Health data processing system and method and computer readable storage medium
The health data processing system with multi-source verification utilizes heterogeneous sensor arrays and dynamic medical knowledge graphs to solve the problem of lack of transparency and credibility in health test results, realize medical endorsement and traceability of test results, and enhance user trust and the authority of the system.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- PING AN TECH (SHENZHEN) CO LTD
- Filing Date
- 2026-01-15
- Publication Date
- 2026-05-12
AI Technical Summary
Existing health testing technologies lack transparency and medical endorsement, resulting in low reliability of results. Users find it difficult to judge the scientific reliability of test results, which severely restricts the application value and user trust of smart devices in disease early warning and assisted diagnosis scenarios.
A health data processing system employing multi-source verification collects multiple physiological signals through a heterogeneous sensor array, performs dynamic consistency checks, and combines a dynamic medical knowledge graph and a medical knowledge fusion reasoning model to generate a health monitoring report with a chain of medical evidence, thereby achieving white-box analysis and logical credibility of the results.
It achieves medical endorsement and traceability of health test results, ensures the physical credibility of input data and the logical credibility of output results, and enhances users' trust in test results and the authority of the system.
Smart Images
Figure CN122025084A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of medical technology and artificial intelligence technology, and in particular to a health data processing system, method and computer-readable storage medium based on multi-source verification. Background Technology
[0002] In the field of medical technology, especially in the application of digital healthcare technology, with the development of artificial intelligence (AI) and the Internet of Things (IoT), intelligent health monitoring devices have been widely used in personal health management. However, existing health monitoring technologies face serious challenges in terms of the reliability and authority of their results.
[0003] Specifically, most current AI health analysis models employ an end-to-end black-box decision-making mechanism, outputting only a number or a simple risk warning, leaving users unaware of the reasoning and logical process behind the result. Furthermore, existing health advice is often trained on general rules or small sample data, lacking direct support from authoritative medical guidelines and clinical evidence-based medicine. This leads to severely generalized advice and even pseudoscientific misleading statements.
[0004] This lack of transparency and medical endorsement makes it difficult for users to judge whether test results are scientifically reliable, severely limiting the application value and user trust of smart devices in disease early warning and assisted diagnosis scenarios. Summary of the Invention
[0005] This application provides a health data processing system, method, and computer-readable storage medium based on multi-source verification, aiming to improve the problems of opaque health monitoring and analysis logic and low reliability of results in the prior art.
[0006] This application provides a health data processing system based on multi-source validation, including: The multi-source heterogeneous acquisition module is configured with a heterogeneous sensor array targeting the same target physiological parameter. The heterogeneous sensor array contains at least three sensors based on different measurement principles for acquiring multiple raw physiological signals from the user. The data confidence verification module is used to receive the multiple raw physiological signals, perform dynamic consistency checks on the multiple raw physiological signals, and determine the valid physiological parameter data based on the check results. The knowledge fusion analysis module stores a pre-built dynamic medical knowledge graph, which is used to extract temporal features from the effective physiological parameter data, match the corresponding clinical phenotype nodes in the dynamic medical knowledge graph based on the temporal features, and input the matched clinical phenotype nodes and the user's covariate information into the medical knowledge fusion reasoning model for calculation to obtain the posterior risk probability value of the target health abnormality. The traceability feedback output module is used to generate a health monitoring report containing the posterior risk probability value, retrieve medical evidence chain information associated with the clinical phenotype node from the dynamic medical knowledge graph, and associate the medical evidence chain information with the health monitoring report for output.
[0007] In one embodiment, the data confidence verification module is configured to: calculate the deviation between any two signal outputs in the multiple raw physiological signals in real time; determine whether the deviation is less than a preset consistency threshold; if it is found that the deviation of at least two independent sensor outputs is less than the consistency threshold, then the corresponding data is identified as the valid physiological parameter data; if it is found that the deviation of all sensor outputs is greater than the consistency threshold, then a confidence downgrade prompt signal is generated, and in response to the confidence downgrade prompt signal, a prompt message is output to guide the user to adjust the contact state between the heterogeneous sensor array and the user or to remain in a resting state.
[0008] In one embodiment, the system further includes an adaptive calibration module, the adaptive calibration module being configured to: Monitor users' real-time physiological status data and identify periods of stable physiological cycles as benchmark calibration windows; Acquire sensor measurement data within the reference calibration window to establish an individualized baseline model; The parameter offset of the heterogeneous sensor array is calculated based on the individualized baseline model, and the parameter offset is used to correct the raw physiological signals acquired subsequently.
