Gastric disease early warning system and method based on traditional chinese medicine syndrome differentiation and ai model

Abstract

The application discloses a stomach disease early warning system and method based on traditional Chinese medicine syndrome differentiation and an AI model, and the system comprises the following modules: a receiving module that receives continuous physiological time series data and structured symptom and behavior data of a target user; an extraction module that extracts western medicine physiological characteristics and traditional Chinese medicine syndrome differentiation quantitative characteristics related to stomach function from the data; an analysis module that performs deep correlation analysis on the western medicine physiological characteristics and the traditional Chinese medicine syndrome differentiation quantitative characteristics by using an AI model, and outputs a probability distribution of stomach disease syndromes and a stomach health risk index; and a generation module that generates graded warning information based on a preset stomach disease traditional Chinese medicine syndrome differentiation knowledge graph according to the probability distribution of stomach disease syndromes and the stomach health risk index, and outputs health intervention suggestions and medical treatment prompts to the target user based on the graded warning information. The seamless connection from data collection to personalized early warning intervention is realized, and a continuous, non-invasive and pre-position daily management scheme is provided for stomach health.

Description

Technical Field

[0001] This invention relates to the field of smart healthcare and digital health technology, and in particular to an early warning system and method for stomach diseases based on traditional Chinese medicine diagnosis and AI models. Background Technology

[0002] Currently, gastric diseases constitute a global health burden, particularly in China, where the incidence and mortality rates of gastric cancer remain high. The gold standard for diagnosing gastric diseases is currently gastroscopy and pathological biopsy, but these methods have limitations such as being invasive, costly, dependent on medical resources, and having low patient compliance, making them unsuitable for large-scale routine screening and early warning. While some existing non-invasive screening methods, such as AI-based imaging screening models based on plain CT scans, represent a significant breakthrough, they are essentially sporadic, hospital-equipped examinations that cannot provide continuous, dynamic tracking of individual health trends. Other technologies attempt to diagnose by analyzing biomarkers such as gastric juice, but sampling is inconvenient, making continuous monitoring difficult. Furthermore, some existing AI-assisted diagnosis and treatment systems in Traditional Chinese Medicine (such as the "Huang Huang Classical Formula AI-Assisted Diagnosis and Treatment System") can differentiate syndromes based on user-inputted symptom information, but their data sources rely on single or intermittent consultations, failing to integrate with the continuous physiological monitoring data streams provided by wearable devices. This results in their inability to capture subtle changes in bodily functions before and after symptom onset, and the lack of continuous background support for early warning. Summary of the Invention

[0003] To address the aforementioned problems, this invention provides an early warning system and method for gastric diseases based on traditional Chinese medicine (TCM) diagnosis and AI models. This addresses the issues mentioned in the background art, such as the discrete nature of precision instrument examinations, the lack of continuous background support for early warnings, and the fact that while consumer-grade continuous health monitoring devices are convenient and non-invasive, they lack the ability to quantify and deeply analyze gastric diseases using TCM diagnosis, resulting in low accuracy of the final analysis results and a reduced user experience.

[0004] An early warning system for stomach diseases based on traditional Chinese medicine diagnosis and AI models, the system includes: The receiving module is used to receive continuous physiological time-series data from the target user's smart wearable device and structured symptom and behavioral data input from the user terminal; The extraction module is used to clean and align continuous physiological time-series data and structured symptom and behavioral data, and extract Western medical psychological features and traditional Chinese medicine syndrome differentiation quantitative features related to gastric function. The analysis module is used to conduct in-depth correlation analysis of Western medical psychological characteristics and TCM syndrome differentiation quantitative characteristics using AI models, and output the probability distribution of gastric disease syndrome types and gastric health risk index. The generation module is used to generate graded early warning information based on the probability distribution of gastric disease syndrome types and the gastric health risk index according to the preset knowledge graph of gastric disease TCM syndrome differentiation, and output health intervention suggestions and medical treatment prompts to the target users based on the graded early warning information.

[0005] Preferably, the receiving module includes: The first receiving submodule is used to establish a communication connection with the smart wearable device and continuously receive the raw physiological time-series data stream collected in real time by the smart wearable device at a preset sampling frequency. The second receiving submodule is used to receive structured data actively input by the target user through the graphical interactive interface of the target user terminal when a preset abnormal physiological data pattern is detected in the user. The data encapsulation submodule is used to timestamp and encapsulate the received raw physiological time-series data stream and structured data; The upload submodule is used to upload the packaged data to the cloud-based intelligent analysis platform in real time via wireless network and store it in a temporary data buffer dedicated to the target user.

[0006] Preferably, the extraction module includes: The data processing submodule is used to evaluate the signal quality of continuous physiological time series data, remove invalid data segments caused by motion artifacts and signal interruptions, smooth the imputation of missing values, and perform logical consistency verification and standardized formatting of structured symptom and behavioral data. The generation submodule is used to synchronize and align the cleaned continuous physiological time series data and structured data with timestamps based on a unified time base, and generate a multimodal time series data stream with consistent time index. The extraction submodule is used to extract a set of Western medical physiological quantitative features that are directly related to gastrointestinal function and associated physiological states from a multimodal time-series data stream. The mapping submodule is used to map symptom descriptions, behavioral records, and specific pulse information identified from continuous physiological time series data into a structured set of TCM diagnostic quantitative features based on a preset TCM diagnostic quantitative rule library.

