Multi-organ health risk assessment and collaborative early warning system and method based on micro-service architecture

Abstract

This invention discloses a multi-organ health risk assessment and collaborative early warning system and method based on a microservice architecture. The system includes: an extraction module that extracts multi-organ digital features from raw, multi-source heterogeneous physiological data; an output module that, based on the multi-organ digital features, executes an adapted artificial intelligence risk assessment model and outputs structured assessment results containing risk levels and key features, generating independent organ early warning events; an analysis module that aggregates structured assessment results from at least two different organs to perform cross-organ risk fusion analysis and generate associated organ early warning events; and an early warning module that, based on preset hierarchical response rules, generates collaborative early warning instructions and issues early warning messages according to independent and associated organ early warning events. This fundamentally overcomes the bottlenecks of rigidity and scalability inherent in traditional monolithic medical early warning systems while ensuring the reliability and accuracy of the analysis results.

Description

Technical Field

[0001] This invention relates to the field of smart healthcare and digital health technology, and in particular to a multi-organ health risk assessment and collaborative early warning system and method based on a microservice architecture. Background Technology

[0002] Currently, with the in-depth implementation of the "Healthy China" strategy and the rapid development of IoT and AI technologies, continuous health monitoring and risk warning services based on smart wearable devices have become an important direction in the field of smart healthcare. By continuously collecting users' physiological signals such as heart rate, blood pressure, blood oxygen, and pulse wave through devices like watches and wristbands, and utilizing big data analysis and AI algorithms for health risk assessment to provide users with early disease warnings, this approach is gradually moving from concept to large-scale application. However, as health management needs expand from single cardiovascular disease risk warnings to comprehensive and coordinated assessments of the functions of multiple organs (or systems) such as the liver, lungs, kidneys, and spleen, existing technological systems face severe challenges in terms of architectural design and business logic. These challenges are mainly reflected in the following aspects: Most existing mainstream health warning systems adopt a traditional integrated, monolithic architecture. Under this architecture, all functional modules (such as data access, feature processing, risk assessment algorithms for different organs, and warning logic) are tightly coupled, compiled, and deployed as a whole. Whenever a new organ risk assessment service needs to be added (e.g., developing a new liver fibrosis risk assessment) or an existing algorithm needs to be optimized, developers must deeply modify and test the entire large and complex codebase, and ultimately redeploy the entire system. This process has a long development cycle, slow iteration speed, and cannot quickly respond to the market's demand for diversified and personalized health services. Furthermore, the lack of cross-organ risk linkage analysis capabilities based on the holistic view of traditional Chinese medicine and modern medical pathophysiology significantly reduces the depth of early warnings and clinical guidance value, thus reducing practicality and adaptability. Summary of the Invention

[0003] To address the aforementioned problems, this invention provides a multi-organ health risk assessment and collaborative early warning system and method based on a microservice architecture. This addresses the issues mentioned in the background art, such as the slow iteration speed of existing organ health assessment architectures, their inability to quickly respond to market demands for diversified and personalized health services, and the lack of cross-organ risk linkage analysis capabilities based on the holistic view of traditional Chinese medicine and modern medical pathophysiology, which significantly reduces the depth of early warning and clinical guidance value, thereby lowering practicality and adaptability.

[0004] A multi-organ health risk assessment and collaborative early warning system based on a microservice architecture, the system comprising: The extraction module is used to receive raw multi-source heterogeneous physiological data from wearable devices and extract standardized digital features of multiple organs; The output module is used to execute an adapted artificial intelligence risk assessment model based on the digital features of multiple organs and output a structured assessment result containing risk level and key features, and generate independent organ early warning events based on the structured assessment result. The analysis module is used to aggregate structured assessment results from at least two different organs and perform cross-organ risk fusion analysis based on a preset knowledge graph of the relationship between traditional Chinese and Western medicine and a preset organ indicator association model. Based on the analysis results, it generates early warning events for related organs. The early warning module is used to generate collaborative early warning instructions and issue early warning messages based on preset hierarchical response rules, independent organ early warning events and related organ early warning events.

[0005] Preferably, the extraction module includes: The receiving submodule is used to receive raw multi-source heterogeneous physiological data streams from wearable devices via MQTT and HTTP protocols and perform cleaning and correction preprocessing on them. The extraction submodule is used to extract the TCM pulse diagnosis features and Western medicine vital signs features from the preprocessed raw multi-source heterogeneous physiological data stream through a multi-dimensional feature extraction engine. The encapsulation submodule is used to format and encapsulate the pulse diagnosis features of traditional Chinese medicine and the vital signs features of Western medicine according to a unified data structure; The first generation submodule is used to generate standardized multi-organ digital features based on the encapsulated TCM pulse diagnosis features and Western medicine vital sign features, and to trigger feature ready events.

[0006] Preferably, the step of extracting the TCM pulse diagnosis features and Western medicine vital sign features from the preprocessed original multi-source heterogeneous physiological data stream using a multi-dimensional feature extraction engine includes: The photoplethysmography (PPG) signal was extracted from the preprocessed raw multi-source heterogeneous physiological data stream using the TCM pulse diagnosis feature extraction model in the multi-dimensional feature extraction engine and decomposed into multiple harmonic components. Based on the pre-set TCM pulse diagnosis knowledge base, N types of pulse features are extracted and quantified from the decomposed harmonic components. According to the TCM viscera theory, the N types of pulse features are mapped to the functional state feature vectors of the corresponding viscera to generate TCM pulse diagnosis features. The Western medicine vital sign feature extraction model in the multi-dimensional feature extraction engine is used to simultaneously extract time-domain and frequency-domain features from the preprocessed raw multi-source heterogeneous physiological data stream. The mean, variance, slope, and periodic variation patterns of time-domain and frequency-domain features within a sliding time window are calculated. Trend features are obtained based on the calculation results, and Western medicine vital signs features are generated based on the trend features.