[0009] In one embodiment, the pre-built dynamic medical knowledge graph in the knowledge fusion analysis module is constructed and updated through the following steps: Medical guidelines, disease classification standards, and clinical decision support libraries are used as structured knowledge sources. Semantic parsing is performed on the structured knowledge sources to extract medical entities and the relationships between entities. Based on the medical entities and the relationships, a dynamic medical knowledge graph covering different human systems is constructed. The system monitors the information update status of international medical organizations' release channels. When a new version of a medical guideline is detected, it automatically captures the content of the new guideline and synchronously maps the content of the new guideline to the corresponding node in the dynamic medical knowledge graph to update the relationship or the medical entity.
[0010] In one embodiment, when the knowledge fusion analysis module inputs the matched clinical phenotype nodes and user covariate information into the medical knowledge fusion reasoning model for calculation, it is configured as follows: The matched clinical phenotype nodes are used as input features of the graph neural network model in the medical knowledge fusion reasoning model; In the graph neural network model, the covariate information is combined for feature propagation and aggregation to output intermediate feature representations; Bayesian inference is performed on the intermediate feature representation to obtain the prior probability of the target's abnormal health condition; By combining the fluctuation amplitude of the time series characteristics with the prior probability, the final posterior risk probability value is calculated.
[0011] In one embodiment, the traceability feedback output module is configured to output medical evidence chain information as follows: Based on the graph path activated by calculating the posterior risk probability value, backtrack the source node in the dynamic medical knowledge graph; Extract the guide clauses, research document numbers, and recommendation level information stored in the source node; The extracted guideline clauses, research literature numbers, and recommendation level information are combined to form the medical evidence chain information.
[0012] In one embodiment, the system further includes a hierarchical response interaction module, which is configured to: The posterior risk probability value is determined by interval; when the posterior risk probability value is in the first risk interval, the temporal feature and the valid physiological parameter data are automatically encrypted and sent to the preset remote medical service terminal, and the feedback data returned by the remote medical service terminal is received. When the posterior risk probability value is in the second risk range, which is higher than the first risk range and is accompanied by a specific critical indicator, the emergency intervention logic is triggered, and a help signal and location information are sent to a preset external terminal.
[0013] In one embodiment, the hierarchical response interaction module is further configured to: The feedback data is converted into a standard document format and stored in the user's digital archive storage area; In response to the user's evaluation instruction, the system collects the user's validity score for the feedback data or the health monitoring report, and transmits the validity score to the knowledge fusion analysis module to optimize the parameters of the medical knowledge fusion reasoning model.
[0014] This application also provides a health data processing method based on multi-source validation, including: Control the heterogeneous sensor array to acquire multiple raw physiological signals targeting the same physiological parameter; A dynamic consistency test is performed on the multiple raw physiological signals, and valid physiological parameter data are determined based on the test results. Temporal features are extracted from the effective physiological parameter data, and corresponding clinical phenotype nodes are matched in a pre-constructed dynamic medical knowledge graph based on the temporal features. The matched clinical phenotype nodes and user covariate information are input into the medical knowledge fusion reasoning model for calculation to obtain the posterior risk probability value of the target health abnormality. Generate a health monitoring report containing the posterior risk probability value, and output the medical evidence chain information corresponding to the posterior risk probability value.
[0015] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described health data processing method.
[0016] The aforementioned technical solution introduces a dynamic medical knowledge graph and a medical knowledge fusion reasoning model, mapping physiological characteristics into phenotypic nodes with medical semantics. Combined with the output of medical evidence chain information, this achieves a white-box analysis process. Simultaneously, multi-source heterogeneous verification at the front end ensures the physical credibility of the input data, while knowledge reasoning at the back end ensures the logical credibility of the output results. This end-to-end processing mechanism, from physical verification to medical evidence-based medicine, provides rigorous medical endorsement and traceability for the test results, effectively solving the problem of low credibility caused by traditional decision-making. Attached Figure Description
[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a schematic diagram of an application environment for a health data processing system based on multi-source verification, as described in one embodiment of this application. Figure 2 This is a schematic diagram of a system module of a health data processing system based on multi-source verification according to an embodiment of this application; Figure 3 This is a flowchart of a health data processing method based on multi-source verification in one embodiment of this application; Figure 4 This is a schematic diagram of the structure of a computer device according to one embodiment of this application. Detailed Implementation
[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0020] like Figure 1 As shown, Figure 1 This is a schematic diagram of the processing framework of a health data processing system based on multi-source verification in an embodiment of this application. The system communicates with at least three sensors based on different measurement principles. These three sensors, based on different measurement principles, are used to collect multiple raw physiological signals from the user. The server is used to collect the signals collected by the sensors. The health data processing system mainly includes the following modules: The multi-source heterogeneous acquisition module is configured with a heterogeneous sensor array targeting the same target physiological parameter. The heterogeneous sensor array contains at least three sensors based on different measurement principles for acquiring multiple raw physiological signals from the user. The data confidence verification module is used to receive the multiple raw physiological signals, perform dynamic consistency checks on the multiple raw physiological signals, and determine the valid physiological parameter data based on the check results. The knowledge fusion analysis module stores a pre-built dynamic medical knowledge graph, which is used to extract temporal features from the effective physiological parameter data, match the corresponding clinical phenotype nodes in the dynamic medical knowledge graph based on the temporal features, and input the matched clinical phenotype nodes and the user's covariate information into the medical knowledge fusion reasoning model for calculation to obtain the posterior risk probability value of the target health abnormality. The traceability feedback output module is used to generate a health monitoring report containing the posterior risk probability value, retrieve medical evidence chain information associated with the clinical phenotype node from the dynamic medical knowledge graph, and associate the medical evidence chain information with the health monitoring report for output.
