Healthcare intelligent medical method based on multi-modal wearable perception and artificial intelligence

Abstract

The application discloses a health intelligent medical method based on multi-modal wearable sensing and artificial intelligence, and comprises the following steps: collecting multi-modal physiological data and modeling to generate a health state vector, and measuring the stability of the physiological state; according to the health state vector, identifying a crisis state and actively warning based on time series prediction; determining whether the medication and the medical order are consistent by uploading the medical order and the drug information; and automatically classifying chronic diseases based on a special disease management library. The application realizes automatic identification and classified management of chronic disease patients by constructing a special disease management library and combining health state vector modeling, can automatically match the corresponding disease management category according to the long-term physiological index characteristics and health state changes of the user, and loads the corresponding management strategy. The method supports dynamic updating and long-term follow-up management of the disease state, and upgrades chronic disease management from static label management to a continuous management mode based on real health data.

Description

Technical Field

[0001] This invention belongs to the field of intelligent medical information systems and artificial intelligence medical engineering technology, specifically relating to a health and intelligent medical method based on multimodal wearable sensing and artificial intelligence. Background Technology

[0002] With the accelerating aging of the population and the continuous rise in the prevalence of chronic non-communicable diseases (such as hypertension, diabetes, and cardiovascular diseases), the traditional hospital-centered, single-visit-based medical model is no longer sufficient to meet the needs of long-term, continuous, and personalized health management. Numerous studies have shown that the occurrence of chronic disease complications and acute events is often closely related to the lack of outpatient monitoring, non-standard medication use, and failure to intervene in risk states in a timely manner.

[0003] The existing technology has the following main shortcomings: 1. Physiological data acquisition is limited in scope and lacks continuity. Current health monitoring methods mostly rely on single devices (such as blood pressure monitors or blood glucose meters), with low data collection frequency, failing to reflect the dynamic changes in the human body's physiological state, especially in nighttime or in situations where monitoring is not readily apparent.

[0004] 2. Fragmented data from multiple devices, lacking a unified modeling and fusion analysis mechanism. Although wearable watches, rings, glasses, and other devices are widely available, they mostly operate as isolated devices with inconsistent data standards, making it difficult to perform system-level fusion modeling and resulting in one-sided health assessment results.

[0005] 3. Medication management relies primarily on reminders, lacking safety verification and intelligent decision-making. Existing medication reminder systems mostly remain at the level of timed check-ins, failing to proactively identify and intervene in high-risk behaviors such as drug mixing, overdosing, and conflicting medical orders.

[0006] 4. Delayed identification of crisis states and passive intervention mechanisms. Physiological crises such as sudden rises in blood pressure and abnormal blood sugar are often only discovered after symptoms become obvious, lacking early warning mechanisms based on trend prediction.

[0007] 5. Disease-specific management lacks systematic knowledge support. Management strategies for different diseases vary significantly, and existing systems have failed to build scalable disease-specific knowledge bases, making it difficult to achieve disease classification, long-term follow-up, and personalized intervention.

[0008] The development of multimodal wearable sensing technology, cloud computing, artificial intelligence, temporal modeling, and medical knowledge graphs has provided a technological foundation for building a new generation of continuous health management systems. Therefore, there is an urgent need for an integrated intelligent health and medical system capable of multi-device collaborative sensing, medication safety control, crisis prediction, and intelligent management of specific diseases.

[0009] In the prior art, a wearable health monitoring-based early warning system (CN109157202A) discloses a system architecture including a multi-physiological signal acquisition device, a transmission device, and a cloud server. It can collect multiple physiological signals from the subject and transmit the data to the cloud for feature fusion and judgment, thereby achieving disease early warning. Although the above-mentioned prior art has achieved the acquisition and preliminary fusion of multi-physiological data, it still has the following significant shortcomings in practical continuous health management applications: Crisis identification is delayed and lacks proactive trend prediction capabilities. Its early warning mechanism mainly relies on whether current physiological data exceeds a fixed safety threshold, essentially acting as a reactive alarm. It lacks the ability to model the dynamic trends of long-term time-series data of physiological indicators, failing to identify potential crises during the evolutionary phase before blood pressure or blood sugar is about to spiral out of control, thus missing the optimal opportunity for proactive intervention.

[0010] There is a complete lack of medication safety intervention and medical order consistency verification mechanisms. Existing technologies are limited to one-way monitoring of physiological signals, which disrupts the most crucial aspect of chronic disease management: medication behavior control. Current systems cannot intelligently identify and safely intercept patients' actual medication behavior, and therefore cannot form a safe medical closed loop of "monitoring-early warning-medication intervention".

[0011] Disease-specific management capabilities are rigid and lack adaptive classification and dynamic follow-up mechanisms. Existing technologies typically only set fixed monitoring logic for a single disease, lacking a standardized and dynamically expandable disease-specific management rule base. When faced with patients with complex chronic diseases or comorbidities, existing systems cannot automatically match disease features and intelligently classify them based on the user's long-term health status vector, making it difficult to output personalized management strategies with long-term follow-up value. This invention addresses these pain points by proposing a novel intelligent treatment method that uses multimodal perception as the entry point and AI prediction and medication safety as the core decision-making engines.