[0007] Preferably, the output module includes: The model retrieval submodule is used to determine multiple organs to be evaluated and the organ identifier of each organ based on the digital features of multiple organs, and to retrieve and load the corresponding artificial intelligence risk assessment model from the model repository based on the organ identifier. The inference calculation submodule is used to execute the artificial intelligence risk assessment model for each organ to be evaluated to perform inference calculations on the digital feature data stream of the organ to be evaluated and obtain the model inference output. The transformation submodule is used to convert the model inference output into a structured evaluation result that includes risk probability values, risk levels, a list of key anomaly features, and confidence indices. The second generation submodule is used to automatically generate an independent organ early warning event that conforms to the event-driven architecture and publish it to the message middleware when the risk level in the structured assessment results exceeds a preset threshold.

[0008] Preferably, the artificial intelligence risk assessment model for each organ to be assessed performs inference calculations on the digital feature data stream of that organ to be assessed, and obtains the model inference output, including: Based on a preset time window strategy, a model input feature matrix is ​​constructed from the digital feature data stream of each organ to be evaluated; The model input feature matrix is ​​fed into the artificial intelligence risk assessment model of each organ to be assessed to perform forward computation with multiple inference modes; Obtain the raw output of the artificial intelligence risk assessment model and perform post-processing to obtain the processed model inference output.

[0009] Preferably, the analysis module includes: The aggregation submodule is used to receive and aggregate structured evaluation result events of two different organs from the same user within the same time window to form a multi-organ joint feature context. The third generation submodule is used to extract the feature set to be analyzed from the context of multi-organ combined features, perform symbolic logic reasoning based on the preset knowledge graph of the relationship between traditional Chinese and Western medicine, and generate the first risk analysis result. The fourth generation submodule is used to call the preset organ indicator association model to perform data-driven analysis, generate the second risk analysis result, and integrate the first risk analysis result and the second risk analysis result to generate collaborative risk decision. The fifth generation submodule is used to generate standardized related organ early warning events based on the decision content when the collaborative risk decision indicates that there is a significant cross-organ risk.

[0010] Preferably, symbolic logic reasoning is performed based on a pre-defined knowledge graph linking traditional Chinese and Western medicine to generate the first risk analysis result, including: Extract discrete and continuous feature states from the feature set that are associated with nodes in the pre-defined knowledge graph of the relationship between traditional Chinese and Western medicine; Based on discrete and continuous feature states, relevant entity nodes are located in the knowledge graph as the starting point for reasoning. Multi-step traversal is performed along the relation edges to explore potential pathological association paths and determine the composite risk patterns corresponding to each pathological association path. The current multi-organ feature status is determined based on the feature set. The current multi-organ feature status is matched with each composite risk pattern to determine the matching degree. The target composite risk pattern with the highest matching degree is then determined. The target composite risk pattern and its target pathological association pathway are encapsulated as the results of the first risk analysis.

[0011] Preferably, a preset organ indicator correlation model is invoked for data-driven analysis to generate a second risk analysis result, including: The feature set is transformed into a unified, machine-readable joint feature vector, which encodes the risk level of each organ, key abnormal feature values, and temporal relationships between features. The constructed joint feature vector is input into the preset organ index association model to perform forward calculation and obtain the original association analysis results output by the model. The original association analysis results are parsed, calibrated, and quantified to generate a second risk analysis result that includes association strength scores, potential compound risk probabilities, and key association patterns discovered through data-driven analysis.

[0012] Preferably, the early warning module includes: The event attributes of independent organ warning events and related organ warning events are determined separately, and the matching response rules are determined by querying the preset hierarchical response rule base based on the event attributes. Based on response rules and event content, an executable collaborative early warning instruction is synthesized, which includes specific message content, distribution channels, notification recipients, and triggering timing. The system invokes the appropriate communication channels, executes warning commands to send warning messages, tracks the execution status of commands, collects feedback, and generates a delivery report.

[0013] A multi-organ health risk assessment and collaborative early warning method based on a microservice architecture includes the following steps: Receive raw, multi-source, heterogeneous physiological data from wearable devices and extract standardized digital features of multiple organs; Based on the digital features of multiple organs, an adapted artificial intelligence risk assessment model is executed and a structured assessment result containing risk level and key features is output. Based on the structured assessment result, an independent organ early warning event is generated. The system aggregates structured assessment results from at least two different organs and performs cross-organ risk fusion analysis based on a pre-set knowledge graph of the relationship between traditional Chinese and Western medicine and a pre-set organ indicator association model. Based on the analysis results, it generates early warning events for related organs. Based on preset hierarchical response rules, collaborative early warning instructions are generated and early warning messages are issued according to early warning events of independent organs and related organs.