[0007] Preferably, the step of synchronizing and aligning the cleaned continuous physiological time-series data with structured data based on a unified time reference to generate a multimodal time-series data stream with consistent time indices includes: Determine the collection timestamp of each data point in the cleaned continuous physiological time series data and the recording timestamp of each record in the structured data; Identify the data type of structured data: if it is a continuous event that can be mapped to a continuous time period, extract its duration; if it is a discrete event, treat it as a point-in-time event. Based on the timeline of continuous physiological time series data, each structured data is converted into a time series feature vector with the same time resolution as the continuous physiological time series data according to the data type and record timestamp of each data. The temporal feature vectors are concatenated and associated with the corresponding continuous physiological time-series data streams along the time axis to generate a multimodal time-series data stream with consistent time indices.

[0008] Preferably, the pre-defined TCM diagnostic quantitative rule base maps symptom descriptions, behavioral records, and specific pulse information identified from continuous physiological time-series data into a structured TCM diagnostic quantitative feature set, including: Extracting symptom description text and behavioral record entries from structured data, and identifying specific pulse characteristic parameters from continuous physiological time series data; Based on a pre-defined TCM syndrome differentiation and quantification rule base, the symptom description text is parsed and mapped into a structured symptom syndrome vector containing disease location attributes, disease nature attributes, and disease intensity. Based on a pre-set TCM syndrome differentiation and quantification rule base, the type, frequency and time attributes of behavioral record items are analyzed to determine the influence index of type, frequency and time attributes on the cold and heat, deficiency and excess and dampness and dryness of the stomach and intestines, and to generate behavioral syndrome feature vectors. Based on a pre-set TCM diagnostic and quantitative rule library, specific pulse feature parameters are converted into TCM pulse syndrome indices that characterize the state of Qi, the strength and weakness of Qi and blood, and the degree of phlegm, dampness and food stagnation, thus generating pulse syndrome feature vectors. The structured symptom vector, behavioral syndrome feature vector, and pulse syndrome feature vector are aligned and merged along the time dimension to form a structured quantitative feature set for TCM syndrome differentiation.

[0009] Preferably, the analysis module includes: The computational submodule is used to extract high-level temporal representations of Western medicine's physiological features through the temporal encoder network in the AI ​​model, and to calculate the attention weight matrix between the high-level temporal representations and the quantitative features of TCM syndrome differentiation through the feature cross-attention network. The first output submodule is used to generate a fusion feature representation containing deep correlation information by weighted fusion of high-level temporal representation and TCM syndrome differentiation quantitative features based on the attention weight matrix. Based on the fusion feature representation, it outputs the probability distribution vector of multiple TCM syndrome types of gastric diseases. The second output submodule is used to output the target user's comprehensive gastric health risk index based on the probability distribution vectors of multiple TCM syndrome types of gastric diseases.

[0010] Preferably, the step of extracting high-level temporal representations of Western medicine psychological features through a temporal encoder network, and calculating the attention weight matrix between the high-level temporal representations and the quantitative features of traditional Chinese medicine syndrome differentiation through a feature cross-attention network, includes: Extract the time series of Western medicine pathological features and divide the time series into multiple fixed-length sliding time windows; The temporal encoder network is used to extract the last hidden state or the output of a specific pooling layer in each time window from the feature sequence in that time window, and use it as the high-level temporal representation corresponding to that time window. High-level temporal representations are used as query vector sequences, and TCM syndrome differentiation quantitative features are used as key vector sequences and value vector sequences, which are then input into a feature cross-attention network. The similarity score between the query vector sequence and the key vector sequence is calculated by using a feature cross-attention network and then normalized to generate an attention weight matrix.

[0011] Preferably, the generation module includes: The reasoning submodule is used to input the probability distribution of gastric disease syndrome types and the gastric health risk index into the preset gastric disease TCM syndrome differentiation knowledge graph, based on the stored syndrome-symptom-risk association rules, pathological evolution logic and intervention threshold, to perform logical reasoning and decision verification. The first determination submodule is used to determine the current risk level based on the reasoning results and generate structured early warning information for the current risk level; The second determination submodule is used to determine the risk level identifier, the explanation of the dominant TCM syndrome type, and the risk summary description based on the structured early warning information; The retrieval submodule is used to retrieve corresponding treatment methods and conditioning principles from a preset TCM syndrome differentiation knowledge graph of stomach diseases based on risk level identification, explanation of dominant TCM syndrome type, and risk summary description, and automatically synthesize personalized health intervention suggestions and medical treatment tips.

[0012] Preferably, the reasoning submodule, in the process of inputting the probability distribution of gastric disease syndrome types and the gastric health risk index into the preset gastric disease TCM syndrome differentiation knowledge graph based on stored syndrome-symptom-risk association rules, pathological evolution logic, and intervention thresholds, and performing logical reasoning and decision verification, also includes: The syndrome types whose probabilities exceed the first preset threshold in the syndrome type probability distribution, along with the health risk index, are matched with the predefined risk-syndrome combination rules in the preset stomach disease TCM syndrome differentiation knowledge graph. By utilizing the correlation between syndrome types and symptoms, pulse signs, and pathogenesis in the pre-set TCM syndrome differentiation knowledge graph for gastric diseases, the logical consistency between the syndrome type probability distribution output by the AI ​​model and the historical input symptom and behavioral data is verified. If there is a potential conflict between the AI ​​model output and the knowledge graph rules, the final risk level will be calibrated according to the pre-set strategy.

