pointers: blood pressure, pulse, and blood viscosity. To a certain extent, it overcomes the limitations of traditional hypertension classification, which relies solely on a single modal indicator or static threshold. It effectively reflects the complex pathophysiological mechanism of hypertension and improves the accuracy and scientific nature of hypertension classification. 2. This invention constructs a multi-stage disease diagnosis network that includes temporal hierarchy and correlation weights, incorporating the order and probability of various abnormal physical data into the analysis framework. It can dynamically trace the pathological evolution path of patients, achieving progress from static classification to dynamic process assessment, and helps to identify high-risk evolution directions early. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0016] Figure 1 This is a flowchart of the method of the present invention.

[0017] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 As shown, a method for classifying primary hypertension based on multimodal data fusion includes the following steps: Step S1: Obtain multimodal hypertension diagnosis records under different age groups, and set traversal pointers. Traverse the multimodal hypertension diagnosis records through various traversal pointers. Obtain multimodal etiology nodes from the multimodal hypertension diagnosis records based on the traversal results, and set disease root nodes according to the disease types corresponding to the multimodal hypertension diagnosis records. Step S2: Based on the correspondence between each multimodal etiology node and the disease root node in the multimodal hypertension diagnosis record, connect each multimodal etiology node and the disease root node to obtain a multi-stage disease diagnosis network, and set the association weight between each multimodal etiology node in the multi-stage disease diagnosis network. Step S3: Collect real-time physical status data of the target patient at multiple time periods, match the real-time physical status data with the multimodal etiology nodes contained in the multi-stage disease diagnosis network, match the matched multimodal etiology nodes with each etiology sequence chain, and accumulate the risk value of each disease root node according to the matching results. Step S4: Preset risk thresholds and output a risk assessment report based on the relationship between the risk values ​​of each disease root node and the risk thresholds.

[0020] Furthermore, step S1 is implemented through the following process: Step S101: Obtain multimodal hypertension diagnosis records for different age groups and set traversal pointers. The specific process includes: Obtain multimodal hypertension diagnosis records covering different age groups such as young adults, middle-aged, and elderly from hospital information systems, regional health information platforms, or specialized disease databases; The multimodal hypertension diagnosis record includes the patient's complete physical data file at the time of diagnosis, specifically covering blood pressure levels (such as office systolic blood pressure, diastolic blood pressure, 24-hour ambulatory blood pressure mean and diurnal rhythm), arterial elasticity indicators (such as carotid-femoral pulse wave velocity cfPWV, reflection wave enhancement index AIx), whole blood viscosity (measured values ​​at different shear rates, such as plasma viscosity), heart rate variability indicators (such as the standard deviation of all normal RR intervals SDNN, root mean square of the difference between adjacent RR intervals RMSSD), etc. Each multimodal hypertension diagnosis record is marked with the corresponding primary hypertension subtype, which serves as the disease label for that multimodal hypertension diagnosis record; Three data traversal pointers are established: blood pressure traversal pointer, pulse traversal pointer, and blood viscosity traversal pointer. Each traversal pointer is associated with a slightly different data range, which is used to extract qualitative features from the above multidimensional body data. The data range associated with the blood pressure traversal pointer includes the ideal blood pressure range, the multi-stage hypertension range (e.g., systolic blood pressure 140-159 mmHg or diastolic blood pressure 90-99 mmHg), and the isolated systolic hypertension range (systolic blood pressure ≥140 mmHg and diastolic blood pressure <90 mmHg): The data range associated with the pulse traversal pointer is divided based on the arterial elasticity index, including the normal arterial stiffness range, the mild hardening range (10m / s ≤ cfPWV < 14m / s), and the severe hardening range (cfPWV ≥ 14m / s). The data range associated with the blood viscosity traversal pointer includes the normal blood viscosity range, the low shear rate blood viscosity increase range, the medium shear rate blood viscosity increase range, the high shear rate blood viscosity increase range, and the plasma viscosity increase range.