[0014] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0017] Figure 1 This is a schematic diagram of the structure of a multi-organ health risk assessment and collaborative early warning system based on a microservice architecture provided by the present invention. Figure 2 This is a schematic diagram of the structure of the extraction module in a multi-organ health risk assessment and collaborative early warning system based on a microservice architecture provided by the present invention. Figure 3 This is a schematic diagram of the structure of the analysis module in a multi-organ health risk assessment and collaborative early warning system based on a microservice architecture provided by the present invention. Figure 4 The flowchart illustrates the workflow of a multi-organ health risk assessment and collaborative early warning method based on a microservice architecture provided by this invention. Detailed Implementation

[0018] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0019] A multi-organ health risk assessment and collaborative early warning system based on a microservice architecture, such as Figure 1 As shown, the system includes: Extraction module 101 is used to receive raw multi-source heterogeneous physiological data from wearable devices and extract standardized multi-organ digital features; The output module 102 is used to execute an adapted artificial intelligence risk assessment model based on the digital features of multiple organs and output a structured assessment result containing risk level and key features, and generate independent organ early warning events based on the structured assessment result. Analysis module 103 is used to aggregate structured assessment results from at least two different organs and perform cross-organ risk fusion analysis based on a preset knowledge graph of the relationship between traditional Chinese and Western medicine and a preset organ indicator association model, and generate related organ early warning events based on the analysis results. The early warning module 104 is used to generate collaborative early warning instructions and issue early warning messages based on preset hierarchical response rules according to early warning events of independent organs and related organs.

[0020] In this embodiment, standardized multi-organ digital features refer to a set of numerical features with uniform dimensions, numerical ranges and formats extracted from the original physiological data after cleaning and correction. These features include quantified TCM pulse characteristics (such as wiry pulse index and slippery pulse index), Western medicine vital signs time-frequency domain characteristics (such as heart rate variability low-frequency power), and behavioral characteristics such as activity and sleep.

[0021] In this embodiment, both independent organ early warning events and related organ early warning events refer to a standardized message body that follows an event-driven architecture. Its data structure includes at least: event type, unique event identifier, user identifier, timestamp risk level, list of involved organs, risk description, and raw data or feature index. Specifically, related organ early warning events also include related pattern identifier and collaborative risk coefficient as unique fields.

[0022] The working principle of the above technical solution is as follows: First, the extraction module receives raw multi-source heterogeneous physiological data from wearable devices and extracts standardized multi-organ digital features; then, based on the multi-organ digital features, the output module executes an adapted artificial intelligence risk assessment model and outputs a structured assessment result containing risk level and key features, and generates independent organ early warning events based on the structured assessment results; second, the analysis module aggregates the structured assessment results from at least two different organs and performs cross-organ risk fusion analysis based on a preset knowledge graph of the relationship between traditional Chinese and Western medicine and a preset organ indicator association model, and generates associated organ early warning events based on the analysis results; finally, the early warning module generates collaborative early warning instructions based on preset hierarchical response rules and independent organ early warning events and associated organ early warning events, and issues early warning messages.

[0023] The beneficial effects of the above technical solution are as follows: By clearly dividing the complex health risk assessment process into four logical stages—data extraction, independent assessment, collaborative analysis, and early warning issuance—and encapsulating them into independent modules, a high degree of cohesion and loose coupling of system functions is achieved. This architecture not only makes the system easier to develop and maintain, but more importantly, it provides independent evolutionary space for subsequent technical upgrades of each link, without affecting the whole system from a single change. This fundamentally overcomes the bottlenecks of rigidity and scalability of traditional standalone medical early warning systems. Furthermore, the combination of traditional Chinese and Western medicine knowledge graphs and organ indicator correlation models for linkage analysis ensures the reliability and accuracy of the analysis results. This addresses the problems mentioned in the background technology, such as the slow iteration speed of existing organ health assessment architectures, their inability to quickly respond to the market's demand for diversified and personalized health services, and the lack of cross-organ risk linkage analysis capabilities based on the holistic view of traditional Chinese medicine and modern medical pathophysiology, which significantly reduces the depth and clinical guidance value of early warnings, thus reducing practicality and adaptability.

[0024] In one embodiment, such as Figure 2 As shown, the extraction module 101 includes: The receiving submodule 1011 is used to receive the raw multi-source heterogeneous physiological data stream from the wearable device via the MQTT and HTTP protocols and perform cleaning and correction preprocessing on it. The extraction submodule 1012 is used to extract the TCM pulse diagnosis features and Western medicine vital sign features of the preprocessed raw multi-source heterogeneous physiological data stream through a multi-dimensional feature extraction engine. Encapsulation submodule 1013 is used to format and encapsulate the characteristics of pulse diagnosis in traditional Chinese medicine and the vital signs characteristics in Western medicine according to a unified data structure; The first generation submodule 1014 is used to generate standardized multi-organ digital features based on the encapsulated TCM pulse diagnosis features and Western medicine vital sign features, and to trigger feature ready events.

[0025] In this embodiment, the raw multi-source heterogeneous physiological data stream includes photoplethysmography (PPG) signals, triaxial accelerometer signals, and raw blood oxygen saturation data.

[0026] In this embodiment, the multi-dimensional feature extraction engine is also used for: By combining accelerometer data, heart rate variability data, and user sleep history, the sleep-wake cycle of users can be identified, with particular extraction of sleep depth, continuity, and number of awakenings during the Chou hour (1-3 AM). Identify user activity types, intensity, and sedentary behavior patterns based on triaxial accelerometer data; By fusing related features from different data sources, composite feature indicators are generated.

[0027] In this embodiment, the training samples for the preset organ index association model are derived from historical data accumulated during the actual deployment and operation of the system of the present invention, specifically including: (1) Source data composition: Multi-organ digital feature database: Stores time-series digital features of multiple organs such as heart, liver, spleen, lung, and kidney after standardization by the extraction module. These features include quantitative features of pulse diagnosis in traditional Chinese medicine (such as wiry pulse index and slippery pulse index), vital signs features in Western medicine (such as HRV time-domain / frequency-domain indicators and PWTT), and user behavior features (such as sleep depth and activity intensity).