[0013] An early warning method for gastric diseases based on traditional Chinese medicine syndrome differentiation and AI models includes the following steps: Receives continuous physiological time-series data from smart wearable devices and structured symptom and behavioral data input from user terminals from target users; The continuous physiological time series data and structured symptom and behavior data were cleaned and aligned, and Western medical psychological features and traditional Chinese medicine syndrome differentiation quantitative features related to gastric function were extracted. Using AI models, we conducted a deep correlation analysis on the physiological characteristics of Western medicine and the quantitative characteristics of syndrome differentiation in traditional Chinese medicine, and output the probability distribution of syndrome types of gastric diseases and the gastric health risk index. Based on a pre-defined TCM knowledge graph for stomach diseases, graded early warning information is generated according to the probability distribution of stomach disease syndromes and the stomach health risk index. Based on the graded early warning information, health intervention suggestions and medical advice are output to the target users.

[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 an early warning system for gastric diseases based on traditional Chinese medicine diagnosis and AI model provided by the present invention. Figure 2 This is a schematic diagram of the extraction module in an early warning system for gastric diseases based on traditional Chinese medicine diagnosis and AI model provided by the present invention. Figure 3 This is a schematic diagram of the structure of the analysis module in an early warning system for gastric diseases based on traditional Chinese medicine diagnosis and AI model provided by the present invention. Figure 4 The flowchart illustrates the workflow of an early warning method for gastric diseases based on traditional Chinese medicine diagnosis and AI models 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] An early warning system for stomach diseases based on traditional Chinese medicine diagnosis and AI models, such as Figure 1 As shown, the system includes: The receiving module 101 is used to receive continuous physiological time-series data from the target user's smart wearable device and structured symptom and behavioral data input from the user terminal; The extraction module 102 is used to clean and align continuous physiological time-series data and structured symptom and behavior data, and extract Western medical psychological features and traditional Chinese medicine syndrome differentiation quantitative features related to gastric function. Analysis module 103 is used to conduct in-depth correlation analysis of Western medical psychological characteristics and TCM syndrome differentiation quantitative characteristics using AI models, and output the probability distribution of gastric disease syndrome types and gastric health risk index. The generation module 104 is used to generate graded early warning information based on the probability distribution of gastric disease syndrome types and the gastric health risk index according to the preset knowledge graph of gastric disease TCM syndrome differentiation, and output health intervention suggestions and medical treatment prompts to the target user based on the graded early warning information.

[0020] In this embodiment, the AI ​​model refers to a specific structure designed for fusing multimodal data from traditional Chinese medicine and Western medicine. Its deep correlation analysis specifically refers to the feature-level deep fusion and inference process achieved by capturing dynamic patterns of physiological signals through a temporal encoder network and dynamically calculating the correlation weights between physiological features and TCM syndrome features using a feature cross-attention network.

[0021] The working principle of the above technical solution is as follows: First, the receiving module receives continuous physiological time-series data from the target user's smart wearable device and structured symptom and behavioral data input from the user terminal; second, the extraction module cleans and aligns the continuous physiological time-series data and structured symptom and behavioral data, and extracts Western medical psychological features and TCM diagnostic quantitative features related to gastric function; then, based on the analysis module, an AI model is used to conduct deep correlation analysis on the Western medical psychological features and TCM diagnostic quantitative features, outputting the probability distribution of gastric disease syndromes and the gastric health risk index; finally, the generation module uses a preset gastric disease TCM syndrome knowledge graph to generate graded early warning information based on the probability distribution of gastric disease syndromes and the gastric health risk index, and outputs health intervention suggestions and medical treatment prompts to the target user based on the graded early warning information.

[0022] The beneficial effects of the above technical solution are as follows: By systematically integrating continuous physiological monitoring from smart wearable devices, symptom and behavioral data actively reported by users, in-depth analysis of Western medical principles and traditional Chinese medicine diagnostic characteristics based on AI, and intelligent decision-making and intervention output based on traditional Chinese medicine knowledge graphs, a single automated process is achieved. This realizes a seamless connection from data collection to personalized early warning and intervention, providing a continuous, non-invasive, and proactive daily management solution for gastric health. It addresses the problems mentioned in the background technology, such as the discrete nature of precision instrument examinations, the lack of continuous background support for early warnings, and the fact that while consumer-grade continuous health monitoring devices are convenient and non-invasive, they lack the ability to quantify and deeply analyze traditional Chinese medicine diagnostic methods for gastric diseases, resulting in low accuracy of the final analysis results and a reduced user experience.

[0023] In one embodiment, the receiving module includes: The first receiving submodule is used to establish a communication connection with the smart wearable device and continuously receive the raw physiological time-series data stream collected in real time by the smart wearable device at a preset sampling frequency. The second receiving submodule is used to receive structured data actively input by the target user through the graphical interactive interface of the target user terminal when a preset abnormal physiological data pattern is detected in the user. The data encapsulation submodule is used to timestamp and encapsulate the received raw physiological time-series data stream and structured data; The upload submodule is used to upload the packaged data to the cloud-based intelligent analysis platform in real time via wireless network and store it in a temporary data buffer dedicated to the target user.

[0024] In this embodiment, the raw physiological time-series data stream includes pulse wave waveform, heart rate, blood oxygen saturation, and body surface temperature; In this embodiment, the structured data includes self-assessment information of gastric symptoms based on standardized scales, dietary behavior logs, and medication records.

