A health management method and system based on physical examination reports

Abstract

This invention discloses a health management method and system based on physical examination reports, belonging to the field of medical and health information technology. It includes acquiring multimodal physical examination data (blood, imaging, vital signs), inputting it into a multimodal health field construction engine, and creating associated virtual data layers according to data types. Each data layer undergoes deep processing to calculate the comprehensive risk field intensity of biochemical indicators, generate a three-dimensional spatial probability distribution map of imaging features, and a vital sign offset vector. These quantitative features are input into a dynamic health status prediction model to simulate the coupled evolution of multiple factors, generating multiple health status evolution trajectories. Finally, key evolution nodes are extracted from the trajectories, and personalized node intervention instruction sets are generated based on the specific state of the virtual field at each node. This method achieves closed-loop management of individual health status from multi-dimensional unified representation and dynamic process prediction to precise time-point intervention.

Description

Technical Field

[0001] This invention belongs to the field of medical and health information technology, specifically a health management method and system based on physical examination reports. Background Technology

[0002] Personal health management relies heavily on medical examination reports, and current technologies primarily employ two methods: independent comparison of indicators and static comprehensive assessment. Independent comparison compares blood biochemistry indicators, imaging examination descriptions, and physical measurement data against preset standard ranges, providing independent recommendations for individual results exceeding the threshold. Static comprehensive assessment, on the other hand, uses preset weights to weight multiple indicators, outputting a single health score or risk level. These methods treat data from different sources and of different types as discrete parameters, and their analysis is based on the assumption of data independence, failing to effectively reveal and utilize the inherent correlations and synergistic effects among multimodal data such as blood, imaging, and vital signs.

[0003] The conclusions drawn from existing health management technologies are essentially static judgments based on a single snapshot of physical examination data. Such methods only reflect a cross-section of the examinee's health status at a specific moment; the output, whether anomaly alerts or comprehensive scores, is a description of the condition at that particular point in time. Due to the lack of modeling for the ongoing interactions between various health indicators and the inability to simulate the dynamic coupling and evolution of these indicators over time, existing technologies struggle to predict potential future health paths and cannot identify specific points of critical intervention value within the potential disease progression chain. This results in existing health recommendations being largely generalized and lagging, making it difficult to achieve precise, proactive intervention targeting the dynamic changes in an individual's health status. Summary of the Invention

[0004] This invention aims to solve at least one of the technical problems existing in the prior art;

[0005] Therefore, this invention proposes a health management method based on physical examination reports, comprising:

[0006] Obtain raw physical examination data that includes multimodal physical examination data;

[0007] The original physical examination data is input into the multimodal health field construction engine, and based on the data type and correlation, a biochemical indicator data layer, an image feature data layer, and a vital sign mapping data layer are created in the virtual health field.

[0008] Anomaly correlation analysis is performed on multiple blood biochemical index values ​​in the biochemical index data layer to calculate the comprehensive risk field intensity. Structured information is extracted from the image examination description text and mapped to the image feature data layer to generate a three-dimensional spatial probability distribution map. Physical measurement data is mapped to the vital sign mapping data layer to generate a vital sign offset vector.

[0009] The comprehensive risk field intensity, three-dimensional spatial probability distribution map, and vital sign offset vector are input into the dynamic health status inference model to simulate the coupled evolution path and generate a health status evolution trajectory.

[0010] Key evolution nodes are extracted from the trajectory of health status evolution. For each key evolution node, a personalized set of node intervention instructions is generated by combining the specific status of the biochemical indicator data layer, image feature data layer and vital sign mapping data layer at that time point.

[0011] Furthermore, the step of performing anomaly correlation analysis on multiple blood biochemical index values ​​in the biochemical index data layer to calculate the comprehensive risk field intensity, extracting structured information from the image examination description text and mapping it to the image feature data layer to generate a three-dimensional spatial probability distribution map, and mapping physical measurement data to the vital sign mapping data layer to generate a vital sign offset vector, specifically includes:

[0012] Anomaly correlation analysis is performed on multiple blood biochemical index values ​​in the biochemical index data layer to identify index groups that deviate from the normal reference range. Based on the direction and magnitude of deviation between each index group, the comprehensive risk field intensity of the index group is calculated.

[0013] Structured information is extracted from the image examination description text to identify key anatomical structures, abnormal descriptive words, and quantitative modifiers. The extracted information is then mapped to the image feature data layer to generate a three-dimensional spatial probability distribution map.

[0014] Anthropometric data is mapped to a vital sign mapping data layer. Based on the difference between the measurement data and the standard body type, a vital sign offset vector representing the physiological structure offset is generated in the vital sign mapping data layer.

[0015] The analysis of abnormal correlations among multiple blood biochemical index values ​​in the biochemical index data layer identifies index groups that deviate from the normal reference range, including:

[0016] Set upper and lower limits for the normal reference range for each blood biochemical indicator in the biochemical indicator data layer;

[0017] Iterate through all blood biochemical indicator values, mark the indicators whose values ​​are higher than the upper limit of their normal reference range or lower than the lower limit of their normal reference range, and form an initial set of abnormal indicators.

[0018] In the initial set of abnormal indicators, calculate the ratio of the magnitude of any two abnormal indicator values ​​deviating from their respective normal reference range center values. If the magnitude ratio is within the preset correlation threshold range, then the two abnormal indicators are grouped into the same abnormal correlation indicator group.

[0019] For each abnormal correlation index group, calculate the vector sum of the standardized deviations of all abnormal index values ​​within the group. The magnitude of the vector sum is the comprehensive risk field strength of the abnormal correlation index group.

[0020] Furthermore, the structured information extraction from the image examination description text, identifying key anatomical structures, abnormal descriptive words, and quantitative modifiers, specifically includes:

[0021] The pre-trained biomedical entity recognition model scans images to examine descriptive text, identifies and labels all anatomical structure names mentioned in the text.