[0028] Risk assessment results library: Stores the structured assessment results of each organ generated by the output module, including risk level, risk probability value, and a list of key abnormal features.

[0029] Clinical Outcome Feedback Database: Collects users' actual health outcomes within a certain period (e.g., 72 hours, 7 days, 30 days) after an alert event is issued, through physician workstations, user follow-ups, or integration with hospital information systems. This includes: Whether a pre-set acute event (such as myocardial infarction or stroke) has occurred; The type and severity of the disease confirmed by medical diagnosis; User's self-reported changes in symptoms and medical records; Doctors label the accuracy of the warning (true positive / false positive / true negative / false negative).

[0030] (2) Sample construction method: Positive samples: Selected cross-organ risk cases that triggered organ warning events (independent or related) and received medical confirmation within the subsequent time window. For example: user data that simultaneously had "cardiac risk level ≥ moderate concern" and "liver syndrome tendency of excessive liver fire", and were diagnosed with hypertensive crisis or arrhythmia within 72 hours.

[0031] Negative samples: User data that did not trigger an alert, or triggered an alert but were confirmed to have no clinical abnormalities. Additionally, long-term monitoring data from a randomly selected group of healthy individuals were used as background negative samples.

[0032] Time window alignment: For each sample, the multi-organ feature sequence of a time window (e.g., 72 hours) before T is taken as the center and used as the model input, and the clinical outcome (whether the target composite risk event occurs) after T is used as the prediction label.

[0033] Feature selection criteria The feature selection for the preset organ index association model is based on the following principles and steps: (1) Prior medical knowledge-driven selection: Based on Traditional Chinese Medicine and Western Medicine Theoretical Basis: Organ associations defined in a pre-defined knowledge graph of Traditional Chinese Medicine and Western Medicine (e.g., "excessive heart and liver fire," "liver and kidney yin deficiency") are selected, along with combinations of organ features relevant to the target composite risk pattern. For example: For the combined risk of "heart and liver fire", a joint feature set of cardiac characteristics (low-frequency power of heart rate variability, nocturnal heart rate trend) and liver characteristics (string pulse index, sleep depth at the hour of Chou) was selected.

[0034] For the combined risk of "liver and kidney yin deficiency", liver characteristics (string pulse index, probability of liver stagnation syndrome) and kidney characteristics (blood pressure waveform characteristics, nocturia-related indicators) were selected.

[0035] (2) Data-driven statistical screening: Correlation analysis: Calculate the Pearson correlation coefficient and mutual information score between the numerical characteristics of each organ and the composite risk outcome, and remove features whose correlation with the target risk is lower than a preset threshold (e.g., |r|<0.1).

[0036] Recursive feature elimination: Using random forest or logistic regression models, iteratively eliminate features that contribute the least to the model's predictions, and retain a subset of features whose cumulative contribution rate reaches a preset threshold (such as 95%).

[0037] Time series importance assessment: For time series features (such as the 24-hour trend of HRV), attention weight analysis using LSTM or Transformer models is used to identify the time segments that contribute the most to risk prediction.

[0038] (3) The final eigenvectors are constructed as follows: The joint feature vector of each sample includes at least the following three types of features: Static features: user age, gender, basic medical history, and TCM constitution identification results.

[0039] Single organ risk characteristics: risk level of each organ (coded as a numerical value from 1 to 5), risk probability value, and key abnormal feature identifiers.

[0040] Cross-organ temporal characteristics: temporal correlation characteristics between indicators of different organs, such as the sliding correlation coefficient between liver wiry pulse index and cardiac HRV low-frequency power in the past 24 hours, and the time lag cross-correlation between liver fire index and blood pressure variability.

[0041] The beneficial effects of the above technical solution are as follows: First, by supporting access via multiple protocols such as MQTT and HTTP, the system's compatibility with wearable devices from different manufacturers and models is enhanced, resolving the issue of data source heterogeneity. Second, by constructing a complete pipeline including cleaning and correction, multi-dimensional feature extraction, and standardized encapsulation, reliable conversion of raw physiological signals into high-quality, standardized digital features is ensured, providing a stable and consistent input foundation for upper-layer AI models. Furthermore, by triggering feature-ready events, asynchronous, event-driven integration with downstream microservices is achieved, improving the overall system response efficiency and throughput.

[0042] In one embodiment, the extraction of TCM pulse diagnosis features and Western medicine vital sign features from the preprocessed raw multi-source heterogeneous physiological data stream using a multi-dimensional feature extraction engine includes: The photoplethysmography (PPG) signal was extracted from the preprocessed raw multi-source heterogeneous physiological data stream using the TCM pulse diagnosis feature extraction model in the multi-dimensional feature extraction engine and decomposed into multiple harmonic components. Based on the pre-set TCM pulse diagnosis knowledge base, N types of pulse features are extracted and quantified from the decomposed harmonic components. According to the TCM viscera theory, the N types of pulse features are mapped to the functional state feature vectors of the corresponding viscera to generate TCM pulse diagnosis features. The Western medicine vital sign feature extraction model in the multi-dimensional feature extraction engine is used to simultaneously extract time-domain and frequency-domain features from the preprocessed raw multi-source heterogeneous physiological data stream. The mean, variance, slope, and periodic variation patterns of time-domain and frequency-domain features within a sliding time window are calculated. Trend features are obtained based on the calculation results, and Western medicine vital signs features are generated based on the trend features.