[0025] In this embodiment, the structured data actively input by the target user can be either text data or voice data. When it is voice data, the symptoms described by the target user's voice are converted into structured fields through voice input and natural language processing components.

[0026] In this embodiment, the preset abnormal physiological data pattern is represented as an abnormal trend pattern based on the deviation of gastric function-related physiological indicators from the individual's dynamic baseline, such as: persistent abnormality in the ratio of low to high frequency postprandial heart rate variability (HRV), disordered body surface temperature rhythm corresponding to the stomach area at night, or right Guan pulse waveform parameters (such as tension) that continuously exceed the individual's baseline range.

[0027] The beneficial effects of the above technical solution are as follows: by establishing a stable communication connection, actively triggering user input when physiological abnormalities are detected, and timestamping and encapsulating the data, the data used for subsequent analysis is ensured to have high timeliness, integrity and traceability, laying a solid data foundation for the system's real-time early warning capability, and realizing real-time and reliable acquisition and preprocessing of multi-source heterogeneous data.

[0028] In one embodiment, such as Figure 2 As shown, the extraction module 102 includes: The data processing submodule 1021 is used to evaluate the signal quality of continuous physiological time series data and remove invalid data segments caused by motion artifacts and signal interruptions, and to smoothly imput missing values, and to perform logical consistency verification and standardized formatting of structured symptom and behavioral data. The generation submodule 1022 is used to synchronize and align the cleaned continuous physiological time series data and structured data with timestamps based on a unified time base, and generate a multimodal time series data stream with consistent time index. Extraction submodule 1023 is used to extract a set of Western medical physiological quantitative features that are directly related to gastrointestinal function and associated physiological state from a multimodal time-series data stream; The mapping submodule 1024 is used to map symptom descriptions, behavioral records, and specific pulse information identified from continuous physiological time series data in structured data into a structured set of quantitative features for TCM diagnosis, based on a preset TCM diagnostic quantification rule library.

[0029] In this embodiment, the Western medicine physical quantitative feature set includes: Autonomic nervous system functional characteristics: Based on heart rate variability analysis, calculate the changes or trends of low-frequency power, high-frequency power and their ratio within a preset time window after eating for the target user; Specific pulse characteristics of the stomach region: Quantitative parameters of waveform morphology, including smoothness, tension, and fullness, are extracted from pulse wave signals processed for the right Guan pulse region. Circadian rhythm characteristics: Analyze the periodic fluctuation amplitude and regularity of the gastric region corresponding to the body surface temperature or microblood flow perfusion signal during specific periods at night.

[0030] In this embodiment, the quantitative feature set for TCM syndrome differentiation includes: Symptom vectorization mapping: The natural language symptom description input by the target user is mapped into a structured vector containing elements of lesion location, lesion nature, and lesion severity through semantic parsing and pattern matching; Behavioral quantification mapping: Analyze users' dietary logs and calculate and output dietary stagnation index, internal heat risk index, or cold and dampness predominance index based on food attributes and intake time; Pulse syndrome correlation mapping: The extracted stomach region-specific pulse characteristics are correlated and mapped to the Qi stagnation index, blood stasis index, or damp-heat tendency index according to the TCM pulse diagnosis theory.

[0031] In this embodiment, the continuous physiological time series data mainly comes from the photoplethysmography (PPG) signals collected by smart wearable devices (such as wrist devices). In particular, targeted analysis is performed on the right Guan pulse (the corresponding part of the stomach in traditional Chinese medicine). From the preprocessed right Guan pulse signal, the key feature points in each pulse cycle are automatically identified by the AI ​​pulse analysis algorithm. Based on the identified feature points, the stomach-specific pulse parameters are calculated. Through the preset TCM syndrome differentiation and quantification rule library, the stomach-specific pulse parameters are converted into a structured TCM syndrome index.

[0032] The beneficial effects of the above technical solution are as follows: through automated data cleaning, alignment and feature engineering, the raw and messy physiological and behavioral data are transformed into high-quality and structured analysis objects, effectively eliminating noise interference, unifying the time reference of multimodal data, and extracting Western medical physiological indicators that directly reflect the state of gastric function and standardized TCM syndrome features, providing clean, aligned and informative input for the accurate analysis of subsequent AI models.

[0033] In one embodiment, the step of synchronizing and aligning the cleaned continuous physiological time-series data with structured data based on a unified time reference to generate a multimodal time-series data stream with consistent time indices includes: Determine the collection timestamp of each data point in the cleaned continuous physiological time series data and the recording timestamp of each record in the structured data; Identify the data type of structured data: if it is a continuous event that can be mapped to a continuous time period, extract its duration; if it is a discrete event, treat it as a point-in-time event. Based on the timeline of continuous physiological time series data, each structured data is converted into a time series feature vector with the same time resolution as the continuous physiological time series data according to the data type and record timestamp of each data. The temporal feature vectors are concatenated and associated with the corresponding continuous physiological time-series data streams along the time axis to generate a multimodal time-series data stream with consistent time indices.

[0034] In this embodiment, for structured data of discrete events, a time tolerance window centered on the recording timestamp is determined on the time axis of continuous physiological time series data; The time-series feature vectors of discrete events are distributed to all or part of the physiological data time points within the tolerance window through reverse assignment or forward / backward padding, thus completing the mapping from discrete events to a continuous time axis. For structured data of continuous events, on the timeline of continuous physiological time series data, a period of time starting from the recorded timestamp and lasting for a specific duration is determined; The temporal feature vectors of persistent events are used to generate continuous or discrete feature sequences that change over time within a given period using an interpolation algorithm.