[0022] Within the context of the identified anatomical structure name entities, extract adjectives and verb phrases used to describe the state of the anatomical structure, and filter out phrases that characterize abnormal states as abnormal descriptive words.

[0023] Further extract degree adverbs or numerical phrases that directly collocate with abnormal descriptive words and parse them into quantitative modifiers;

[0024] The step of mapping the extracted information to the image feature data layer to generate a three-dimensional spatial probability distribution map specifically includes:

[0025] Based on the identified anatomical structure names, locate the corresponding three-dimensional spatial coordinate range in the standard human three-dimensional anatomical atlas;

[0026] Calculate the probability value of the anatomical structure exhibiting a specific abnormal state based on the abnormal descriptive words and quantification modifiers;

[0027] Centered on the three-dimensional spatial coordinate range of the anatomical structure, a Gaussian probability distribution cloud is generated according to the magnitude of the probability value. The Gaussian probability distribution cloud is the three-dimensional spatial probability distribution map of the abnormal state on the anatomical structure.

[0028] If the same anatomical structure corresponds to multiple abnormal states, multiple superimposed Gaussian probability distribution clouds will be generated.

[0029] Furthermore, generating a vital sign offset vector representing the physiological structural shift in the vital sign mapping data layer specifically includes:

[0030] Establish a standard physical characteristic parameter model, which includes standard values ​​and allowable fluctuation ranges for various physical measurement data of different age groups and genders;

[0031] The user's physical measurement data is compared item by item with the standard physical characteristic parameter model of the same age and gender, and the difference between each measurement data and its standard value is calculated.

[0032] Using the physiological dimensions corresponding to each measurement data as the basis, and the differences as the components on the basis, a multidimensional vital sign offset vector is constructed.

[0033] Furthermore, the dynamic health status simulation model is specifically used for:

[0034] The dynamic health status simulation model simulates the coupled evolution path of the comprehensive risk field intensity, three-dimensional spatial probability distribution map, and vital sign offset vector on a preset future time scale.

[0035] The current comprehensive risk field intensity, three-dimensional spatial probability distribution map, and vital sign offset vector are used as the initial state of the coupling evolution path;

[0036] Based on the pathophysiological association rules defined in the medical knowledge graph, we establish the influence relationship between the intensity of the comprehensive risk field and the probability of abnormality of a specific anatomical structure, as well as the influence relationship between the sign offset vector and the rate of change of the intensity of the comprehensive risk field.

[0037] The simulation is advanced in discrete time steps. Within each time step, the numerical value of the comprehensive risk field intensity, the intensity and range of the probability cloud in the three-dimensional spatial probability distribution map, and the direction and magnitude of the symptom offset vector are updated according to the aforementioned influence relationship.

[0038] The simulation is continuously advanced until the preset future timescale endpoint is reached. The continuous changes of the above-mentioned state variables are recorded throughout the simulation process, forming a complete trajectory of the health status evolution.

[0039] Furthermore, the extraction of key evolution nodes from the health status evolution trajectory includes:

[0040] The key evolution node is defined as the moment when the intensity of the comprehensive risk field, the abnormal probability of a specific anatomical structure, or the magnitude of the sign offset vector in the health status evolution trajectory undergoes a step change.

[0041] The curve showing the change of the intensity of the comprehensive risk field over time in the trajectory of the health status evolution;

[0042] Mark the point in time when the absolute value of the slope of the comprehensive risk field intensity curve exceeds the preset slope threshold;

[0043] Simultaneously, monitor the curve of the abnormal probability value corresponding to a specific anatomical structure in the three-dimensional spatial probability distribution map changing over time, and mark the moment when the curve changes abruptly.

[0044] Monitor the curve of the magnitude of the vital sign offset vector changing over time, and mark the moment when the magnitude curve undergoes a step jump.

[0045] By merging all the marked time points, removing duplicates, and sorting them in chronological order, a series of key evolution nodes are obtained.

[0046] Furthermore, for each key evolution node, a personalized node intervention instruction set is generated based on the specific state of the biochemical indicator data layer, image feature data layer, and vital sign mapping data layer at that time point, including:

[0047] For key evolution nodes, read the state snapshot on the health status evolution trajectory corresponding to the key evolution node. The state snapshot includes the comprehensive risk field intensity value at the corresponding time, the details of the three-dimensional spatial probability distribution map, and the specific components of the vital sign offset vector.

[0048] Based on the abnormal correlation index group associated with the comprehensive risk field intensity value in the status snapshot, laboratory re-examination suggestions and lifestyle adjustment suggestions are generated for the abnormal correlation index group.

[0049] Based on the high probability abnormalities of specific anatomical structures shown in the three-dimensional spatial probability distribution map in the status snapshot, targeted imaging review suggestions or specialist consultation suggestions are generated.

[0050] Based on the specific components of the vital sign offset vector in the status snapshot, generate targeted exercise training suggestions or nutritional diet suggestions;

[0051] All suggestions are structured and integrated into a set of node intervention instructions that includes specific action items, expected goals, and execution cycles.

[0052] Furthermore, after generating the personalized node intervention instruction set, the process also includes:

[0053] The node intervention instruction set is compiled into an executable health management task sequence. Based on the health management task sequence, the user interaction terminal is driven to push task reminders to the user and collect task execution feedback data. The task execution feedback data is used to update the user's original physical examination data, forming an iterative health management closed loop.

[0054] The process of compiling the node intervention instruction set into an executable health management task sequence includes:

[0055] The health management task sequence includes specific action items, target values, and execution time points;

[0056] Each suggestion in the node intervention instruction set is analyzed and decomposed into one or more basic task units with clear start and end times, execution frequency, execution content, and quantifiable objectives.

[0057] Considering the logical order and resource conflicts between different basic task units, schedule and prioritize all basic task units according to time.