[0021] Step S102: Traverse the multimodal hypertension diagnosis records using various traversal pointers, obtain the multimodal etiology nodes from the multimodal hypertension diagnosis records based on the traversal results, and set the disease root node according to the corresponding disease type of the multimodal hypertension diagnosis records. The specific process includes: The blood pressure traversal pointer, pulse traversal pointer, and blood viscosity traversal pointer are used to traverse the corresponding body data in all multimodal hypertension diagnosis records in turn. During the interval traversal, when a certain body data value in a multimodal hypertension diagnosis record falls into a data interval associated with any traversal pointer, the corresponding interval label is marked on the body data. Based on the interval traversal results, identify and record all combinations of blood pressure level interval labels, arterial elasticity index interval labels, and blood viscosity interval labels that appear in the multimodal hypertension diagnosis records and have significant distinguishing significance. Each unique combination, as long as it appears repeatedly in the record group and has a directional relationship with certain subtypes, is defined as an independent multimodal etiological node. The multimodal etiology node includes a range of body data values. The range of body data values ​​is a set of precise numerical boundaries of the interval labels obtained by the three traversal pointers mentioned above. For example, a multimodal etiology node can be defined as "Grade 1 hypertension interval", "severe sclerosis interval", and "low shear rate blood viscosity increase interval". Each multimodal etiology node contains a different range or type of body data values. That is, the range of body data values ​​is the same, but the types are necessarily different. If the range of body data values ​​is partially the same or completely different, then the corresponding types are the same or different. For example, "Grade 1 hypertension + severe sclerosis" and "Grade 1 hypertension + elevated low-shear blood viscosity" belong to different multimodal etiological nodes because they contain different combinations of indicators. However, if two records both meet the numerical range of "Grade 2 hypertension + mild sclerosis", then they correspond to the same multimodal etiological node. Simultaneously, based on the disease tags accompanying each multimodal hypertension diagnosis record, corresponding disease root nodes are set. Each disease root node corresponds to a hypertension type characteristic, such as refractory hypertension, salt-sensitive hypertension, renovascular hypertension, primary aldosteronism-related hypertension, sleep apnea syndrome-related hypertension, systolic hypertension in the elderly, obesity-related hypertension, etc.

[0022] Furthermore, step S2 is implemented through the following process: Step S201: Based on the correspondence between each multimodal etiology node and the disease root node in the multimodal hypertension diagnosis record, connect each multimodal etiology node and the disease root node to obtain a multi-stage disease diagnosis network. The specific process includes: All disease root nodes and all multimodal etiology nodes are grouped into a network node set. This network is a heterogeneous network containing two levels: an etiology layer and a disease layer. For each multimodal hypertension diagnosis record, the order in which various abnormal physical data appear is traced along the entire timeline from early abnormal physiological indicators to final diagnosis. Based on the order of appearance, a disease root node corresponding to the record is directionally connected to a series of multimodal etiology nodes to form an etiology → disease propagation link. For any disease root node, such as "salt-sensitive hypertension", analyze all multimodal hypertension diagnosis records pointing to the disease root node, extract the propagation path of the multimodal etiology node evolution in each multimodal hypertension diagnosis record, and define it as etiology sequence chain; Each etiology sequence chain consists of multiple multimodal etiology nodes connected sequentially in chronological order, and a directed connection line is used to represent the progressive relationship between nodes. Therefore, each disease root node is connected to multiple etiology sequence chains at the same time. The connection order of the multimodal etiology nodes contained in each etiology sequence chain must be partially different, which is used to represent that there may be multiple different pathophysiological evolution paths for the same hypertension subtype. For example, in salt-sensitive hypertension, one etiological sequence chain might start with the "high-normal blood pressure" node, progress to the "grade 1 hypertension" node, accompanied by the "increased low-shear blood viscosity" node, and then the "high plasma renin activity" node. Another etiological sequence chain might start directly with the "grade 1 hypertension" node, appearing together with the "hypernatremia" node.