[0035] The beneficial effects of the above technical solution are as follows: by intelligently identifying data types and employing corresponding mapping algorithms, asynchronous and discrete user-reported events are successfully and accurately correlated with continuous physiological signals on the time axis. This solves the problem of asynchronous behavioral and physiological data in wearable health monitoring, and the generated multimodal time-series data stream can more realistically reflect the causal relationship between physiological events and behavioral factors, greatly improving the accuracy and reliability of subsequent correlation analysis.

[0036] In one embodiment, the preset TCM diagnostic quantitative rule base maps symptom descriptions, behavioral records, and specific pulse information identified from continuous physiological time-series data into a structured TCM diagnostic quantitative feature set, including: Extracting symptom description text and behavioral record entries from structured data, and identifying specific pulse characteristic parameters from continuous physiological time series data; Based on a pre-defined TCM syndrome differentiation and quantification rule base, the symptom description text is parsed and mapped into a structured symptom syndrome vector containing disease location attributes, disease nature attributes, and disease intensity. Based on a pre-set TCM syndrome differentiation and quantification rule base, the type, frequency and time attributes of behavioral record items are analyzed to determine the influence index of type, frequency and time attributes on the cold and heat, deficiency and excess and dampness and dryness of the stomach and intestines, and to generate behavioral syndrome feature vectors. Based on a pre-set TCM diagnostic and quantitative rule library, specific pulse feature parameters are converted into TCM pulse syndrome indices that characterize the state of Qi, the strength and weakness of Qi and blood, and the degree of phlegm, dampness and food stagnation, thus generating pulse syndrome feature vectors. The structured symptom vector, behavioral syndrome feature vector, and pulse syndrome feature vector are aligned and merged along the time dimension to form a structured quantitative feature set for TCM syndrome differentiation.

[0037] In this embodiment, the preset TCM syndrome differentiation quantitative rule base includes a multi-dimensional mapping relationship between symptom description, behavior record, pulse characteristics and TCM syndrome elements.

[0038] In this embodiment, the symptom description text is parsed and mapped into a structured symptom-syndrome vector containing lesion location attributes, lesion nature attributes, and lesion intensity, including: The symptom description text is segmented and keywords are extracted to identify the main complaint location, pain nature, triggering factors, and relieving factors. The identified elements are matched with the symptom-syndrome mapping table in the rule base to determine the corresponding combination of TCM syndrome elements; Adjust the quantitative values ​​of each syndrome element based on the intensity adverbs, frequency adverbs, and context of the symptom description; Output a multidimensional vector, where each dimension corresponds to a TCM syndrome element, and its value is the quantification intensity of that element.

[0039] In this embodiment, the type, frequency, and time attributes of behavioral record entries are analyzed to determine the influence index of type, frequency, and time attributes on the cold / heat, deficiency / excess, and dampness / dryness states of the stomach and intestines, generating a behavioral syndrome feature vector, including: Based on the dietary behavior records, the cold / hot index, rich / sweet index, and spicy / stimulating index of a single meal were calculated according to the properties of food in the classification of properties and meridians in traditional Chinese medicine. By combining the eating time and digestion cycle model, the curves of the above index decaying over time are generated; Based on the emotional and behavioral records, the system queries the mapping relationship of corresponding syndromes in the rule base according to the type and intensity of emotions, and outputs the Liver Qi Stagnation Index or the Heart Fire Excess Index.

[0040] In this embodiment, specific pulse characteristic parameters are converted into TCM pulse syndrome indices that characterize the state of Qi, the abundance or deficiency of Qi and blood, and the degree of phlegm-dampness and food stagnation, generating a pulse syndrome feature vector, including: Receive parameters of smoothness, tension, and fullness extracted from the right Guan pulse waveform; The slipperiness parameter is mapped to the phlegm-dampness index and food stagnation index through a piecewise linear function; The tension parameter is mapped to the stagnation index and pain index through a nonlinear transformation. The fullness parameter is compared with the target user's personal baseline and mapped to a Qi and Blood Deficiency Index or an Excess Heat Index.

[0041] The beneficial effects of the above technical solution are as follows: through a pre-set quantitative rule base, unstructured TCM symptom language, vague daily behaviors, and abstract pulse characteristics are transformed into unified, computable structured feature vectors. This achieves the digitization, standardization, and reusability of TCM diagnostic experience.

[0042] In one embodiment, such as Figure 3 As shown, the analysis module 103 includes: The computational submodule 1031 is used to extract high-level temporal representations of Western medicine pathological features through the temporal encoder network in the AI ​​model, and to calculate the attention weight matrix between the high-level temporal representations and the quantitative features of TCM syndrome differentiation through the feature cross-attention network. The first output submodule 1032 is used to generate a fusion feature representation containing deep correlation information by weighted fusion of high-level temporal representation and TCM syndrome differentiation quantitative features based on the attention weight matrix, and output the probability distribution vector of multiple TCM syndrome types of gastric diseases based on the fusion feature representation. The second output submodule 1033 is used to output the comprehensive gastric health risk index of the target user based on the probability distribution vectors of multiple TCM syndrome types of gastric diseases.