[0058] Organize the scheduled and sorted basic task units along a timeline to form a coherent sequence of health management tasks that can be checked off one by one.

[0059] Furthermore, the step of updating the user's original physical examination data using task execution feedback data to form an iterative health management closed loop includes:

[0060] Collect task execution feedback data generated by users during the execution of health management task sequences. The task execution feedback data includes task completion records, self-reported new vital signs data, and re-examination and test data conducted midway.

[0061] The new vital signs data and re-examination test data in the task execution feedback data are normalized according to the format of the original physical examination data to form incremental physical examination data.

[0062] The incremental physical examination data is merged with the historical original physical examination data to generate updated user original physical examination data;

[0063] Using the updated user's original physical examination data as new input, the process of constructing a multimodal health field, inferring dynamic health status, and generating health management tasks is restarted to achieve periodic iterative optimization of health management.

[0064] Furthermore, the present invention also includes a health management system based on physical examination reports, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the health management method based on physical examination reports described above.

[0065] Compared with the prior art, the beneficial effects of the present invention are:

[0066] By constructing a multimodal virtual health field comprising a biochemical indicator data layer, an image feature data layer, and a vital sign mapping data layer, blood indicators, image text, and vital sign data are uniformly spatialized and represented based on their inherent correlations. Anomaly correlation analysis of biochemical indicators yields a comprehensive risk field intensity, image text is transformed into a three-dimensional spatial probability distribution map, and vital sign data is quantified into vital sign offset vectors. This approach transforms previously heterogeneous and isolated data into a holistic view with spatiotemporal and logical connections within a unified virtual field. Its quantitative characteristics more profoundly reflect the complex interactions between multiple systems, providing structured high-dimensional input for subsequent dynamic simulations. This reduces the computational overhead of real-time or near-real-time simulation of ultra-large-scale medical correlation matrices, improving the system's processing performance and feasibility.

[0067] By inputting quantitative features such as the comprehensive risk field intensity, three-dimensional spatial probability distribution map, and vital sign offset vectors into a dynamic model, and simulating their coupled evolution process, it is possible to generate health status evolution trajectories reflecting multiple possibilities. Key evolution nodes identified from these trajectories represent crucial moments when a qualitative change in health status may occur or when intervention is necessary. For each node, combined with the specific state parameters of each data layer in the virtual health field at that moment, the generated intervention instruction set is directly linked to a specific risk field intensity distribution, organ probability cloud map morphology, or vital sign vector direction. This transforms health management from general recommendations based on a single point in time to precise control strategies targeting specific stages in the dynamic development of health status. Attached Figure Description

[0068] Figure 1 This is a flowchart illustrating the steps of a health management method based on a physical examination report as described in this invention.

[0069] Figure 2 A flowchart for refining biochemical indicators, imaging, and physical data;

[0070] Figure 3 This is a logical framework diagram of a health management system based on physical examination reports as described in this invention;

[0071] Figure 4 A flowchart for generating the vital sign offset vector;

[0072] Figure 5 A trend chart of the intensity of the comprehensive risk field under the evolution of the health situation;

[0073] Figure 6 This is a trend chart of multi-dimensional improvement rates under the iterative cycle of health management. Detailed Implementation

[0074] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and 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.

[0075] See Figure 1The system acquires raw physical examination data containing multimodal data. This raw data is input into a multimodal health field construction engine. Based on data types and their inherent correlations, the engine creates a biochemical indicator data layer, an imaging feature data layer, and a vital sign mapping data layer within a virtual health field, thus unifying the heterogeneous physical examination data into a structured digital space. Anomaly correlation analysis is performed on multiple blood biochemical indicator values ​​in the biochemical indicator data layer to calculate the comprehensive risk field strength. Structured information is extracted from the imaging examination description text and mapped to the imaging feature data layer to generate a three-dimensional spatial probability distribution map. Anthropometric data is mapped to the vital sign mapping data layer to generate vital sign offset vectors. This processed field information is input into a dynamic health status prediction model. This model simulates the coupling evolution path between the comprehensive risk field strength, the three-dimensional spatial probability distribution map, and the vital sign offset vectors based on medical rules, generating a health status evolution trajectory that predicts future changes in health status. Key evolution nodes are extracted from the generated health status evolution trajectory. For each key evolution node, a personalized node intervention instruction set containing specific action suggestions is generated by combining the specific state snapshots of the biochemical indicator data layer, image feature data layer, and vital sign mapping data layer at that point in time.