[0043] In this embodiment, the N types of pulse characteristics include: floating pulse index, deep pulse index, slow pulse index, rapid pulse index, wiry pulse index, and slippery pulse index.

[0044] In this embodiment, the time-domain features include: heart rate, standard deviation of RR interval, root mean square of the difference between adjacent RR intervals, pulse wave conduction time, pulse wave amplitude, etc.

[0045] In this embodiment, the frequency domain features include: the low-frequency power, high-frequency power, and the ratio of low-frequency power to high-frequency power of heart rate variability.

[0046] The beneficial effects of the above technical solution are as follows: through signal decomposition and knowledge base mapping, abstract pulse symptoms are transformed into computable feature vectors, enabling the integration of TCM diagnostic experience into modern data-driven models. Simultaneously, the synchronous extraction of time-domain, frequency-domain, and trend features constitutes a multi-layered and comprehensive characterization of vital signs. This extraction method, which integrates TCM and Western medicine features, provides a unique and rich data foundation for the subsequent construction of assessment models that can both understand abnormalities in Western medicine indicators and identify the risks of TCM syndromes.

[0047] In one embodiment, the output module includes: The model retrieval submodule is used to determine multiple organs to be evaluated and the organ identifier of each organ based on the digital features of multiple organs, and to retrieve and load the corresponding artificial intelligence risk assessment model from the model repository based on the organ identifier. The inference calculation submodule is used to execute the artificial intelligence risk assessment model for each organ to be evaluated to perform inference calculations on the digital feature data stream of the organ to be evaluated and obtain the model inference output. The transformation submodule is used to convert the model inference output into a structured evaluation result that includes risk probability values, risk levels, a list of key anomaly features, and confidence indices. The second generation submodule is used to automatically generate an independent organ early warning event that conforms to the event-driven architecture and publish it to the message middleware when the risk level in the structured assessment results exceeds a preset threshold.

[0048] In this embodiment, the organs to be evaluated include the heart, liver, lungs, kidneys, spleen, etc., and the artificial intelligence risk assessment model corresponding to each organ is selected from the following types: Convolutional neural networks are used to process pulse waveform image features; Recurrent neural networks, or long short-term memory networks, are used to process physiological time-series data; Gradient boosting decision trees or random forests are used to handle combinations of structured features; Multimodal fusion neural networks are used to fuse features from multiple data sources.

[0049] In this embodiment, the model repository is a centralized microservice for model registration and management, used to store artificial intelligence model files of different versions and different organs, and to provide interface for model version control, metadata management and on-demand loading.

[0050] In this embodiment, the independent organ early warning event includes the following data structure: event type, event source: the organ risk assessment service identifier that generated the event, event payload: the complete structured assessment result, event priority: dynamically calculated according to the risk level, event timestamp: the time when the event was generated, and event ID: a globally unique event identifier.

[0051] In this embodiment, the following techniques are used for identifying key anomaly features: Gradient-based feature importance analysis: Evaluates feature importance by calculating the gradient of the model output with respect to the input features; Impact analysis based on perturbation: By systematically perturbing the input characteristics, the degree of change in the model output is observed; Feature selection based on attention weights: directly extracting feature weights from the attention mechanism model; Approximate interpretation based on surrogate models: approximate the behavior of complex models using simple, interpretable models, and extract feature importance from the models.

[0052] The beneficial effects of the above technical solution are as follows: By retrieving a dedicated risk assessment model through organ identification, different organs such as the heart and liver can use algorithms best suited to their physiological characteristics and data patterns, greatly improving the accuracy of single-organ risk assessment. Simultaneously, structuring the model output and automatically generating standardized early warning events not only makes the assessment results readable and interpretable, but also, through an event-driven mechanism, transforms the risk assessment service into an independently deployable and scalable microservice, achieving elastic utilization of system resources and high-concurrency processing.

[0053] In one embodiment, the artificial intelligence risk assessment model for each organ to be assessed performs inference calculations on the digital feature data stream of the organ to be assessed, and obtains the model inference output, including: Based on a preset time window strategy, a model input feature matrix is ​​constructed from the digital feature data stream of each organ to be evaluated; The model input feature matrix is ​​fed into the artificial intelligence risk assessment model of each organ to be assessed to perform forward computation with multiple inference modes; Obtain the raw output of the artificial intelligence risk assessment model and perform post-processing to obtain the processed model inference output.

[0054] In this embodiment, the artificial intelligence risk assessment model includes: Cardiac risk assessment model: It adopts a structure that combines temporal convolutional networks with attention mechanisms, focusing on processing electrocardiogram features and pulse wave features; Liver risk assessment model: Employs a multi-task learning framework to simultaneously output the probability distributions of three TCM syndrome types: liver stagnation, liver fire, and liver yin deficiency. Kidney risk assessment model: Based on gradient boosting tree ensemble algorithm, focusing on the analysis of blood pressure waveform characteristics and urine-related indicators; Lung risk assessment model: Recurrent neural network is used to process respiratory waveform sequences and combined with static features for risk prediction.

[0055] In this embodiment, the various inference modes include: Standard inference mode: Performs complete model forward computation to obtain an accurate risk assessment; Rapid screening mode: Uses a simplified version of the model or an early exit mechanism to quickly filter low-risk samples; Integrated inference model: For high-risk samples, multiple related models are called simultaneously for voting or weighted averaging; Incremental update mode: During model execution, the model parameters are fine-tuned based on real-time feedback.