[0043] In this embodiment, the attention weight matrix is ​​used to quantify the correlation strength between each time point or feature dimension in the advanced time series representation and each syndrome element in the TCM syndrome differentiation quantitative features.

[0044] In this embodiment, based on the attention weight matrix, high-level temporal representation and TCM syndrome differentiation quantitative features are weighted and fused to generate a fused feature representation containing deep correlation information, including: The attention weight matrix is ​​used as a modulation signal to perform gating or weighting operations on the high-level temporal representation, thereby enhancing the physiological temporal features corresponding to highly correlated TCM syndrome elements. At the same time, after transposing the attention weight matrix, information filtering is performed on the quantitative features of TCM syndrome differentiation to highlight the syndrome elements that are highly concerned by the physiological temporal sequence. The enhanced features are concatenated or added element by element to generate a fused feature representation.

[0045] The beneficial effects of the above technical solution are as follows: By introducing an AI model that combines a temporal encoder network and a feature cross-attention network, it can go beyond simple feature superposition and deeply explore the complex and nonlinear mapping relationship between objective indicators of Western medicine and subjective symptoms of traditional Chinese medicine. This makes the output syndrome probability distribution and health risk index not only based on data statistics, but also contain pathophysiological insights from the integration of traditional Chinese and Western medicine, significantly improving the accuracy of early warning and medical interpretability.

[0046] In one embodiment, the step of extracting high-level temporal representations of Western medicine psychological features through a temporal encoder network, and calculating the attention weight matrix between the high-level temporal representations and the quantitative features of traditional Chinese medicine syndrome differentiation through a feature cross-attention network, includes: Extract the time series of Western medicine pathological features and divide the time series into multiple fixed-length sliding time windows; The temporal encoder network is used to extract the last hidden state or the output of a specific pooling layer in each time window from the feature sequence in that time window, and use it as the high-level temporal representation corresponding to that time window. High-level temporal representations are used as query vector sequences, and TCM syndrome differentiation quantitative features are used as key vector sequences and value vector sequences, which are then input into a feature cross-attention network. The similarity score between the query vector sequence and the key vector sequence is calculated by using a feature cross-attention network and then normalized to generate an attention weight matrix.

[0047] The beneficial effects of the above technical solution are as follows: by processing physiological time sequences through a sliding time window and utilizing an attention mechanism to calculate an attention weight matrix, the model can dynamically and selectively focus on the most relevant combinations of physiological and symptom characteristics at different time points. This mimics the comprehensive analytical thinking in traditional Chinese medicine diagnosis, making the model's analysis process more focused and flexible, and enabling it to more keenly capture subtle changes in gastric function and key risk signals.

[0048] In one embodiment, the generation module includes: The reasoning submodule is used to input the probability distribution of gastric disease syndrome types and the gastric health risk index into the preset gastric disease TCM syndrome differentiation knowledge graph, based on the stored syndrome-symptom-risk association rules, pathological evolution logic and intervention threshold, to perform logical reasoning and decision verification. The first determination submodule is used to determine the current risk level based on the reasoning results and generate structured early warning information for the current risk level; The second determination submodule is used to determine the risk level identifier, the explanation of the dominant TCM syndrome type, and the risk summary description based on the structured early warning information; The retrieval submodule is used to retrieve corresponding treatment methods and conditioning principles from a preset TCM syndrome differentiation knowledge graph of stomach diseases based on risk level identification, explanation of dominant TCM syndrome type, and risk summary description, and automatically synthesize personalized health intervention suggestions and medical treatment tips.

[0049] The beneficial effects of the above technical solution are as follows: By combining the data-driven results of the AI ​​model with the symbolic rules of the TCM syndrome differentiation knowledge graph for gastric diseases for collaborative decision-making, this ensures that the early warning decision is not only based on data probability, but also conforms to the internal logic of TCM theory and clinical experience. Through reasoning and retrieval of the knowledge graph, the system can automatically generate early warning information that integrates specific syndrome explanations, risk descriptions, and personalized conditioning or medical treatment suggestions, achieving an intelligent leap from risk identification to actionable guidance and improving the system's practicality.

[0050] In one embodiment, the reasoning submodule, in the process of inputting the probability distribution of gastric disease syndrome types and the gastric health risk index into a preset gastric disease TCM syndrome differentiation knowledge graph based on stored syndrome-symptom-risk association rules, pathological evolution logic, and intervention thresholds, and performing logical reasoning and decision verification, also includes: The syndrome types whose probabilities exceed the first preset threshold in the syndrome type probability distribution, along with the health risk index, are matched with the predefined risk-syndrome combination rules in the preset stomach disease TCM syndrome differentiation knowledge graph. By utilizing the correlation between syndrome types and symptoms, pulse signs, and pathogenesis in the pre-set TCM syndrome differentiation knowledge graph for gastric diseases, the logical consistency between the syndrome type probability distribution output by the AI ​​model and the historical input symptom and behavioral data is verified. If there is a potential conflict between the AI ​​model output and the knowledge graph rules, the final risk level will be calibrated according to the pre-set strategy.

[0051] In this embodiment, the preset strategy is based on the knowledge graph.

[0052] The beneficial effects of the above technical solution are as follows: when the data-driven AI output has a potential conflict with the medical logic fixed in the knowledge graph, the system can perform calibration or labeling, effectively preventing false warnings caused by model misjudgment or data noise, and greatly improving the robustness, security and credibility of the system's decision-making.