[0023] Step S202: Set the association weights between various multimodal etiology nodes in the multi-stage disease diagnosis network. The specific process includes: Within each etiology sequence chain, it is divided into multiple temporal levels according to the time window of the first appearance of the multimodal etiology nodes. The first temporal level contains one or more of the first multimodal etiology nodes, the second temporal level contains all the multimodal etiology nodes that appear in the next time window after the first level, and so on. The multimodal etiological nodes contained in the same time series level are all different, and there are connecting lines between adjacent time series levels to represent the stage crossing of the pathophysiological process; It should be noted that in a multi-stage disease diagnosis network, each multimodal etiology node exists independently and uniquely. Therefore, a multimodal etiology node may exist simultaneously in different levels of different etiology sequence chains of the same disease root node. Each link in the etiological sequence chain has a correlation weight, which is obtained based on the statistical probability of occurrence of multimodal hypertension diagnosis records corresponding to the same disease root node. The specific calculation method is as follows: For the c-th etiology sequence chain pointing to the disease root node T, its initial association weight W c = m / num, where m represents the total number of multimodal hypertension diagnosis records that present the pathophysiological evolution pattern of the c-th etiological sequence chain associated with the disease root node T, and num represents the total number of all multimodal hypertension diagnosis records pointing to the disease root node T.

[0024] Furthermore, step S3 is implemented through the following process: Step S301: Collect real-time physical status data of the target patient over multiple time periods, and match the real-time physical status data with the multimodal etiology nodes included in the multi-stage disease diagnosis network. The specific process includes: For a target patient with essential hypertension to be evaluated, develop a multi-time period data collection plan, with the collection period covering at least three time points: T0 at the first visit, T1 after 24 hours of dynamic monitoring, and T2 at the re-evaluation after lifestyle intervention or drug washout period. Each real-time body status data item collected must be completely consistent with the types of body data on which the multimodal etiology node is constructed, specifically including: blood pressure level, arterial elasticity index, whole blood viscosity and plasma viscosity, heart rate variability index, serum electrolyte concentration and RAAS hormone level, and other body status data. All real-time physical status data collected from the target patient during time period T0 are matched one by one with all multimodal etiology nodes in the multi-stage disease diagnosis network. The matching process includes: if a certain real-time physical status data value of the patient falls within the range of physical data values ​​defined by a certain multimodal etiology node, then the data is matched with that multimodal etiology node; otherwise, it is judged as not matching. The data for each time period (T0, T1, T2) are processed sequentially to obtain the set of matched multimodal etiological nodes for each time period.

[0025] Step S302: Match the matched multimodal etiology nodes with each etiology sequence chain, and accumulate the risk value of each disease root node according to the matching results. The specific process includes: For all etiological sequence chains associated with a certain disease root node (such as salt-sensitive hypertension), check each one to see if it contains multimodal etiological nodes matched by the target patient at different time periods; When matching any etiology sequence chain, the temporal constraint must be satisfied: the set of nodes matched by the patient at time T0 must have an intersection with the multimodal etiology nodes in the first temporal level of the etiology sequence chain, and the temporal constraints of the multimodal etiology nodes in subsequent temporal levels are similar. If any of the patient's matched multimodal etiology node sets has no intersection with the corresponding temporal hierarchy, then the patient fails to match this etiology sequence chain. For any successfully matched etiology sequence chain, its cumulative association weight is calculated based on the association weight on the temporal hierarchy connection line where the matched multimodal etiology node is located and the proportion of the number of matches. The specific calculation formula is as follows: Obtain the temporal hierarchy L of the etiology sequence chain y, where the etiology sequence chain points to the disease root node t, and define the association weight w for the inter-layer connection line i (from temporal hierarchy i to i+1). i ; For the target patient P, the number of multimodal etiological nodes successfully matched at time level i is n. i The total number of nodes defined at time level i is M. i Then the proportion of matching quantity p at time sequence level i i = n i / M i ; The cumulative association weight V of the etiological sequence chain y with respect to patient P (y，P) The result is obtained by weighted summation of the contributions from all matched levels: V (y，P) = Σ L i (p) i × w i ), where the summation iterates through all levels i that patient P successfully matched; Repeat this process to calculate the cumulative association weight of patient P for each matching etiological sequence chain; For each disease root node T, the cumulative association weights of the disease sequence chains that were successfully matched with the target patient in all the disease sequence chains associated with it are added together to obtain the risk value of the disease root node T.