[0076] See Figure 2 and Figure 3In one embodiment of the present invention, the system includes a multimodal health field construction engine, a dynamic health status simulation model, and a health management task generation and execution module. The multimodal health field construction engine receives raw blood biochemical indicators, imaging examination description text, and anthropometric data, and creates a biochemical indicator data layer, an imaging feature data layer, and a vital sign mapping data layer in a virtual health field. The dynamic health status simulation model receives the comprehensive risk field intensity, three-dimensional spatial probability distribution map, and vital sign offset vector output from the multimodal health field construction engine, simulates their coupled evolution, and generates a health status evolution trajectory. The health management task generation and execution module extracts key evolution nodes from the trajectory, generates a node intervention instruction set, and compiles it into an executable health management task sequence. This sequence drives task execution through a user interaction terminal and collects feedback data. Finally, the incremental physical examination data formed by the feedback data is sent back to the system front end to update the original physical examination data, thereby forming an iterative health management closed loop. Taking a sedentary office worker as an example, their original physical examination data might show mild dyslipidemia, slight changes in cervical spine curvature in imaging, and excessive weight and body fat percentage. After constructing a virtual health field, the system's simulation model might demonstrate that over the next 6-12 months, the intensity of the comprehensive risk field for dyslipidemia will gradually increase due to lack of exercise, coupling with the probability of cervical spine abnormalities, thus deducing a health trajectory of 'worsening neck and shoulder discomfort accompanied by an increased risk of metabolic syndrome'. The system might extract the 3rd and 8th months as key nodes from the trajectory and generate a set of node intervention instructions, including 'increasing intermittent standing activities', 'targeted cervical spine relaxation training', and 'recommending a follow-up examination of four dyslipidemia tests'. Multimodal data is processed to generate core field features; abnormal correlation analysis is performed on multiple blood biochemical indicator values ​​in the biochemical indicator data layer to calculate the comprehensive risk field intensity; structured information is extracted from the imaging examination description text to generate a three-dimensional spatial probability distribution map; and anthropometric data is mapped to the vital sign mapping data layer to generate a vital sign offset vector. In practice, the abnormal correlation analysis includes setting upper and lower limits of the normal reference range for each blood biochemical indicator in the biochemical indicator data layer, traversing all blood biochemical indicator values ​​and marking indicators whose values ​​are higher than the upper limit of their normal reference range or lower than the lower limit of their normal reference range to form an initial abnormal indicator set, calculating the ratio of the magnitude of any two abnormal indicator values ​​deviating from their respective normal reference range center values ​​in the initial abnormal indicator set, and if the magnitude ratio is within the preset correlation threshold range, then the two abnormal indicators are grouped into the same abnormal correlation indicator group, and calculating the vector sum of the standardized deviation values ​​of all abnormal indicator values ​​in each abnormal correlation indicator group, and using the magnitude of the vector sum as the comprehensive risk field strength of the abnormal correlation indicator group.In some embodiments, the standardized deviation is calculated based on the ratio of the difference between the indicator value and the center value of the normal reference range to the width of the normal reference range. The association threshold interval is defined as a fixed numerical range used to determine abnormal associations between indicators. It can be understood that the association threshold interval is pre-configured based on medical statistical knowledge to reflect abnormal patterns of indicator synergy.

[0077] In practical implementation, the following formula is used to calculate the intensity of the comprehensive risk field:

[0078]

[0079] in: Indicates the overall risk field strength. Indicates the first The standardized deviation vector of an abnormal indicator is defined by the standardized distance and direction of the indicator value from the center value of its normal reference range. This indicates the number of abnormal indicators contained in the abnormal correlation indicator group. This represents the Euclidean modulus of the computed vector. In a specific implementation, structured information extraction is performed on the image examination description text. A pre-trained biomedical entity recognition model scans the text to identify and label all anatomical structure names mentioned. Within the context of the identified anatomical structure names, adjectives and verb phrases describing the state of the anatomical structures are extracted, and phrases representing abnormal states are selected as anomalous descriptors. Further, degree adverbs or numerical phrases directly paired with these anomalous descriptors are extracted and parsed into quantitative modifiers. In some embodiments, the pre-trained biomedical entity recognition model employs an architecture based on a bidirectional long short-term memory network combined with a conditional random field to achieve entity recognition. Optionally, the selection of anomalous descriptors is performed by matching against a predefined medical anomaly dictionary to improve extraction accuracy.

[0080] In practice, the process of mapping the extracted information to the image feature data layer and generating a 3D spatial probability distribution map involves locating the corresponding 3D spatial coordinate range of the identified anatomical structure name entity in the standard human 3D anatomical atlas, calculating the probability value of a specific abnormal state of the anatomical structure based on the anomaly descriptor and quantification modifier, and generating a Gaussian probability distribution cloud centered on the 3D spatial coordinate range of the anatomical structure according to the probability value. This serves as the 3D spatial probability distribution map of the abnormal state on the anatomical structure. If the same anatomical structure corresponds to multiple abnormal states, multiple superimposed Gaussian probability distribution clouds are generated. Optionally, the probability value is calculated by querying a pre-set anomaly descriptor-probability mapping table, and the quantification modifier is used to linearly scale the probability value. In practice, the steps of mapping anthropometric data to a vital sign mapping data layer to generate a vital sign offset vector include: establishing a standard vital sign parameter model containing standard values ​​and allowable fluctuation ranges for various anthropometric data of different age groups and genders; comparing the user's anthropometric data with the standard vital sign parameter model of the same age and gender item by item and calculating the difference between each measurement and its standard value; and constructing a multidimensional vital sign offset vector using the physiological dimensions corresponding to each measurement as the basis and the components based on the differences. Physiological dimensions include height, weight, body fat percentage, and waist circumference. The multidimensional vital sign offset vector mathematically represents the deviation of the user's vital signs from the standard model.

[0081] See Figure 4 In one embodiment of the present invention, the structured information extraction of image examination description text involves scanning the text using a pre-trained biomedical entity recognition model to identify and label all anatomical structure names mentioned in the text. Within the context of the identified anatomical structure names, adjectives and verb phrases describing the state of the anatomical structures are extracted, and phrases representing abnormal states are selected as anomalous descriptive words. Furthermore, degree adverbs or numerical phrases directly paired with the anomalous descriptive words are extracted and parsed into quantified modifiers. In some embodiments, the pre-trained biomedical entity recognition model implements entity recognition based on a bidirectional long short-term memory network and a conditional random field architecture. The selection of anomalous descriptive words is accomplished by matching a predefined medical anomaly dictionary. The parsing of quantified modifiers includes mapping degree adverbs to probability adjustment coefficients and converting numerical phrases into standardized values. It is understood that the medical anomaly dictionary contains medical terms representing abnormal states, and the probability adjustment coefficients are predefined based on linguistic rules.