[0056] The beneficial effects of the above technical solution are as follows: By constructing a feature matrix through a multi-scale time window strategy, the model can simultaneously capture instantaneous anomalies, short-term fluctuations, and long-term trends of physiological indicators, avoiding the limitations of single-time-scale analysis. Furthermore, by supporting multiple inference modes, it can efficiently filter a large amount of low-risk data while ensuring the accuracy of core high-risk case analysis, achieving an optimized balance between accuracy and computational efficiency, making the system more suitable for continuous health monitoring scenarios with high real-time requirements.

[0057] In one embodiment, such as Figure 3 As shown, the analysis module 103 includes: The aggregation submodule 1031 is used to receive and aggregate structured evaluation result events of two different organs within the same time window of the same user to form a multi-organ joint feature context. The third generation submodule 1032 is used to extract the feature set to be analyzed from the context of multi-organ joint features, perform symbolic logic reasoning based on the preset knowledge graph of the relationship between traditional Chinese and Western medicine, and generate the first risk analysis result. The fourth generation submodule 1033 is used to call the preset organ indicator association model to perform data-driven analysis, generate the second risk analysis result, and integrate the first risk analysis result and the second risk analysis result to generate collaborative risk decision. The fifth generation submodule 1034 is used to generate standardized related organ early warning events based on the decision content when the collaborative risk decision indicates that there is a significant cross-organ risk.

[0058] In this embodiment, the multi-organ joint feature context is represented as a user-specific data structure maintained in memory or cache, used to temporarily store and associate the structured evaluation results and original features of multiple organs within the same time window for correlation analysis.

[0059] In this embodiment, symbolic logic reasoning is represented as using a rule engine graph query and reasoning framework based on a preset knowledge graph of the relationship between traditional Chinese medicine and Western medicine to execute logical rules in the form of "IF-THEN" in order to deduce composite risks.

[0060] In this embodiment, a pre-defined knowledge graph of the relationship between traditional Chinese medicine and Western medicine defines the physiological and pathological relationships between organs, the corresponding mapping between combinations of traditional Chinese medicine syndromes and the pathophysiological links in Western medicine, and the early warning rules for cross-organ complex risks, for example: Organ entities and their attributes: physiological functions, common syndromes, and related Western medical indicators of organs such as the heart, liver, spleen, lungs, and kidneys; Types of relationships: These include the relationships of mutual generation, mutual restraint, mutual overacting and mutual insult among the five elements in traditional Chinese medicine, the relationships of internal organs and exterior-interior relationships, the relationships of meridians and collaterals, and the relationships of pathophysiological influence in Western medicine; Composite risk model: a predefined cross-organ high-risk combination model, including syndrome combinations such as "heart and liver fire", "liver and kidney yin deficiency" and "spleen and lung qi deficiency", and associated with specific abnormal indicator threshold combinations; Inference rule base: A set of rules stored in the form of "IF-THEN" to infer the existence and level of complex risks based on the status of multiple organ indicators.

[0061] In this embodiment, the preset organ indicator association model stores at least one organ indicator association model trained based on machine learning, which is used to discover and quantify undefined potential cross-organ indicator statistical association patterns in the knowledge graph. It is constructed using at least one of the following machine learning paradigms: Graph Neural Networks: Treat organs as nodes and relationships between organs as edges, and learn the complex mapping between node features (organ indicators) and edge weights (relationship strength); Attention mechanism network: learns the attention weights between different organ indicators and identifies the cross-organ indicator combination that contributes the most to a specific composite risk; Bayesian networks: model the conditional probability dependencies between organ indicators, and are used to infer the risk probability of other organs when some indicators are abnormal; Temporal association rule mining model: mining the sequential or co-occurrence patterns of cross-organ indicator abnormalities from historical time series data.

[0062] The beneficial effects of the above technical solution are as follows: by aggregating multi-organ assessment results and introducing dual-mode fusion analysis, the system can proactively discover and quantify potential risk correlations between organs. This simulates the comprehensive judgment process of medical experts based on a holistic perspective, enabling the identification of complex risk patterns and achieving a qualitative leap from single-point alarms to systemic risk warnings, significantly improving the depth of warnings and their clinical reference value.

[0063] In one embodiment, symbolic logic reasoning is performed based on a pre-defined knowledge graph linking traditional Chinese and Western medicine to generate a first risk analysis result, including: Extract discrete and continuous feature states from the feature set that are associated with nodes in the pre-defined knowledge graph of the relationship between traditional Chinese and Western medicine; Based on discrete and continuous feature states, relevant entity nodes are located in the knowledge graph as the starting point for reasoning. Multi-step traversal is performed along the relation edges to explore potential pathological association paths and determine the composite risk patterns corresponding to each pathological association path. The current multi-organ feature status is determined based on the feature set. The current multi-organ feature status is matched with each composite risk pattern to determine the matching degree. The target composite risk pattern with the highest matching degree is then determined. The target composite risk pattern and its target pathological association pathway are encapsulated as the results of the first risk analysis.

[0064] The beneficial effects of the above technical solution are as follows: through graph traversal and pattern matching, the system can perform logical deduction like an expert, discovering cross-organ risk pathways that conform to known medical theories. This makes the generated analysis results highly interpretable and the reasoning chain clear, making them easy for medical professionals to understand and verify, thus enhancing the system's credibility and acceptability.