[0012] In existing technology, a smart follow-up management system based on a medical system (CN112466446A) discloses modules such as an IoT detection platform, medication reminders, intelligent follow-up plan formulation, and a chronic disease knowledge base. It can upload the patient's physical data at home via the IoT and provide medication reminders and intervention tracking based on the patient's chronic disease records. Although it achieves IoT-based chronic disease data uploading and basic medication follow-up, it has significant shortcomings in the security depth and dynamic adaptability of intelligent decision-making. Medication reminders lack safety interception and doctor's order consistency verification mechanisms. They primarily rely on routine medication plan pushes, lacking safety logic verification of medication behavior. They cannot perform structured drug component conflict (incompaniment) detection or daily total dose exceedance determination before / during medication administration. For implicit deviations between actual medication use and doctor's orders, this existing technology cannot perform closed-loop identification and hazard warning, greatly increasing the risk of medication safety incidents for patients outside of hospitals.

[0013] Chronic disease intervention strategies lack the dynamic evolution capability based on multimodal health state vectors. Existing intervention schemes rely more on static chronic disease knowledge bases and manual doctor-patient interactions. They lack the core mechanism of this invention—constructing a health state vector by continuously extracting multimodal features and automatically calculating disease category matching scores. This results in lagging and rigid intervention strategies, unable to adaptively update and load corresponding disease-specific management strategies when a patient's health status just begins to deteriorate. Based on the limitations of existing technologies, this invention proposes a closed-loop method based on multimodal wearable sensing and artificial intelligence decision-making, truly filling the key technological gap in outpatient chronic disease management for trend crisis prediction and intelligent verification of medication safety. Summary of the Invention

[0014] The purpose of this invention is to overcome the problems of fragmented monitoring, unsafe medication, delayed crisis identification, and insufficient disease management capabilities in existing technologies. It proposes a smart health care method and system that uses multimodal wearable devices as the sensing entry point, artificial intelligence as the core decision-making engine, and medication safety and risk intervention as the core objectives. This smart health care system and its implementation method integrate multi-form wearable physiological monitoring devices, a cloud-based artificial intelligence analysis engine, medication safety management, and intelligent intervention mechanisms for specific diseases. It enables continuous monitoring, intelligent analysis, proactive intervention, and closed-loop management of users' health status, and is suitable for continuous health management and risk warning for patients with chronic diseases, sub-healthy individuals, and high-risk groups.

[0015] The present invention is achieved by at least one of the following technical solutions.

[0016] A health and intelligent medical approach based on multimodal wearable sensing and artificial intelligence includes the following steps: S1. Collect multimodal physiological data and model to generate health status vectors, and measure the stability of physiological status; S2. Based on the health status vector, identify crisis states and proactively warn of potential emergencies using time-series prediction; S3. By uploading medical orders and medication information, determine whether the medication is consistent with the medical orders; S4. Automatically classify chronic diseases based on the disease management database.

[0017] Furthermore, the modeling in step 1 generates a vector reflecting the user's overall health status at a certain point in time through linear mapping and nonlinear activation functions.

[0018] Furthermore, in step S1, the stability of physiological state is measured by the health status change rate index. When the health status change rate index exceeds the individualized threshold, it is determined that there is abnormal fluctuation in health status.

[0019] Furthermore, in step S2, identifying a crisis state includes the following steps: First, a long short-term memory network is used to model the time series of physiological indicators in order to capture the characteristics of long-term dependence and short-term fluctuations and output the predicted values ​​of future time steps. Then, a crisis risk score is calculated: the predicted value is compared with the individual safety baseline to obtain a risk score; when the risk score is greater than or equal to the threshold, a potential crisis state is determined.

[0020] Further, step S3 includes the following steps: (31) After OCR and information extraction, each drug is represented as a structured object:

[0021] in, Represents the structured feature object of the j-th drug. Indicates the name of the drug. Indicates a single dose. Indicates the frequency of use. Indicates the duration of administration; (32) Determine the total dosage of medication and whether it is an overdose: Calculate the total actual dosage of medication taken per unit time. :

[0022] Let the drug The safe maximum dose is ,when:

[0023] The risk of overdose was identified. (33) Identify whether drugs are mixed or have conflicting ingredients: Introducing a collection of drug components express:

[0024] When two drugs are present ,satisfy:

[0025] This is identified as a risk of duplicate drug use; Meanwhile, based on the pharmacological action rule base, if the following exists:

[0026] This triggers a drug interaction warning, in which, , represents a drug interaction discriminant function based on a pharmacological action rule base, used to evaluate drug components. Are there any pharmacological conflicts or contraindications between them? (34) Use the medical order consistency verification model to determine whether there is a deviation in medical order execution: Representing users' actual medication use behavior as a time function:

[0027] In the formula Indicates that the user is Always be aware of the drug Actual medication use behavior status indicator value; when the user is... When performing the medication action at all times, The value is 1, if the user is... If medication is not taken for a period of time, then The value is 0; The prescribed medication time is as follows ,when:

[0028] The system then determines that there is a deviation in the execution of medical orders, among which This is the compliance threshold; (35) Comprehensive risk score for combined drug use: Multiple risk categories are combined into a unified score:

[0029] in, This indicates the overall risk score for combined drug use. For the first Risk indicators For the corresponding weights; when When a medication risk warning is triggered, This indicates the system's preset threshold for triggering medication risk alerts.