[0021] It should be noted that the health data processing system based on multi-source verification provided in this application embodiment can be widely used in application scenarios in the field of medical technology, such as medical auxiliary equipment or systems in hospitals, home medical auxiliary terminals, etc., and is not specifically limited.
[0022] This system introduces a dynamic medical knowledge graph and a medical knowledge fusion reasoning model, mapping physiological features to phenotypic nodes with medical semantics. Combined with the output of medical evidence chain information, this achieves a white-box analysis process. Simultaneously, multi-source heterogeneous verification at the front end ensures the physical credibility of the input data, while knowledge reasoning at the back end ensures the logical credibility of the output results. This end-to-end processing mechanism, from physical verification to medical evidence-based medicine, provides rigorous medical endorsement and traceability for the test results, effectively solving the problem of low credibility caused by traditional decision-making. It should be noted that the above system can be implemented through a server, which can be a standalone server or a server cluster consisting of multiple servers; there is no specific limitation. Furthermore, it should be noted that all data operations such as user information / data collection and processing involved in the embodiments of this application must be performed based on user consent or authorization.
[0023] The solutions provided in this application will now be described in detail through various embodiments.
[0024] This application provides a health data processing system based on multi-source verification. Please refer to... Figure 2 ,include: The multi-source heterogeneous acquisition module is configured with a heterogeneous sensor array targeting the same target physiological parameter. The heterogeneous sensor array contains at least three sensors based on different measurement principles for acquiring multiple raw physiological signals from the user. The data confidence verification module is used to receive the multiple raw physiological signals, perform dynamic consistency checks on the multiple raw physiological signals, and determine the valid physiological parameter data based on the check results. The knowledge fusion analysis module stores a pre-built dynamic medical knowledge graph, which is used to extract temporal features from the effective physiological parameter data, match the corresponding clinical phenotype nodes in the dynamic medical knowledge graph based on the temporal features, and input the matched clinical phenotype nodes and the user's covariate information into the medical knowledge fusion reasoning model for calculation to obtain the posterior risk probability value of the target health abnormality. The traceability feedback output module is used to generate a health monitoring report containing the posterior risk probability value, and retrieve medical evidence chain information associated with the clinical phenotype node from the dynamic medical knowledge graph, and associate the medical evidence chain information with the health monitoring report and output it 32.
[0025] In the above embodiments, the multi-source heterogeneous acquisition module addresses the physical signal-to-noise ratio problem at the data source. A heterogeneous sensor array refers to a solution that, for the same physiological signal (such as heart rate) from a user, does not rely on a single technological path but integrates at least three sensors based on completely different physical mechanisms. For example: Heart rate acquisition uses optical sensors (PPG, which detects changes in blood flow volume and is susceptible to light interference), electrical sensors (ECG, which detects cardiac electrical activity and is susceptible to dryness interference), and mechanical sensors (such as accelerometers or piezoelectric sensors, which detect body movement or pulse wave vibrations and are susceptible to motion interference).
[0026] Blood pressure trends were measured using PPG pulse wave conduction time (PTT) + phonocardiography (PCG) + body position sensor. Respiratory rate detection employs a chest strap strain sensor, radar micro-motion detection, and audio analysis.
[0027] The core principle of this design lies in orthogonal redundancy: because sensors based on different physical principles have different sensitivity characteristics to environmental noise (orthogonality), it is difficult for them to produce the same erroneous readings simultaneously due to the same environmental factor (such as a simple change in light).