[0065] In one embodiment, a preset organ indicator correlation model is invoked to perform data-driven analysis and generate a second risk analysis result, including: The feature set is transformed into a unified, machine-readable joint feature vector, which encodes the risk level of each organ, key abnormal feature values, and temporal relationships between features. The constructed joint feature vector is input into the preset organ index association model to perform forward calculation and obtain the original association analysis results output by the model. The original association analysis results are parsed, calibrated, and quantified to generate a second risk analysis result that includes association strength scores, potential compound risk probabilities, and key association patterns discovered through data-driven analysis.

[0066] The beneficial effects of the above technical solution are as follows: by directly mining potential statistical association patterns that are not explicitly defined by knowledge graphs from massive historical data through machine learning models, it is possible to discover novel risk combinations or quantify the unique association strength of specific populations. This data-driven insight and knowledge reasoning conclusions corroborate or complement each other, forming a more robust and comprehensive analytical capability.

[0067] In one embodiment, the early warning module is used to generate a coordinated early warning instruction and issue an early warning message based on preset hierarchical response rules according to independent organ early warning events and related organ early warning events, including: The event attributes of independent organ warning events and related organ warning events are determined separately, and the matching response rules are determined by querying the preset hierarchical response rule base based on the event attributes. Based on response rules and event content, an executable collaborative early warning instruction is synthesized, which includes specific message content, distribution channels, notification recipients, and triggering timing. The system invokes the appropriate communication channels, executes warning commands to send warning messages, tracks the execution status of commands, collects feedback, and generates a delivery report.

[0068] The beneficial effects of the above technical solution are as follows: Through a pre-set hierarchical response rule base, the system can automatically match the optimal response strategy based on risk level and event type. Especially for related alerts, mandatory escalation of notification channels and recipients ensures emergency handling of higher-risk situations. End-to-end status tracking and feedback constitute an operational closed loop, guaranteeing the reliable delivery of alert messages and continuously optimizing response rules, thereby improving the overall effectiveness of the alert service and user experience.

[0069] In one embodiment, this embodiment also discloses a multi-organ health risk assessment and collaborative early warning method based on a microservice architecture, such as... Figure 4 As shown, it includes the following steps: Step S401: Receive raw multi-source heterogeneous physiological data from wearable devices and extract standardized multi-organ digital features; Step S402: Based on the digital features of multiple organs, execute the adapted artificial intelligence risk assessment model and output the structured assessment results containing risk level and key features, and generate independent organ early warning events based on the structured assessment results; Step S403: Aggregate structured assessment results from at least two different organs and perform cross-organ risk fusion analysis based on a preset knowledge graph of the relationship between traditional Chinese and Western medicine and a preset organ indicator association model. Generate early warning events for related organs based on the analysis results. Step S404: Based on the preset hierarchical response rules, generate collaborative early warning instructions and issue early warning messages according to independent organ early warning events and related organ early warning events.

[0070] The working principle and beneficial effects of the above technical solution have been explained in the system embodiments and will not be repeated here.

[0071] Those skilled in the art should understand that the terms "first" and "second" in this invention simply refer to different application stages.

[0072] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0073] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A multi-organ health risk assessment and collaborative early warning system based on a micro-service architecture, characterized in that, The system includes: The extraction module is used to receive raw multi-source heterogeneous physiological data from wearable devices and extract standardized digital features of multiple organs; The output module is used to execute an adapted artificial intelligence risk assessment model based on the digital features of multiple organs and output a structured assessment result containing risk level and key features, and generate independent organ early warning events based on the structured assessment result. The analysis module is used to aggregate structured assessment results from at least two different organs and perform cross-organ risk fusion analysis based on a preset knowledge graph of the relationship between traditional Chinese and Western medicine and a preset organ indicator association model. Based on the analysis results, it generates early warning events for related organs. The early warning module is used to generate collaborative early warning instructions and issue early warning messages based on preset hierarchical response rules, independent organ early warning events and related organ early warning events.

2. The multi-organ health risk assessment and collaborative early warning system based on micro-service architecture according to claim 1, characterized in that, The extraction module includes: The receiving submodule is used to receive raw multi-source heterogeneous physiological data streams from wearable devices via MQTT and HTTP protocols and perform cleaning and correction preprocessing on them. The extraction submodule is used to extract the TCM pulse diagnosis features and Western medicine vital signs features from the preprocessed raw multi-source heterogeneous physiological data stream through a multi-dimensional feature extraction engine. The encapsulation submodule is used to format and encapsulate the pulse diagnosis features of traditional Chinese medicine and the vital signs features of Western medicine according to a unified data structure; The first generation submodule is used to generate standardized multi-organ digital features based on the encapsulated TCM pulse diagnosis features and Western medicine vital sign features, and to trigger feature ready events.

3. The multi-organ health risk assessment and collaborative early warning system based on micro-service architecture according to claim 2, characterized in that, The extraction of TCM pulse diagnosis features and Western medicine vital signs features from the preprocessed raw multi-source heterogeneous physiological data stream using a multi-dimensional feature extraction engine includes: The photoplethysmography (PPG) signal was extracted from the preprocessed raw multi-source heterogeneous physiological data stream using the TCM pulse diagnosis feature extraction model in the multi-dimensional feature extraction engine and decomposed into multiple harmonic components. Based on the pre-set TCM pulse diagnosis knowledge base, N types of pulse features are extracted and quantified from the decomposed harmonic components. According to the TCM viscera theory, the N types of pulse features are mapped to the functional state feature vectors of the corresponding viscera to generate TCM pulse diagnosis features. The Western medicine vital sign feature extraction model in the multi-dimensional feature extraction engine is used to simultaneously extract time-domain and frequency-domain features from the preprocessed raw multi-source heterogeneous physiological data stream. The mean, variance, slope, and periodic variation patterns of time-domain and frequency-domain features within a sliding time window are calculated. Trend features are obtained based on the calculation results, and Western medicine vital signs features are generated based on the trend features.