[0026] Furthermore, step S4 is implemented through the following process: A risk assessment report is generated based on a preset risk threshold and the relationship between the risk value of each disease root node and the risk threshold. The specific process includes: Based on epidemiological data, clinical experience, and diagnostic network test results, a team of clinical experts preset a first risk threshold and a second risk threshold for the root node of the disease corresponding to each type of hypertension. The first risk threshold is lower than the second risk threshold. For example, the first risk threshold is set at 0.8 for refractory hypertension and 0.6 for salt-sensitive hypertension. For any disease root node, compare its risk value with a preset risk threshold: If the risk value is greater than or equal to the second risk threshold, the target patient is determined to have a high probability of developing primary hypertension of the corresponding hypertension type and is marked as "high risk". If the risk value is less than or equal to the first risk threshold, the target patient is determined to have a low probability of developing primary hypertension of the corresponding hypertension type and is marked as "low risk". If the risk value is greater than the first risk threshold and less than the second risk threshold, the target patient is determined to have primary hypertension that has partially developed into the corresponding hypertension subtype and is marked as "medium risk". Collect risk values ​​and corresponding risk levels for all disease root nodes, and generate a risk assessment report, which includes: It lists summaries of the target patient's physical condition data over multiple time periods, visually displays dynamic trend graphs of key indicators such as blood pressure, arterial elasticity, and blood viscosity, and presents the risk values ​​and risk levels (high, medium, and low) of all disease root nodes in the form of tables or radar charts. For disease root nodes judged as "high risk" or "medium risk", it traces back and extracts the top 1-2 etiological sequence chains that contributed the largest cumulative association weight to the risk value.

[0027] Please see Figure 2 As shown, a primary hypertension classification system based on multimodal data fusion includes a data collection module, an assessment network construction module, and a tiered assessment module. The data collection module is used to acquire multimodal hypertension diagnosis records of different age groups, as well as collect real-time physical status data of target patients at multiple time periods; The evaluation network construction module is used to set traversal pointers, traverse multimodal hypertension diagnosis records through various traversal pointers, obtain multimodal etiology nodes from multimodal hypertension diagnosis records based on traversal results, set disease root nodes based on the disease types corresponding to multimodal hypertension diagnosis records, connect each multimodal etiology node and disease root node to each other based on the correspondence between each multimodal etiology node and disease root node in multimodal hypertension diagnosis records to obtain a multi-stage disease diagnosis network, and set the association weights between each multimodal etiology node in the multi-stage disease diagnosis network. The graded assessment module is used to match real-time physical status data with multimodal etiology nodes included in the multi-stage disease diagnosis network, match the matched multimodal etiology nodes with each etiology sequence chain, accumulate the risk value of each disease root node according to the matching results, preset the risk threshold, and output a risk assessment report based on the relationship between the risk value of each disease root node and the risk threshold.

[0028] The above description is merely a specific embodiment 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. A method for classifying primary hypertension based on multimodal data fusion, characterized in that, Includes the following steps: Step S1: Obtain multimodal hypertension diagnosis records under different age groups, set traversal pointers, and then traverse the multimodal hypertension diagnosis records through various traversal pointers. Based on the traversal results, obtain multimodal etiology nodes from the multimodal hypertension diagnosis records, and set disease root nodes according to the disease types corresponding to the multimodal hypertension diagnosis records. Step S2: Based on the correspondence between each multimodal etiology node and the disease root node in the multimodal hypertension diagnosis record, connect each multimodal etiology node and the disease root node to obtain a multi-stage disease diagnosis network, and set the association weight between each multimodal etiology node in the multi-stage disease diagnosis network. Step S3: Collect real-time physical status data of the target patient at multiple time periods, match the real-time physical status data with the multimodal etiology nodes contained in the multi-stage disease diagnosis network, match the matched multimodal etiology nodes with each etiology sequence chain, and accumulate the risk value of each disease root node according to the matching results. Step S4: Preset risk thresholds and output a risk assessment report based on the relationship between the risk values ​​of each disease root node and the risk thresholds.