[0028] Dynamic Consistency Validation is the core process of data cleaning. Its principle is based on real-time cross-verification of multiple signals. The system does not simply perform addition, but rather calculates the logical consistency between the output values of different sensors. Only when data from different physical sources converge mathematically (i.e., mutually verify each other) does the system consider the data at that moment credible, thus determining it as valid physiological parameter data.
[0029] In the knowledge fusion analysis module, covariates refer to user background variables that significantly influence the prior probability of disease, in addition to real-time sensor data. These include, but are not limited to, age, gender, BMI, smoking history, and family medical history. In medical statistics, the same physiological characteristic (such as decreased heart rate variability) represents drastically different risks in different covariate populations. Clinical phenotype nodes are semantic units in the knowledge graph, mapping digital temporal features collected by sensors (such as ST segment depression of 0.1mV) to medically common terminology nodes, realizing the conversion from engineering language to medical language. The posterior risk probability value can be the final result derived from Bayesian inference, representing a correction value for the user's current disease probability based on the known prior probability (baseline prevalence) and combined with currently observed evidence (sensor features + covariates).
[0030] The medical evidence chain information output by the traceability feedback output module refers to the complete evidence traceability path that supports the system in making risk judgments. It includes the specific medical guideline clauses on which the judgment is based (such as Article x of the ACC / AHA 2017 guidelines), the authoritative document number (DOI) cited, and the recommendation level of the evidence (such as evidence level I, grade A).
[0031] As can be seen, this embodiment provides a health data processing system based on multi-source verification, which can be applied to intelligent interaction scenarios in finance and medical settings. By heterogeneous data acquisition at the hardware level and dynamic consistency verification at the software level, it eliminates environmental noise at the source by utilizing the differences in physical principles, thus solving the problem of inaccurate data. By introducing covariate model reasoning, it achieves personalized analysis that conforms to medical logic. By outputting a medical evidence chain, it endows the test results with extremely high authority and interpretability, thereby solving the problem of the lack of credibility of health monitoring results in existing technologies.
[0032] In one embodiment, the data confidence verification module is configured to perform dynamic consistency checks as follows: The deviation between any two signals output from the multiple raw physiological signals is calculated in real time; it is then determined whether the deviation is less than a preset consistency threshold. If it is found that the deviation values of at least two independent sensor outputs are less than the consistency threshold, then the corresponding data is identified as the valid physiological parameter data; If the deviation values of all sensor outputs are found to be greater than the consistency threshold, a confidence downgrade prompt signal is generated, and in response to the confidence downgrade prompt signal, a prompt message is output to guide the user to adjust the contact state between the heterogeneous sensor array and the user or to keep it in a resting state.
[0033] In the above embodiments, the dynamic consistency check process employs differential comparison logic. The system receives multiple raw signal streams from a heterogeneous sensor array in parallel and calculates their respective independent measurement values in real time. For example, the heart rate value (HR) calculated by PPG. ppg Heart rate (HR) calculated from ECG ecg Next, the data confidence verification module calculates the absolute deviation between any two signals: =|HR ppg -HR ecg |
[0034] The preset consistency threshold can be set based on the physiologically permissible error range (e.g., ±5% or ±3 bpm).
[0035] When the data confidence verification module identifies that the deviation values of at least two sensors based on different physical principles (such as PPG and ECG) are less than the threshold, according to the independent source principle in probability theory, the probability that two sensors affected by different interference factors will produce the same error at the same time is extremely low. Therefore, the data confidence verification module determines that the two signals reflect the real physiological condition, and can mark their weighted average as valid physiological parameter data and input it into the knowledge fusion analysis module.
[0036] Conversely, if the deviation values between all pairs of sensors exceed the threshold (e.g., PPG measured at 80 due to strong light interference, ECG at 140 due to sweat interference, and the accelerometer detecting high-frequency vibration), it indicates that the current measurement environment is extremely harsh, the signal-to-noise ratio is extremely low, and no algorithm compensation can restore the true value. In this case, the data confidence verification module generates a confidence downgrade prompt signal, triggering the user interaction logic. The device will output clear prompts via the screen or vibration (e.g., if severe shaking is detected, please remain still or wipe away sweat), proactively guiding the user to improve physical acquisition conditions. In specific applications, the system will activate the confidence downgrade mode, mark the data as pending verification, and prompt the user to re-wear the device or perform a resting measurement on the corresponding APP interface.