[0082] In practice, the process of mapping the extracted information to the image feature data layer and generating a 3D spatial probability distribution map involves locating the corresponding 3D spatial coordinate range of the identified anatomical structure name entity in the standard human 3D anatomical atlas, calculating the probability value of a specific abnormal state of the anatomical structure based on the anomaly descriptor and quantification modifier, and generating a Gaussian probability distribution cloud centered on the 3D spatial coordinate range of the anatomical structure according to the probability value. This serves as the 3D spatial probability distribution map of the abnormal state on the anatomical structure. If the same anatomical structure corresponds to multiple abnormal states, multiple superimposed Gaussian probability distribution clouds are generated. Optionally, the probability value is calculated by querying a preset anomaly descriptor-probability mapping table, and the quantification modifier is used to linearly scale the base probability value. The Gaussian probability distribution cloud of the 3D spatial probability distribution map is defined by the following formula:

[0083]

[0084] in: Represents spatial coordinates The probability density at that location, This represents the calculated probability value of the abnormal state. This indicates the center coordinates of the anatomical structure in a standard three-dimensional human anatomy atlas. The standard deviation represents the Gaussian distribution, and its value is related to the spatial dimensions of the anatomical structure. In some embodiments, the standard deviation... The length of the diagonal of the bounding box of the anatomical structure in the standard three-dimensional human anatomy atlas is determined based on the proportion of the anatomical structure.

[0085] In practice, the generation of the vital sign offset vector relies on a pre-established standard vital sign parameter model. This model includes standard values ​​and allowable fluctuation ranges for various physical measurements of different age groups and genders. The user's physical measurements are compared item by item with the standard vital sign parameter model for the same age and gender, and the difference between each measurement and its standard value is calculated. A multidimensional vital sign offset vector is constructed using the physiological dimensions corresponding to each measurement as the basis and the components based on the differences. Optionally, physiological dimensions include height, weight, body fat percentage, and waist circumference. The standard values ​​are calculated based on large-scale population statistics, and the allowable fluctuation range is defined as plus or minus one standard deviation of the standard value. The multidimensional vital sign offset vector is represented as follows: ,in Indicates the first The standardized difference between physical measurement data and standard values. It can be understood that the standardized difference is dimensionless by dividing by half the allowable range of variation to eliminate the influence of dimensions.

[0086] In one embodiment of the present invention, the dynamic health status prediction model operates by simulating the coupled evolution path of the comprehensive risk field intensity, the three-dimensional spatial probability distribution map, and the vital sign offset vector at a preset future time scale. The dynamic health status prediction model uses the current comprehensive risk field intensity, the three-dimensional spatial probability distribution map, and the vital sign offset vector as the initial state of the coupled evolution path. In specific implementation, the dynamic health status prediction model establishes the influence relationship between the comprehensive risk field intensity and the probability of abnormality of specific anatomical structures based on the pathophysiological association rules defined in the medical knowledge graph, and establishes the influence relationship between the vital sign offset vector and the rate of change of the comprehensive risk field intensity. The medical knowledge graph stores the relationships between entities in the form of triples, such as the positive influence between hypertension and the probability of kidney abnormalities. The dynamic health status prediction model advances the simulation in discrete time steps. Within each time step, it updates the value of the comprehensive risk field intensity, the intensity and range of the probability cloud in the three-dimensional spatial probability distribution map, and the direction and magnitude of the vital sign offset vector according to the influence relationship. The simulation is continuously advanced until the preset future time scale endpoint is reached, and the continuous changes of the above-mentioned state quantities are recorded throughout the simulation process to form a complete health status evolution trajectory.

[0087] In some embodiments, the dynamic health status simulation model updates its state at each time step through a state transition function, where an example component of the state transition function describes the change in the intensity of the comprehensive risk field. It can be understood that the discrete time step is set to one month to simulate medium- to long-term health evolution. In a specific implementation, the update of the comprehensive risk field intensity is calculated using the following formula:

[0088]

[0089] in: Indicates the time step The change in the intensity of the comprehensive risk field at that time This represents the influence coefficient of the abnormal probability of a specific anatomical structure on the change in the intensity of the comprehensive risk field in the three-dimensional spatial probability distribution map defined in the medical knowledge graph. Indicates the time step The probability value of abnormality of a specific anatomical structure at that time. This represents the influence coefficient of the vital sign offset vector defined in the medical knowledge graph on the change in the intensity of the comprehensive risk field. Indicates the time step The Euclidean magnitude of the time-signal offset vector. The overall risk field intensity over the time step. numerical value Depend on Calculations show that This represents the actual time interval corresponding to the time step. The intensity and range updates of the probability cloud in the 3D spatial probability distribution map are adjusted using a preset mapping function based on the updated comprehensive risk field intensity value. The direction and magnitude updates of the characteristic offset vector depend on the dynamic relationship between its current state and the change in the comprehensive risk field intensity. Optionally, the dynamic relationship is defined by a set of differential equations and solved numerically.

[0090] In one embodiment of the present invention, the process of extracting key evolution nodes from the health status evolution trajectory and generating intervention instructions involves defining key evolution nodes as the points in the health status evolution trajectory where the intensity of the comprehensive risk field, the abnormal probability of a specific anatomical structure, or the magnitude of the vital sign offset vector undergoes a step change. The process involves monitoring the curve of the comprehensive risk field intensity changing over time on the health status evolution trajectory and marking the points where the absolute value of the slope of the comprehensive risk field intensity curve exceeds a preset slope threshold. Simultaneously, the process involves monitoring the curve of the abnormal probability value corresponding to a preset specific anatomical structure changing over time in the three-dimensional spatial probability distribution map and marking the points where the curve undergoes abrupt changes. The process also involves monitoring the curve of the magnitude of the vital sign offset vector changing over time and marking the points where the magnitude curve undergoes a step change. All marked points are merged, deduplicated, and sorted chronologically to obtain a series of key evolution nodes. In some embodiments, the preset slope threshold is dynamically adjusted according to the health risk assessment level. The moment when the curve changes abruptly is determined by detecting that the first difference of the abnormal probability value exceeds the preset abrupt change threshold. The moment when a step jump occurs is identified by detecting that the second difference of the magnitude of the vital sign offset vector exceeds the preset jump threshold. The key evolution nodes are extracted using the following formula for slope determination:

[0091]

[0092] in: Indicates a point in time The monitored curve is on the trajectory of the health status evolution. The instantaneous slope, This indicates the curve The preset change threshold, curve This can be a comprehensive risk field intensity curve, a probability curve of specific anatomical structural abnormalities, or a magnitude curve of vital sign offset vectors. The merging operation of marked time points removes duplicate timestamps and sorts them in ascending chronological order, outputting the final list of key evolution nodes. This list is stored in tabular form for easy subsequent processing. See Table 1 for an example of key evolution node marking; the table data is for illustrative purposes only.