2. The method for classifying primary hypertension based on multimodal data fusion according to claim 1, characterized in that, The process of acquiring multimodal hypertension diagnosis records across different age groups and setting up traversal pointers includes: We obtained multimodal hypertension diagnosis records for different age groups. Each multimodal hypertension diagnosis record was marked with the corresponding primary hypertension subtype, which served as the disease label for that multimodal hypertension diagnosis record. The system is configured to retrieve multimodal hypertension diagnosis records for different age groups, and to set up traversal pointers to iterate through the multimodal hypertension diagnosis records. Based on the traversal results, multimodal etiology nodes are retrieved from the multimodal hypertension diagnosis records, and disease root nodes are set according to the disease types corresponding to the multimodal hypertension diagnosis records.

3. The method for classifying primary hypertension based on multimodal data fusion according to claim 2, characterized in that, The process of traversing multimodal hypertension diagnosis records using various traversal pointers, obtaining multimodal etiology nodes from these records based on the traversal results, and setting disease root nodes includes: The blood pressure traversal pointer, pulse traversal pointer, and blood viscosity traversal pointer are used to traverse the corresponding body data in all multimodal hypertension diagnosis records in turn. During the interval traversal, when a certain body data value in a multimodal hypertension diagnosis record falls into a data interval associated with any traversal pointer, the corresponding interval label is marked on the body data. Based on the interval traversal results, identify and record all combinations of blood pressure level interval labels, arterial elasticity index interval labels, and blood viscosity interval labels that appear in the multimodal hypertension diagnosis records. Each unique combination, as long as it appears repeatedly in the record group and has a directional relationship with certain subtypes, is defined as an independent multimodal etiological node. The multimodal etiology node includes a range of body data values, which is a set of precise numerical boundaries of the interval labels obtained by the three traversal pointers mentioned above. Each multimodal etiology node contains a different range or type of body data values. At the same time, based on the disease label of each multimodal hypertension diagnosis record, a corresponding disease root node is set, and each disease root node corresponds to a hypertension type feature.

4. The method for classifying primary hypertension based on multimodal data fusion according to claim 3, characterized in that, Based on the correspondence between each multimodal etiology node and the disease root node in the multimodal hypertension diagnosis record, the multimodal etiology nodes and the disease root node are interconnected to obtain a multi-stage disease diagnosis network. The specific process includes: All disease root nodes and all multimodal etiology nodes are treated as a set of network nodes. This network is a heterogeneous network containing two levels: etiology layer and disease layer. For each multimodal hypertension diagnosis record, the order of occurrence of various abnormal physical data is traced along the entire timeline from early abnormal physiological indicators to final diagnosis. Based on the order of occurrence, a disease root node corresponding to the record is directionally connected to a series of multimodal etiology nodes to form a propagation link from etiology to disease. For any disease root node, analyze all multimodal hypertension diagnosis records pointing to the disease root node, extract the propagation path of the evolution of multimodal etiology nodes in each multimodal hypertension diagnosis record, and define it as etiology sequence chain; Each etiology sequence chain consists of multiple multimodal etiology nodes connected sequentially in chronological order, and a directed connection line represents the progressive relationship between the nodes. Therefore, each disease root node is connected to multiple etiology sequence chains at the same time, and the connection order of the multimodal etiology nodes contained in each etiology sequence chain must be partially different.

5. The method for classifying primary hypertension based on multimodal data fusion according to claim 4, characterized in that, The process of setting association weights between various multimodal etiology nodes in a multi-stage disease diagnosis network includes: Within each etiology sequence chain, it is divided into multiple temporal levels according to the time window of the first appearance of the multimodal etiology nodes. The first temporal level contains one or more of the first multimodal etiology nodes, the second temporal level contains all the multimodal etiology nodes that appear in the next time window after the first level, and so on. The multimodal etiological nodes contained in the same temporal hierarchy are all different, and there are connecting lines between adjacent temporal hierarchy levels to represent the stage crossing of the pathophysiological process. Each etiological sequence chain has an associated weight on the connecting line.