[0037] This embodiment breaks away from the traditional logic of forcibly outputting readings regardless of the environment. It accurately eliminates artifact data through a real-time mathematical verification mechanism, significantly reducing the false alarm rate. At the same time, it incorporates the user into a data quality control closed loop. If the deviation value of all sensor output results is found to be greater than the consistency threshold, a confidence downgrade prompt signal is generated. In response to the confidence downgrade prompt signal, prompt information is output to guide the user to adjust the contact state between the heterogeneous sensor array and the user or to keep it in a resting state. This ensures the reliability of the input data from the source and lays a data foundation for the credibility of the final result.
[0038] In one embodiment, the system further includes an adaptive calibration module, the adaptive calibration module being configured to: Monitor users' real-time physiological status data and identify periods of stable physiological cycles as benchmark calibration windows; Acquire sensor measurement data within the reference calibration window to establish an individualized baseline model; The parameter offset of the heterogeneous sensor array is calculated based on the individualized baseline model, and the parameter offset is used to correct the raw physiological signals acquired subsequently.
[0039] In the above embodiments, the adaptive calibration module aims to address the zero-point drift problem of sensors after long-term use. Its processing is based on the stability of human circadian rhythms. The adaptive calibration module continuously monitors the user's acceleration (to determine body movement) and heart rate characteristics, identifying stable physiological cycles, most typically the NREM (Nighttime Deep Sleep) stage. During this period, the human body is dominated by the parasympathetic nervous system, with a constant metabolic rate, minimal fluctuations in vital signs, and minimal external interference; this period is defined by the system as the baseline calibration window.
[0040] Within this window, the adaptive calibration module collects sensor readings to construct a personalized baseline model. For example, if the adaptive calibration module detects that a user's PPG signal baseline during deep sleep over the past month exhibits a non-physiological linear monotonic drift (such as a gradual increase in baseline reading due to light intensity decay caused by LED aging), the adaptive calibration module will determine that this has resulted in a parameter offset. It will calculate the specific value of this offset and use it as a correction factor in subsequent daytime activity measurements, subtracting or compensating for it during the original signal preprocessing stage.
[0041] As can be seen, this embodiment achieves self-maintenance of the device, which can automatically correct measurement errors caused by hardware aging without the need for external professional standard sources. At the same time, this calibration is based on the user's own data, so it can adapt to individual differences among different users (such as skin color and blood vessel stiffness), ensuring the longitudinal consistency and accuracy of the device's measurements throughout its entire life cycle.
[0042] In one embodiment, the pre-built dynamic medical knowledge graph in the knowledge fusion analysis module is constructed and updated through the following steps: Medical guidelines, disease classification standards, and clinical decision support libraries are used as structured knowledge sources. Semantic parsing is performed on the structured knowledge sources to extract medical entities and the relationships between entities. Based on the medical entities and the relationships, a dynamic medical knowledge graph covering different human systems is constructed. The system monitors the information update status of international medical organizations' release channels. When a new version of a medical guideline is detected, it automatically captures the content of the new guideline and synchronously maps the content of the new guideline to the corresponding node in the dynamic medical knowledge graph to update the relationship or the medical entity.
[0043] In the above embodiments, the construction process of the dynamic medical knowledge graph utilizes Natural Language Processing (NLP) technology and accesses authoritative structured knowledge sources, including, but not limited to, the International Classification of Diseases (ICD-10 / 11), clinical decision support systems (such as UpToDate), and guidelines published by medical associations in various countries (such as ACC / AHA). The system extracts standardized medical entities (such as systolic blood pressure, atrial fibrillation, and QT interval) and their logical relationships (such as threshold determination relationships and causal risk relationships) from massive amounts of text using entity recognition (NER) and relation extraction (RE) algorithms. This constructs a complex semantic network covering cardiovascular, respiratory, and other systems, thereby building a dynamic medical knowledge graph covering different human systems. Its update mechanism can be based on web crawling technology. The system monitors the official release channels of international medical organizations in real time. When a new version of the guidelines is detected (for example, the diagnostic threshold for a certain disease is adjusted from 140 mmHg to 130 mmHg), the system will automatically capture the change information, locate the corresponding medical entity node in the graph, and dynamically update its attribute values or logical rules.
[0044] As can be seen, this embodiment ensures that the system's judgment logic is always based on the latest global medical authoritative standards, rather than outdated algorithm models. This up-to-date professionalism is the core of establishing the authority of the results, solving the problem of misjudgment caused by lagging medical standards, and making it more intelligent.
[0045] In one embodiment, when the knowledge fusion analysis module inputs the matched clinical phenotype nodes and user covariate information into the medical knowledge fusion reasoning model for calculation, it is configured as follows: The matched clinical phenotype nodes are used as input features of the graph neural network model in the medical knowledge fusion reasoning model; In the graph neural network model, the covariate information is combined for feature propagation and aggregation to output intermediate feature representations; Bayesian inference is performed on the intermediate feature representation to obtain the prior probability of the target's abnormal health condition; By combining the fluctuation amplitude of the time series characteristics with the prior probability, the final posterior risk probability value is calculated.