[0093] Table 1: Key Evolution Node Marker Table

[0094]

[0095] In practice, for each key evolution node, a state snapshot of the health status evolution trajectory corresponding to the key evolution node is read. The state snapshot includes the comprehensive risk field intensity value at the corresponding time, the details of the three-dimensional spatial probability distribution map, and the specific components of the vital sign offset vector. Based on the abnormal correlation index group associated with the comprehensive risk field intensity value in the state snapshot, laboratory re-examination suggestions and lifestyle adjustment suggestions are generated for the abnormal correlation index group. Based on the high probability abnormality of specific anatomical structures shown in the three-dimensional spatial probability distribution map in the state snapshot, targeted imaging re-examination suggestions or specialist consultation suggestions are generated. Based on the specific components of the vital sign offset vector in the state snapshot, targeted exercise training suggestions or nutritional diet suggestions are generated. All suggestions are structured and integrated to form a node intervention instruction set that includes specific action items, expected goals, and execution cycles. Optionally, laboratory follow-up recommendations include blood biochemistry indicators to be reviewed and suggested follow-up time windows; lifestyle adjustment recommendations include specific guidance on dietary and sleep adjustments; imaging follow-up recommendations specify the types of imaging examinations and target anatomical structures to be reviewed; specialist consultation recommendations recommend corresponding medical specialties; exercise training recommendations include exercise types and intensity planning; and nutritional dietary recommendations include guidance on nutrient intake and food selection. It can be understood that the node intervention instruction set is stored in a structured data format, containing fields such as action items, target values, and execution cycles, facilitating subsequent compilation into a health management task sequence. In some embodiments, structured integration is achieved through template filling; different types of recommendations are mapped to preset instruction templates and filled with specific parameters to generate readable text instructions. Optionally, the execution cycle of the node intervention instruction set is calculated and set based on the time points of key evolution nodes and the simulated time scale of the health status evolution trajectory. It can be understood that the node intervention instruction set is generated independently for each key evolution node, forming a series of personalized health intervention guidelines arranged in chronological order.

[0096] See Figure 5 This is a trend chart of the comprehensive risk field intensity under the evolution of the health status, clearly showing the dynamic changes of the comprehensive risk field intensity and the identification results of key nodes within a 0-24 month period. The risk intensity continuously rises from 0 to 12 months, reflecting the deteriorating correlation of abnormal indicator groups; the risk intensity reaches its peak at 12 months, with the slope exceeding a preset threshold, marking it as a key node; the risk intensity remains above the threshold from 12 to 18 months, triggering the key node marking again; and the risk intensity gradually declines from 18 to 24 months, moving away from the high-risk range. By dynamically extrapolating the risk intensity curve, key intervention opportunities can be identified at early nodes (12 and 18 months) when risk undergoes a step change, avoiding the problem of intervention only after the risk has deteriorated to an irreversible stage. The mapping relationship between "slope threshold → risk threshold line" in the chart clarifies the quantitative implementation method of the "preset slope threshold," avoiding the ambiguity of abstract concepts.

[0097] In one embodiment of the present invention, the step of transforming intervention instructions into executable tasks and forming a management closed loop involves generating a node intervention instruction set and then compiling it into an executable health management task sequence. The health management task sequence includes specific action items, target values, and execution time points. Each suggestion in the node intervention instruction set is parsed and decomposed into one or more basic task units with clearly defined start and end times, execution frequency, execution content, and quantifiable goals. Considering the logical order and resource conflicts between different basic task units, all basic task units are scheduled and prioritized. The scheduled and prioritized basic task units are then organized chronologically to form a coherent health management task sequence that can be completed item by item. In some embodiments, the decomposition of basic task units is implemented based on a preset task template library, the time scheduling uses a scheduling algorithm based on time window constraints, and the priority ranking is determined by calculating the priority score of each basic task unit. Calculated using the following formula:

[0098]

[0099] in: Indicates the first Priority scores for each basic task unit. Indicates the first The urgency coefficient of the health risk corresponding to each basic task unit. Indicates the first The estimated execution time required for each basic task unit This represents the maximum estimated execution time among all basic task units. and These are preset weighting coefficients used to balance the impact of urgency and execution time. It can be understood that the logical order includes dependencies between tasks; for example, a review task must be executed after a motion intervention task. Resource conflicts consider the user's time availability and execution capacity.

[0100] In practical implementation, the user interaction terminal is driven by the health management task sequence to push task reminders to users and collect task execution feedback data. This feedback data is then used to update the user's original physical examination data, forming an iterative health management closed loop. Task execution feedback data generated by the user during the execution of the health management task sequence is collected. This data includes task completion records, self-reported new vital signs, and data from follow-up examinations. The new vital signs and follow-up examination data in the task execution feedback data are normalized according to the format of the original physical examination data to form incremental physical examination data. This incremental data is then merged with historical original physical examination data to generate updated user original physical examination data. This updated data serves as the new input to restart the multimodal health field construction, dynamic health status simulation, and health management task generation process, achieving periodic iterative optimization of health management. In some embodiments, the normalization process includes data format standardization, unit unification, and reference range alignment. The fusion operation appends the incremental physical examination data to the historical original physical examination dataset using timestamps. Optionally, the triggering condition for periodic iterative optimization can be a fixed time period or automatic triggering when the changes in key indicators contained in the incremental physical examination data exceed a preset threshold. It can be understood that the restarted process uses updated user-original physical examination data to overwrite the initial data from the previous analysis, thus forming a closed-loop feedback loop.