[0053] In one embodiment, this embodiment discloses an early warning method for gastric diseases based on traditional Chinese medicine syndrome differentiation and an AI model, such as... Figure 4 As shown, it includes the following steps: Step S401: Receive continuous physiological time-series data from the target user's smart wearable device and structured symptom and behavioral data input from the user terminal; Step S402: Clean and align the continuous physiological time series data and structured symptom and behavior data, and extract the Western medical psychological features and traditional Chinese medicine syndrome differentiation quantitative features related to gastric function. Step S403: Use AI models to conduct in-depth correlation analysis on the psychological characteristics of Western medicine and the quantitative characteristics of TCM syndrome differentiation, and output the probability distribution of gastric disease syndromes and the gastric health risk index. Step S404: Based on the preset knowledge graph of TCM syndrome differentiation for stomach diseases, generate graded early warning information according to the probability distribution of stomach disease syndromes and stomach health risk index, and output health intervention suggestions and medical treatment prompts to the target user based on the graded early warning information.

[0054] The working principle and beneficial effects of the above technical solution have been explained in the method embodiments, and will not be repeated here.

[0055] Those skilled in the art should understand that the terms "first" and "second" in this invention simply refer to different application stages.

[0056] 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.

[0057] 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. An early warning system for gastric diseases based on traditional Chinese medicine syndrome differentiation and AI models, characterized in that, The system includes: The receiving module is used to receive continuous physiological time-series data from the target user's smart wearable device and structured symptom and behavioral data input from the user terminal; The extraction module is used to clean and align continuous physiological time-series data and structured symptom and behavioral data, and extract Western medical psychological features and traditional Chinese medicine syndrome differentiation quantitative features related to gastric function. The analysis module is used to conduct in-depth correlation analysis of Western medical psychological characteristics and TCM syndrome differentiation quantitative characteristics using AI models, and output the probability distribution of gastric disease syndromes and gastric health risk index. The generation module is used to generate graded early warning information based on the probability distribution of gastric disease syndrome types and the gastric health risk index according to the preset knowledge graph of gastric disease TCM syndrome differentiation, and output health intervention suggestions and medical treatment prompts to the target users based on the graded early warning information.

2. The early warning system for gastric diseases based on traditional Chinese medicine syndrome differentiation and AI model according to claim 1, characterized in that, The receiving module includes: The first receiving submodule is used to establish a communication connection with the smart wearable device and continuously receive the raw physiological time-series data stream collected in real time by the smart wearable device at a preset sampling frequency. The second receiving submodule is used to receive structured data actively input by the target user through the graphical interactive interface of the target user terminal when a preset abnormal physiological data pattern is detected in the user. The data encapsulation submodule is used to timestamp and encapsulate the received raw physiological time-series data stream and structured data; The upload submodule is used to upload the packaged data to the cloud-based intelligent analysis platform in real time via wireless network and store it in a temporary data buffer dedicated to the target user.

3. The early warning system for gastric diseases based on traditional Chinese medicine syndrome differentiation and AI model according to claim 1, characterized in that, The extraction module includes: The data processing submodule is used to evaluate the signal quality of continuous physiological time series data, remove invalid data segments caused by motion artifacts and signal interruptions, smooth the imputation of missing values, and perform logical consistency verification and standardized formatting of structured symptom and behavioral data. The generation submodule is used to synchronize and align the cleaned continuous physiological time series data and structured data with timestamps based on a unified time base, and generate a multimodal time series data stream with consistent time index. The extraction submodule is used to extract a set of Western medical physiological quantitative features that are directly related to gastrointestinal function and associated physiological states from a multimodal time-series data stream. The mapping submodule is used to map symptom descriptions, behavioral records, and specific pulse information identified from continuous physiological time series data into a structured set of TCM diagnostic quantitative features based on a preset TCM diagnostic quantitative rule library.

4. The early warning system for gastric diseases based on traditional Chinese medicine syndrome differentiation and AI model according to any one of claims 2-3, characterized in that, The process, based on a unified time reference, involves synchronizing and aligning the cleaned continuous physiological time-series data with structured data using timestamps, generating a multimodal time-series data stream with consistent time indices, including: Determine the collection timestamp of each data point in the cleaned continuous physiological time series data and the recording timestamp of each record in the structured data; Identify the data type of structured data: if it is a continuous event that can be mapped to a continuous time period, extract its duration; if it is a discrete event, treat it as a point-in-time event. Based on the timeline of continuous physiological time series data, each structured data is converted into a time series feature vector with the same time resolution as the continuous physiological time series data according to the data type and record timestamp of each data. The temporal feature vectors are concatenated and associated with the corresponding continuous physiological time-series data streams along the time axis to generate a multimodal time-series data stream with consistent time indices.

5. The early warning system for gastric diseases based on traditional Chinese medicine syndrome differentiation and AI model according to claim 3, characterized in that, The pre-defined TCM diagnostic quantitative rule base maps symptom descriptions, behavioral records, and specific pulse information identified from continuous physiological time-series data in structured data into a structured TCM diagnostic quantitative feature set, including: Extracting symptom description text and behavioral record entries from structured data, and identifying specific pulse characteristic parameters from continuous physiological time series data; Based on a pre-defined TCM syndrome differentiation and quantification rule base, the symptom description text is parsed and mapped into a structured symptom syndrome vector containing disease location attributes, disease nature attributes, and disease intensity. Based on a pre-set TCM syndrome differentiation and quantification rule base, the type, frequency and time attributes of behavioral record items are analyzed to determine the influence index of type, frequency and time attributes on the cold and heat, deficiency and excess and dampness and dryness of the stomach and intestines, and to generate behavioral syndrome feature vectors. Based on a pre-set TCM diagnostic and quantitative rule library, specific pulse feature parameters are converted into TCM pulse syndrome indices that characterize the state of Qi, the strength and weakness of Qi and blood, and the degree of phlegm, dampness and food stagnation, thus generating pulse syndrome feature vectors. The structured symptom vector, behavioral syndrome feature vector, and pulse syndrome feature vector are aligned and merged along the time dimension to form a structured quantitative feature set for TCM syndrome differentiation.