[0046] In the above embodiments, as one implementation method, the medical knowledge fusion reasoning model adopts a hybrid architecture of graph neural network (GNN) and Bayesian inference, aiming to simulate the clinical thinking of doctors. The specific processing procedure is as follows: First, the knowledge fusion analysis module extracts and maps clinical phenotypic nodes (such as detected ST-segment depression) from sensor data and inputs them into the graph neural network. During feature propagation in the graph, user covariate information (such as age, gender, and history of diabetes) is used as parameters to adjust edge weights. This means that the same phenotypic node will have different propagation paths and intensities in the graph under different covariate backgrounds.
[0047] Subsequently, the intermediate feature representations containing contextual information generated through graph aggregation are input into the Bayesian inference module. This module, based on Bayes' theorem, combines prior probabilities with the likelihood of the current temporal features to calculate the posterior risk probability. For example, for the feature of chest pain, if the covariates indicate that the user is elderly and smokes, the Bayesian module will significantly amplify the posterior probability of them having acute coronary syndrome based on prior knowledge.
[0048] For example, raw physiological data can be acquired from multiple sensors, and temporal features (such as heart rate variability (HRV) and oxygen saturation fluctuation amplitude) can be extracted. Abnormal features can be matched with clinical phenotype nodes in D-MKG, such as mapping "frequent hypoxemia at night + snoring spectrum concentrated in 50–100Hz" to "high-risk clues for obstructive sleep apnea". Graph neural networks (GNN) combined with Bayesian inference can be used to calculate the prior probability and posterior risk of various diseases, taking into account covariates such as age, gender, BMI, and past medical history.
[0049] As can be seen, the technical effect of this embodiment is that it overcomes the drawback of traditional algorithms that rely solely on a single signal threshold for a one-size-fits-all judgment. Instead, by integrating individual user covariate information and probabilistic reasoning, the system can provide highly personalized health assessments for people with different risk backgrounds, thus solving the problem of inaccurate generalization of results.
[0050] In one embodiment, the traceability feedback output module is configured to output medical evidence chain information as follows: Based on the graph path activated by calculating the posterior risk probability value, backtrack the source node in the dynamic medical knowledge graph; Extract the guide clauses, research document numbers, and recommendation level information stored in the source node; The extracted guideline clauses, research literature numbers, and recommendation level information are combined to form the medical evidence chain information.
[0051] In the above embodiments, the generation of medical evidence chain information employs a graph path backtracking processing mechanism. In this embodiment, when the reasoning model calculates a high-risk probability value, the system records the reasoning path actually activated in the knowledge graph during the calculation process (i.e., the path from the symptom node, through which intermediate logical nodes, and finally to the disease node). The system backtracks along this path until it locates the source node that serves as the foundation of knowledge. These source nodes are associated with metadata during the graph construction phase, including: the specific medical guideline clauses on which the judgment logic is based (e.g., based on the ESC2020 atrial fibrillation management guidelines), the publication number (DOI) of the research supporting the logic, and the evidence-based medicine recommendation level of the evidence (e.g., Class I, Level A). The system extracts and combines this metadata to form a complete evidence chain, thus obtaining the medical evidence chain information.
[0052] In this embodiment, the AI decision-making process is displayed. Users can not only know the risk outcome, but also clearly know which authoritative guideline the outcome is based on. This transparent display eliminates users' doubts about the report results, greatly enhances the system's authority, and provides doctors with easily verifiable reference materials and records, solving the problem of unreliable results.
[0053] In one embodiment, the system further includes a hierarchical response interaction module, which is configured to: The posterior risk probability value is determined by interval determination; When the posterior risk probability value is within the first risk range, the time series features and the valid physiological parameter data are automatically encrypted and sent to the preset remote medical service terminal, and the feedback data returned by the remote medical service terminal is received. When the posterior risk probability value is in the second risk range, which is higher than the first risk range and is accompanied by a specific critical indicator, the emergency intervention logic is triggered, and a help signal and location information are sent to a preset external terminal.
[0054] In the above embodiments, the tiered response interaction module establishes a service hierarchy from routine monitoring to medical intervention through interval determination logic. The system monitors and calculates the posterior risk probability value in real time.
[0055] If the value falls within the first risk range (e.g., medium risk, indicating potential health risks but no immediate life-threatening danger), the system determines that professional medical intervention is required but emergency treatment is not necessary. Therefore, it performs a data upload operation, connects to a remote medical service (such as an internet hospital platform or medical terminal), and obtains the doctor's review opinions and feedback data.