6. The method for classifying primary hypertension based on multimodal data fusion according to claim 5, characterized in that, The process of collecting real-time physical status data of target patients over multiple time periods and matching this data with multimodal etiological nodes included in a multi-stage disease diagnosis network includes: For a target patient with essential hypertension to be evaluated, develop a multi-time period data collection plan, with the collection period covering at least three time points: T0 at the first visit, T1 after 24 hours of dynamic monitoring, and T2 after re-evaluation following lifestyle intervention or drug washout. Each real-time body status data item collected must be completely consistent with the types of body data on which the multimodal etiology nodes are constructed; All real-time physical status data collected from the target patient during time period T0 are matched one by one with all multimodal etiology nodes in the multi-stage disease diagnosis network. The matching process includes: if a certain real-time physical status data value of the patient falls within the range of physical data values ​​defined by a certain multimodal etiology node, then the data is matched with that multimodal etiology node; otherwise, it is judged as not matching. The data for each time period (T0, T1, T2) are processed sequentially to obtain the set of matched multimodal etiological nodes for each time period.

7. The method for classifying primary hypertension based on multimodal data fusion according to claim 6, characterized in that, The process of matching multimodal etiology nodes with each etiology sequence chain and accumulating the risk values ​​of each disease root node based on the matching results includes: For all etiological sequence chains associated with a certain disease root node, check each one to see if it contains multimodal etiological nodes matched by the target patient at different time periods; When matching any etiology sequence chain, the temporal constraint must be satisfied: the set of nodes matched by the patient at time T0 must have an intersection with the multimodal etiology nodes in the first temporal level of the etiology sequence chain, and the temporal constraints of the multimodal etiology nodes in subsequent temporal levels are similar. If any of the patient's matched multimodal etiology node sets has no intersection with the corresponding temporal hierarchy, then the patient fails to match this etiology sequence chain. For any successfully matched etiology sequence chain, its cumulative association weight is calculated based on the association weight on the temporal hierarchy connection line where the matched multimodal etiology node is located and the proportion of the number of matches. For each disease root node, the cumulative association weights of the etiology sequence chains that were successfully matched by the target patient in all the etiology sequence chains associated with it are added together to obtain the risk value of each disease root node.

8. The method for classifying primary hypertension based on multimodal data fusion according to claim 7, characterized in that, The process of setting a risk threshold and generating a risk assessment report based on the relationship between the risk value of each disease root node and the risk threshold includes: For each type of hypertension, a first risk threshold and a second risk threshold are preset for the disease root node corresponding to the characteristics, wherein the first risk threshold is less than the second risk threshold; For any disease root node, compare its risk value with a preset risk threshold. Based on the comparison results, set a risk level for each disease root node, collect the risk values ​​and corresponding risk levels of all disease root nodes, and generate a risk assessment report.

9. A primary hypertension classification system based on multimodal data fusion, comprising the primary hypertension classification method based on multimodal data fusion according to any one of claims 1-8, characterized in that, It includes a data collection module, an evaluation network construction module, and a hierarchical evaluation module; The data collection module is used to acquire multimodal hypertension diagnosis records of different age groups, as well as collect real-time physical status data of target patients at multiple time periods; The evaluation network construction module is used to set traversal pointers, traverse multimodal hypertension diagnosis records through various traversal pointers, obtain multimodal etiology nodes from multimodal hypertension diagnosis records based on traversal results, set disease root nodes based on the disease types corresponding to multimodal hypertension diagnosis records, connect each multimodal etiology node and disease root node to each other based on the correspondence between each multimodal etiology node and disease root node in multimodal hypertension diagnosis records to obtain a multi-stage disease diagnosis network, and set the association weights between each multimodal etiology node in the multi-stage disease diagnosis network. The graded assessment module is used to match real-time physical status data with multimodal etiology nodes included in the multi-stage disease diagnosis network, match the matched multimodal etiology nodes with each etiology sequence chain, accumulate the risk value of each disease root node according to the matching results, preset the risk threshold, and output a risk assessment report based on the relationship between the risk value of each disease root node and the risk threshold.

Images