6. The early warning system for gastric diseases based on traditional Chinese medicine syndrome differentiation and AI model according to claim 1, characterized in that, The analysis module includes: The computational submodule is used to extract high-level temporal representations of Western medicine's physiological features through the temporal encoder network in the AI ​​model, and to calculate the attention weight matrix between the high-level temporal representations and the quantitative features of TCM syndrome differentiation through the feature cross-attention network. The first output submodule is used to generate a fusion feature representation containing deep correlation information by weighted fusion of high-level temporal representation and TCM syndrome differentiation quantitative features based on the attention weight matrix. Based on the fusion feature representation, it outputs the probability distribution vector of multiple TCM syndrome types of gastric diseases. The second output submodule is used to output the comprehensive gastric health risk index of the target user based on the probability distribution vectors of multiple TCM syndrome types of gastric diseases.

7. The early warning system for gastric diseases based on traditional Chinese medicine syndrome differentiation and AI model according to claim 6, characterized in that, The process involves extracting high-level temporal representations of Western medicine's physiological features using a temporal encoder network, and then calculating the attention weight matrix between these high-level temporal representations and the quantitative features of traditional Chinese medicine's syndrome differentiation using a feature cross-attention network. This includes: Extract the time series of Western medicine pathological features and divide the time series into multiple fixed-length sliding time windows; The temporal encoder network is used to extract the last hidden state or the output of a specific pooling layer in each time window from the feature sequence in that time window, and use it as the high-level temporal representation corresponding to that time window. High-level temporal representations are used as query vector sequences, and TCM syndrome differentiation quantitative features are used as key vector sequences and value vector sequences, which are then input into a feature cross-attention network. The similarity score between the query vector sequence and the key vector sequence is calculated by using a feature cross-attention network and then normalized to generate an attention weight matrix.

8. The early warning system for gastric diseases based on traditional Chinese medicine syndrome differentiation and AI model according to claim 1, characterized in that, The generation module includes: The reasoning submodule is used to input the probability distribution of gastric disease syndrome types and the gastric health risk index into the preset gastric disease TCM syndrome differentiation knowledge graph, based on the stored syndrome-symptom-risk association rules, pathological evolution logic and intervention threshold, to perform logical reasoning and decision verification. The first determination submodule is used to determine the current risk level based on the reasoning results and generate structured early warning information for the current risk level; The second determination submodule is used to determine the risk level identifier, the explanation of the dominant TCM syndrome type, and the risk summary description based on the structured early warning information; The retrieval submodule is used to retrieve corresponding treatment methods and conditioning principles from a preset TCM syndrome differentiation knowledge graph of stomach diseases based on risk level identification, explanation of dominant TCM syndrome type, and risk summary description, and automatically synthesize personalized health intervention suggestions and medical treatment tips.

9. The early warning system for gastric diseases based on traditional Chinese medicine syndrome differentiation and AI model according to claim 8, characterized in that, The reasoning submodule, in the process of inputting the probability distribution of gastric disease syndrome types and the gastric health risk index into the preset gastric disease TCM syndrome differentiation knowledge graph, based on the stored syndrome-symptom-risk association rules, pathological evolution logic, and intervention thresholds, and performing logical reasoning and decision verification, also includes: The syndrome types whose probabilities exceed the first preset threshold in the syndrome type probability distribution, along with the health risk index, are matched with the predefined risk-syndrome combination rules in the preset stomach disease TCM syndrome differentiation knowledge graph. By utilizing the correlation between syndrome types and symptoms, pulse signs, and pathogenesis in the pre-set TCM syndrome differentiation knowledge graph for gastric diseases, the logical consistency between the syndrome type probability distribution output by the AI ​​model and the historical input symptom and behavioral data is verified. If there is a potential conflict between the AI ​​model output and the knowledge graph rules, the final risk level will be calibrated according to the pre-set strategy.

10. A method for early warning of gastric diseases based on traditional Chinese medicine syndrome differentiation and AI model, characterized in that, Includes the following steps: Receives continuous physiological time-series data from smart wearable devices and structured symptom and behavioral data input from user terminals from target users; The continuous physiological time series data and structured symptom and behavior data were cleaned and aligned, and Western medical psychological features and traditional Chinese medicine syndrome differentiation quantitative features related to gastric function were extracted. Using AI models, we conducted a deep correlation analysis on the physiological characteristics of Western medicine and the quantitative characteristics of syndrome differentiation in traditional Chinese medicine, and output the probability distribution of syndrome types of gastric diseases and the gastric health risk index. Based on a pre-defined TCM knowledge graph for stomach diseases, graded early warning information is generated according to the probability distribution of stomach disease syndromes and the stomach health risk index. Based on the graded early warning information, health intervention suggestions and medical advice are output to the target users.

Images