[0056] If the value falls within the second risk range (e.g., extremely high risk) and specific critical indicators are detected simultaneously (e.g., severe ST segment abnormalities on ECG and complete loss of body movement as shown by the accelerometer), the system determines that the user may be in a state of acute illness with loss of mobility. At this time, the system immediately triggers emergency intervention logic and automatically sends a distress signal containing GPS location to a preset external terminal (e.g., 120 emergency center or a pre-set family member's mobile phone).
[0057] This embodiment breaks down the information silos of health monitoring devices. Through a hierarchical linkage mechanism, it can provide timely medical guidance for sub-healthy users and ensure that critically ill users receive timely treatment, thus constructing a complete medical service closed loop and enhancing the system's practical value and security.
[0058] In one embodiment, the hierarchical response interaction module is further configured to: The feedback data is converted into a standard document format and stored in the user's digital archive storage area; In response to the user's evaluation instruction, the system collects the user's validity score for the feedback data or the health monitoring report, and transmits the validity score to the knowledge fusion analysis module to optimize the parameters of the medical knowledge fusion reasoning model.
[0059] In the above embodiments, the system incorporates a data standardization and feedback learning mechanism. The feedback data returned by doctors is converted by the system's built-in conversion engine into internationally recognized standard document formats such as HL7 / FHIR and stored in the user's digital profile, ensuring data can be transferred across institutions.
[0060] Simultaneously, the system incorporates a Human Feedback Reinforcement Learning (RLHF) mechanism. After a user receives a report or doctor's advice, the system collects the user's evaluation of the report's accuracy (e.g., marking it as a false positive or helpful). This evaluation is then transmitted back to the cloud as a reward signal to fine-tune the weight parameters of the neural network in the inference model.
[0061] In this embodiment, not only is the interoperability of data improved, making it easier for users to seek medical treatment across institutions, but the system is also endowed with the ability to self-evolve, enabling it to continuously optimize algorithm accuracy based on real-world human feedback, becoming more precise and intelligent over time.
[0062] The above modules can be embedded in the processor of the computer device in hardware form or independent of it, or they can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of the above modules.
[0063] In one embodiment, such as Figure 3As shown, a health data processing method based on the system described in any one of claims 1-8 is provided, the method comprising the following steps: S10. Control the heterogeneous sensor array to acquire multiple raw physiological signals targeting the same physiological parameter; S20. Perform a dynamic consistency check on the multiple raw physiological signals and determine the valid physiological parameter data based on the check results; S30. Extract temporal features from the effective physiological parameter data, and match the corresponding clinical phenotype nodes in the pre-constructed dynamic medical knowledge graph based on the temporal features; S40. Input the matched clinical phenotype nodes and user covariate information into the medical knowledge fusion reasoning model for calculation to obtain the posterior risk probability value of the target health abnormality. S50. Generate a health monitoring report containing the posterior risk probability value, and output the medical evidence chain information corresponding to the posterior risk probability value.
[0064] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.
[0065] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and the database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. When executed by the processor, the computer program implements the functions of a health data processing system based on multi-source verification, or, when executed, implements the steps of a health data processing method based on multi-source verification.
[0066] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the functions of a health data processing system based on multi-source verification, or executes the steps of a health data processing system method based on multi-source verification, to avoid repetition.
[0067] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When executed by a processor, the computer program implements the functions of a health data processing system based on multi-source verification, or when executed, implements the steps of a health data processing method based on multi-source verification. To avoid repetition, these steps will not be repeated here.
[0068] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0069] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.
[0070] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.
Claims
1. A health data processing system based on multi-source validation, characterized in that, include: The multi-source heterogeneous acquisition module is configured with a heterogeneous sensor array targeting the same target physiological parameter. The heterogeneous sensor array contains at least three sensors based on different measurement principles for acquiring multiple raw physiological signals from the user. The data confidence verification module is used to receive the multiple raw physiological signals, perform dynamic consistency checks on the multiple raw physiological signals, and determine the valid physiological parameter data based on the check results. The knowledge fusion analysis module stores a pre-built dynamic medical knowledge graph, which is used to extract temporal features from the effective physiological parameter data, match the corresponding clinical phenotype nodes in the dynamic medical knowledge graph based on the temporal features, and input the matched clinical phenotype nodes and the user's covariate information into the medical knowledge fusion reasoning model for calculation to obtain the posterior risk probability value of the target health abnormality. The traceability feedback output module is used to generate a health monitoring report containing the posterior risk probability value, retrieve medical evidence chain information associated with the clinical phenotype node from the dynamic medical knowledge graph, and associate the medical evidence chain information with the health monitoring report for output.