[0101] See Figure 6 This is a trend chart of multi-dimensional improvement rates across health management iteration cycles, visually demonstrating the changes in the improvement rates of blood lipids, blood pressure, weight, and comprehensive health scores within 1–5 iteration cycles. The chart shows a continuous improvement trend for blood lipids, blood pressure, weight, and comprehensive health scores within 1–5 iteration cycles; the improvement rate of blood lipids consistently leads, followed closely by the improvement rate of the comprehensive health score, reflecting the synergistic improvement of multiple indicators. The growth rate significantly increases in cycles 3–4, corresponding to the feedback optimization process after the intervention instruction set takes effect at key nodes. All four curves show a continuous upward trend, with the improvement rate of blood lipids consistently leading, followed closely by the improvement rate of the comprehensive health score, reflecting the technical effect of "synergistic improvement of multi-dimensional health indicators," corresponding to the update of the original physical examination data by the health management task sequence execution feedback data after the intervention instruction set takes effect at key evolution nodes.

[0102] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

[0103] It should be noted that the methods and systems provided by this invention are intended for personal daily health management and early warning of potential health risks. Their output results are health advice reminders based on data analysis, and do not involve clinical diagnosis of specific diseases, nor should they be used as the sole basis for any medical decision.

Claims

1. A health management method based on a physical examination report, characterized by, include: Obtain raw physical examination data that includes multimodal physical examination data; The original physical examination data is input into the multimodal health field construction engine, and based on the data type and correlation, a biochemical indicator data layer, an image feature data layer, and a vital sign mapping data layer are created in the virtual health field. Anomaly correlation analysis is performed on multiple blood biochemical index values ​​in the biochemical index data layer to calculate the comprehensive risk field intensity. Structured information is extracted from the image examination description text and mapped to the image feature data layer to generate a three-dimensional spatial probability distribution map. Physical measurement data is mapped to the vital sign mapping data layer to generate a vital sign offset vector. The comprehensive risk field intensity, three-dimensional spatial probability distribution map, and vital sign offset vector are input into the dynamic health status inference model to simulate the coupled evolution path and generate a health status evolution trajectory. Key evolution nodes are extracted from the trajectory of health status evolution. For each key evolution node, a personalized set of node intervention instructions is generated by combining the specific status of the biochemical indicator data layer, image feature data layer and vital sign mapping data layer at that time point.

2. The health management method based on physical examination report according to claim 1, characterized in that, The process involves performing anomaly correlation analysis on multiple blood biochemical index values ​​in the biochemical index data layer to calculate the comprehensive risk field intensity; extracting structured information from the image examination description text and mapping it to the image feature data layer to generate a three-dimensional spatial probability distribution map; and mapping physical measurement data to the vital sign mapping data layer to generate a vital sign offset vector. Specifically, this includes: Anomaly correlation analysis is performed on multiple blood biochemical index values ​​in the biochemical index data layer to identify index groups that deviate from the normal reference range. Based on the direction and magnitude of deviation between each index group, the comprehensive risk field intensity of the index group is calculated. Structured information is extracted from the image examination description text to identify key anatomical structures, abnormal descriptive words, and quantitative modifiers. The extracted information is then mapped to the image feature data layer to generate a three-dimensional spatial probability distribution map. The anatomy measurement data is mapped to the vital sign mapping data layer. Based on the difference between the measurement data and the standard body type, a vital sign offset vector representing the physiological structure offset is generated in the vital sign mapping data layer. The analysis of abnormal correlations among multiple blood biochemical index values ​​in the biochemical index data layer identifies index groups that deviate from the normal reference range, including: Set upper and lower limits for the normal reference range for each blood biochemical indicator in the biochemical indicator data layer; Iterate through all blood biochemical indicator values, mark the indicators whose values ​​are higher than the upper limit of their normal reference range or lower than the lower limit of their normal reference range, and form an initial set of abnormal indicators. In the initial set of abnormal indicators, calculate the ratio of the magnitude of any two abnormal indicator values ​​deviating from their respective normal reference range center values. If the magnitude ratio is within the preset correlation threshold range, then the two abnormal indicators are grouped into the same abnormal correlation indicator group. For each abnormal correlation index group, calculate the vector sum of the standardized deviations of all abnormal index values ​​within the group. The magnitude of the vector sum is the comprehensive risk field strength of the abnormal correlation index group.

3. The health management method based on physical examination reports according to claim 2, characterized in that, The process of extracting structured information from the image examination description text, identifying key anatomical structures, abnormal descriptive words, and quantitative modifiers, specifically includes: The pre-trained biomedical entity recognition model scans images to examine descriptive text, identifies and labels all anatomical structure names mentioned in the text. Within the context of the identified anatomical structure name entities, extract adjectives and verb phrases used to describe the state of the anatomical structure, and filter out phrases that characterize abnormal states as abnormal descriptive words. Further extract degree adverbs or numerical phrases that directly collocate with abnormal descriptive words and parse them into quantitative modifiers; The step of mapping the extracted information to the image feature data layer to generate a three-dimensional spatial probability distribution map specifically includes: Based on the identified anatomical structure names, locate the corresponding three-dimensional spatial coordinate range in the standard human three-dimensional anatomical atlas; Calculate the probability value of the anatomical structure exhibiting a specific abnormal state based on the abnormal descriptive words and quantification modifiers; Centered on the three-dimensional spatial coordinate range of the anatomical structure, a Gaussian probability distribution cloud is generated according to the magnitude of the probability value. The Gaussian probability distribution cloud is the three-dimensional spatial probability distribution map of the abnormal state on the anatomical structure. If the same anatomical structure corresponds to multiple abnormal states, multiple superimposed Gaussian probability distribution clouds will be generated.