4. The multi-organ health risk assessment and collaborative early warning system based on micro-service architecture according to claim 1, characterized in that, The output module includes: The model retrieval submodule is used to determine multiple organs to be evaluated and the organ identifier of each organ based on the digital features of multiple organs, and to retrieve and load the corresponding artificial intelligence risk assessment model from the model repository based on the organ identifier. The inference calculation submodule is used to execute the artificial intelligence risk assessment model for each organ to be evaluated to perform inference calculations on the digital feature data stream of the organ to be evaluated and obtain the model inference output. The transformation submodule is used to convert the model inference output into a structured evaluation result that includes risk probability values, risk levels, a list of key anomaly features, and confidence indices. The second generation submodule is used to automatically generate an independent organ early warning event that conforms to the event-driven architecture and publish it to the message middleware when the risk level in the structured assessment results exceeds a preset threshold.

5. The multi-organ health risk assessment and collaborative early warning system based on micro-service architecture according to claim 4, characterized in that, The AI ​​risk assessment model for each organ to be assessed performs inference calculations on the digital feature data stream of that organ, and obtains the model inference output, including: Based on a preset time window strategy, a model input feature matrix is ​​constructed from the digital feature data stream of each organ to be evaluated; The model input feature matrix is ​​fed into the artificial intelligence risk assessment model of each organ to be assessed to perform forward computation with multiple inference modes; Obtain the raw output of the artificial intelligence risk assessment model and perform post-processing to obtain the processed model inference output.

6. The multi-organ health risk assessment and collaborative early warning system based on microservice architecture according to claim 1, characterized in that, The analysis module includes: The aggregation submodule is used to receive and aggregate structured evaluation result events of two different organs from the same user within the same time window to form a multi-organ joint feature context. The third generation submodule is used to extract the feature set to be analyzed from the context of multi-organ combined features, perform symbolic logic reasoning based on the preset knowledge graph of the relationship between traditional Chinese and Western medicine, and generate the first risk analysis result. The fourth generation submodule is used to call the preset organ indicator association model to perform data-driven analysis, generate the second risk analysis result, and integrate the first risk analysis result and the second risk analysis result to generate collaborative risk decision. The fifth generation submodule is used to generate standardized related organ early warning events based on the decision content when the collaborative risk decision indicates that there is a significant cross-organ risk.

7. The multi-organ health risk assessment and collaborative early warning system based on microservice architecture according to claim 6, characterized in that, Based on a pre-defined knowledge graph linking traditional Chinese and Western medicine, symbolic logic reasoning is used to generate the first risk analysis result, including: Extract discrete and continuous feature states from the feature set that are associated with nodes in the pre-defined knowledge graph of the relationship between traditional Chinese and Western medicine; Based on discrete and continuous feature states, relevant entity nodes are located in the knowledge graph as the starting point for reasoning. Multi-step traversal is performed along the relation edges to explore potential pathological association paths and determine the composite risk patterns corresponding to each pathological association path. The current multi-organ feature status is determined based on the feature set. The current multi-organ feature status is matched with each composite risk pattern to determine the matching degree. The target composite risk pattern with the highest matching degree is then determined. The target composite risk pattern and its target pathological association pathway are encapsulated as the results of the first risk analysis.

8. The multi-organ health risk assessment and collaborative early warning system based on microservice architecture according to claim 6, characterized in that, The system invokes a pre-defined organ indicator correlation model to perform data-driven analysis, generating a second risk analysis result, including: The feature set is transformed into a unified, machine-readable joint feature vector, which encodes the risk level of each organ, key abnormal feature values, and temporal relationships between features. The constructed joint feature vector is input into the preset organ index association model to perform forward calculation and obtain the original association analysis results output by the model. The original association analysis results are parsed, calibrated, and quantified to generate a second risk analysis result that includes association strength scores, potential compound risk probabilities, and key association patterns discovered through data-driven analysis.

9. The multi-organ health risk assessment and collaborative early warning system based on microservice architecture according to claim 1, characterized in that, The early warning module includes: The determination submodule is used to determine the event attributes of independent organ warning events and related organ warning events respectively, and to determine the matching response rules based on the event attributes by querying the preset hierarchical response rule library. The synthesis submodule is used to synthesize executable collaborative early warning instructions that include specific message content, distribution channels, notification objects, and triggering times based on response rules and event content. The calling submodule is used to call the corresponding communication channels, execute the warning command to complete the issuance of the warning message, track the execution status of the command, collect feedback and generate a reach report.

10. A multi-organ health risk assessment and collaborative early warning method based on a microservice architecture, characterized in that, Includes the following steps: Receive raw, multi-source, heterogeneous physiological data from wearable devices and extract standardized digital features of multiple organs; Based on the digital features of multiple organs, an adapted artificial intelligence risk assessment model is executed and a structured assessment result containing risk level and key features is output. Based on the structured assessment result, an independent organ early warning event is generated. The system aggregates structured assessment results from at least two different organs and performs cross-organ risk fusion analysis based on a pre-set knowledge graph of the relationship between traditional Chinese and Western medicine and a pre-set organ indicator association model. Based on the analysis results, it generates early warning events for related organs. Based on preset hierarchical response rules, collaborative early warning instructions are generated and early warning messages are issued according to early warning events of independent organs and related organs.

Images