2. The system according to claim 1, characterized in that, The data confidence verification module is configured to perform dynamic consistency checks as follows: Calculate the deviation between any two output signals from the multiple raw physiological signals in real time; determine whether the deviation is less than a preset consistency threshold; If it is found that the deviation values of at least two independent sensor outputs are less than the consistency threshold, then the corresponding data is identified as the valid physiological parameter data; If the deviation values of all sensor outputs are found to be greater than the consistency threshold, a confidence downgrade prompt signal is generated, and in response to the confidence downgrade prompt signal, a prompt message is output to guide the user to adjust the contact state between the heterogeneous sensor array and the user or to keep it in a resting state.
3. The system according to claim 1, characterized in that, The system also includes an adaptive calibration module, which is configured to: Monitor users' real-time physiological status data and identify periods of stable physiological cycles as benchmark calibration windows; Acquire sensor measurement data within the reference calibration window to establish an individualized baseline model; The parameter offset of the heterogeneous sensor array is calculated based on the individualized baseline model, and the parameter offset is used to correct the raw physiological signals acquired subsequently.
4. The system according to claim 1, characterized in that, The pre-built dynamic medical knowledge graph in the knowledge fusion and analysis module is constructed and updated through the following steps: Medical guidelines, disease classification standards, and clinical decision support libraries are used as structured knowledge sources. Semantic parsing is performed on the structured knowledge sources to extract medical entities and the relationships between entities. Based on the medical entities and the relationships, a dynamic medical knowledge graph covering different human systems is constructed. The system monitors the information update status of international medical organizations' release channels. When a new version of a medical guideline is detected, it automatically captures the content of the new guideline and synchronously maps the content of the new guideline to the corresponding node in the dynamic medical knowledge graph to update the relationship or the medical entity.
5. The system according to claim 1, characterized in that, When the knowledge fusion analysis module inputs the matched clinical phenotype nodes and user covariate information into the medical knowledge fusion reasoning model for calculation, it is configured as follows: The matched clinical phenotype nodes are used as input features of the graph neural network model in the medical knowledge fusion reasoning model; In the graph neural network model, the covariate information is combined for feature propagation and aggregation to output intermediate feature representations; Bayesian inference is performed on the intermediate feature representation to obtain the prior probability of the target's abnormal health condition; By combining the fluctuation amplitude of the time series characteristics with the prior probability, the final posterior risk probability value is calculated.
6. The system according to claim 1, characterized in that, The traceability feedback output module is configured to output medical evidence chain information as follows: Based on the graph path activated by calculating the posterior risk probability value, backtrack the source node in the dynamic medical knowledge graph; Extract the guide clauses, research document numbers, and recommendation level information stored in the source node; The extracted guideline clauses, research literature numbers, and recommendation level information are combined to form the medical evidence chain information.
7. The system according to any one of claims 1-6, characterized in that, The system also includes a hierarchical response interaction module, which is configured as follows: The posterior risk probability value is determined by interval determination; When the posterior risk probability value is within the first risk range, the time series features and the valid physiological parameter data are automatically encrypted and sent to the preset remote medical service terminal, and the feedback data returned by the remote medical service terminal is received. When the posterior risk probability value is in the second risk range, which is higher than the first risk range, and a specific critical indicator is detected, the emergency intervention logic is triggered, and a distress signal and location information are sent to a preset external terminal.
8. The system according to claim 7, characterized in that, The hierarchical response interaction module is also configured to: The feedback data is converted into a standard document format and stored in the user's digital archive storage area; In response to the user's evaluation instruction, the system collects the user's validity score for the feedback data or the health monitoring report, and transmits the validity score to the knowledge fusion analysis module to optimize the parameters of the medical knowledge fusion reasoning model.
9. A health data processing method based on the system described in any one of claims 1-8, characterized in that, include: Control the heterogeneous sensor array to acquire multiple raw physiological signals targeting the same physiological parameter; A dynamic consistency test is performed on the multiple raw physiological signals, and valid physiological parameter data are determined based on the test results. Temporal features are extracted from the effective physiological parameter data, and corresponding clinical phenotype nodes are matched in a pre-constructed dynamic medical knowledge graph based on the temporal features. The matched clinical phenotype nodes and user covariate information are input into the medical knowledge fusion reasoning model for calculation to obtain the posterior risk probability value of the target health abnormality. Generate a health monitoring report containing the posterior risk probability value, and output the medical evidence chain information corresponding to the posterior risk probability value.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in claim 9.