4. The health management method based on physical examination reports according to claim 3, characterized in that, The generation of a vital sign offset vector representing the physiological structural shift in the vital sign mapping data layer specifically includes: Establish a standard physical characteristic parameter model, which includes standard values ​​and allowable fluctuation ranges for various physical measurement data of different age groups and genders; The user's physical measurement data is compared item by item with the standard physical characteristic parameter model of the same age and gender, and the difference between each measurement data and its standard value is calculated. Using the physiological dimensions corresponding to each measurement data as the basis, and the differences as the components on the basis, a multidimensional vital sign offset vector is constructed.

5. A health management method based on a physical examination report according to claim 4, characterized in that, The dynamic health status simulation model is specifically used for: The dynamic health status simulation model simulates the coupled evolution path of the comprehensive risk field intensity, three-dimensional spatial probability distribution map, and vital sign offset vector on a preset future time scale. The current comprehensive risk field intensity, three-dimensional spatial probability distribution map, and vital sign offset vector are used as the initial state of the coupling evolution path; Based on the pathophysiological association rules defined in the medical knowledge graph, we establish the influence relationship between the intensity of the comprehensive risk field and the probability of abnormality of a specific anatomical structure, as well as the influence relationship between the sign offset vector and the rate of change of the intensity of the comprehensive risk field. The simulation is advanced in discrete time steps. Within each time step, the numerical value of the comprehensive risk field intensity, the intensity and range of the probability cloud in the three-dimensional spatial probability distribution map, and the direction and magnitude of the symptom offset vector are updated according to the aforementioned influence relationship. The simulation is continuously advanced until the preset future timescale endpoint is reached. The numerical value of the comprehensive risk field intensity, the intensity and range of the probability cloud in the three-dimensional spatial probability distribution map, and the continuous changes in the direction and magnitude of the vital sign offset vector are recorded throughout the simulation process, forming a complete trajectory of the health status evolution.

6. A health management method based on a physical examination report according to claim 5, characterized in that, The extraction of key evolution nodes from the health status evolution trajectory includes: The key evolution node is defined as the moment when the intensity of the comprehensive risk field, the abnormal probability of a specific anatomical structure, or the magnitude of the sign offset vector in the health status evolution trajectory undergoes a step change. The curve showing the change of the intensity of the comprehensive risk field over time in the trajectory of the health status evolution; Mark the point in time when the absolute value of the slope of the comprehensive risk field intensity curve exceeds the preset slope threshold; Simultaneously, monitor the curve of the abnormal probability value corresponding to a specific anatomical structure in the three-dimensional spatial probability distribution map changing over time, and mark the moment when the curve changes abruptly. Monitor the curve of the magnitude of the vital sign offset vector changing over time, and mark the moment when the magnitude curve undergoes a step jump. By merging all the marked time points, removing duplicates, and sorting them in chronological order, a series of key evolution nodes are obtained.

7. A health management method based on a physical examination report according to claim 6, characterized in that, For each key evolution node, a personalized node intervention instruction set is generated based on the specific status of the biochemical indicator data layer, image feature data layer, and vital sign mapping data layer at that time point, including: For key evolution nodes, read the state snapshot on the health status evolution trajectory corresponding to the key evolution node. The state snapshot includes the comprehensive risk field intensity value at the corresponding time, the details of the three-dimensional spatial probability distribution map, and the specific components of the vital sign offset vector. Based on the abnormal correlation index group associated with the comprehensive risk field intensity value in the status snapshot, laboratory re-examination suggestions and lifestyle adjustment suggestions are generated for the abnormal correlation index group. Based on the high probability abnormalities of specific anatomical structures shown in the three-dimensional spatial probability distribution map in the status snapshot, targeted imaging review suggestions or specialist consultation suggestions are generated. Based on the specific components of the vital sign offset vector in the status snapshot, generate targeted exercise training suggestions or nutritional diet suggestions; All suggestions are structured and integrated into a set of node intervention instructions that includes specific action items, expected goals, and execution cycles.

8. A health management method based on a physical examination report according to claim 7, characterized in that, The process of generating a personalized set of node intervention instructions also includes: The node intervention instruction set is compiled into an executable health management task sequence. Based on the health management task sequence, the user interaction terminal is driven to push task reminders to the user and collect task execution feedback data. The task execution feedback data is used to update the user's original physical examination data, forming an iterative health management closed loop. The process of compiling the node intervention instruction set into an executable health management task sequence includes: The health management task sequence includes specific action items, target values, and execution time points; Each suggestion in the node intervention instruction set is analyzed and decomposed into one or more basic task units with clear start and end times, execution frequency, execution content, and quantifiable objectives. Considering the logical order and resource conflicts between different basic task units, schedule and prioritize all basic task units according to time. Organize the scheduled and sorted basic task units along a timeline to form a coherent sequence of health management tasks that can be checked off one by one.

9. A health management method based on a physical examination report according to claim 8, characterized in that, The process of updating the user's original physical examination data using task execution feedback data to form an iterative health management closed loop includes: Collect task execution feedback data generated by users during the execution of health management task sequences. The task execution feedback data includes task completion records, self-reported new vital signs data, and re-examination and test data conducted midway. The new vital signs data and re-examination test data in the task execution feedback data are normalized according to the format of the original physical examination data to form incremental physical examination data. The incremental physical examination data is merged with the historical original physical examination data to generate updated user original physical examination data; Using the updated user's original physical examination data as new input, the process of constructing a multimodal health field, inferring dynamic health status, and generating health management tasks is restarted to achieve periodic iterative optimization of health management.

10. A health management system based on physical examination reports, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the health management method based on physical examination reports as described in any one of claims 1 to 9.

Images