[0030] Furthermore, in step S4, based on the user's health status vector and historical diagnostic information, the user is matched for chronic diseases using a disease matching function, and the highest matching score is selected as the disease category.

[0031] Furthermore, the disease management database is constructed as follows: Each type of disease is represented as a structured object as follows:

[0032] in, Indicates the first in the disease management database Structured feature objects of chronic diseases Indicates the disease name and code. This represents a set of key physiological indicators. Indicates the risk indicator threshold range, This represents a set of recommended management strategies.

[0033] The system for implementing the aforementioned health and intelligent medical method based on multimodal wearable sensing and artificial intelligence includes: The data acquisition module is used to collect multimodal physiological signals and generate individualized health status vectors; The crisis prediction module is used to identify crisis states and proactively issue warnings based on individualized health status vectors; The medication detection module is used to intelligently determine whether medication safety is consistent with the doctor's orders; The chronic disease classification module is used to automatically classify users' chronic diseases. The AI ​​medical assistant module is used to answer patients' medication-related questions, explain the reasons for risk warnings, help patients understand medical advice, and provide non-diagnostic health management advice.

[0034] A computer device according to the present invention includes a memory and a processor, the memory being electrically connected to the processor, the memory storing a computer program, which, when executed by the processor, causes the processor to implement the method described herein.

[0035] The present invention provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor implements the method described herein.

[0036] Compared with the prior art, the present invention has at least the following significant advantages: 1. Achieving Continuous Health Monitoring Through Multi-Device Collaboration. Compared to existing health monitoring solutions that rely on a single wearable device or single-point measurement method, this invention achieves multi-channel, distributed, and continuous acquisition of key physiological indicators by collaborating with various wearable devices such as watches, rings, wearable glasses, and necklaces. The complementarity of different devices in terms of wearing location, sampling frequency, and applicable scenarios enables the system to maintain stable data acquisition capabilities in various life scenarios such as user exercise, rest, and sleep, significantly reducing the risk of data interruption caused by single device detachment, obstruction, or signal loss. Through multimodal data fusion modeling, this invention effectively improves the continuity, reliability, and noise resistance of health monitoring data, providing a high-quality data foundation for subsequent health assessment and risk prediction.

[0037] 2. Shifting from reactive alerts to predictive crisis intervention. Existing health monitoring systems mostly employ reactive alert mechanisms based on fixed thresholds, triggering alerts only when physiological indicators are clearly abnormal, often missing the optimal intervention window. This invention introduces time-series modeling and trend prediction algorithms to model and analyze the changing trends of key physiological indicators such as blood pressure and blood sugar. It can identify potential abnormal evolution risks before these indicators cross safe thresholds. This predictive crisis identification mechanism enables the system to trigger tiered warnings and intervention recommendations in advance, shifting from "passive response" to "proactive prevention," effectively reducing the probability of acute risk events (such as sudden increases in blood pressure or uncontrolled blood sugar), and significantly enhancing the safety value of the health management system in practical applications.

[0038] 3. Medication management is upgraded from "reminders" to "safe decision-making." The system not only reminds users of medication times but also identifies drug mixing, overdosing, and deviations from prescriptions, effectively reducing medication safety risks. Compared to existing medication management solutions that only provide timed reminders, this invention introduces pharmacological rule verification and prescription consistency analysis mechanisms into the medication management process, achieving a comprehensive assessment of the safety, rationality, and compliance of medication use. The system not only provides medication time reminders according to prescriptions but also automatically identifies drug mixing, duplicate ingredients, dosage exceeding limits, and deviations between actual medication use and prescriptions, promptly alerting users when potential risks are detected. This mechanism effectively compensates for patients' lack of pharmaceutical knowledge, significantly reducing health risks caused by mis-dosing, missed doses, or incorrect medication use.

[0039] 4. Automated Classification and Long-Term Follow-up for Disease-Specific Management. This invention achieves automated identification and classification management of chronic disease patients by constructing a disease-specific management database and combining it with health status vector modeling. The system can automatically match the corresponding disease management category and load the appropriate management strategy based on the user's long-term physiological indicators and changes in health status. Compared with existing management methods that rely on manual judgment or single diagnostic labels, this method supports dynamic updates of disease status and long-term follow-up management, upgrading chronic disease management from static label management to a continuous management model driven by real health data, significantly improving the targeting and effectiveness of management.

[0040] 5. AI Medical Assistant Improves Accessibility to Healthcare Services but Does Not Replace Doctors. The AI ​​medical assistant in this invention is strictly designed and constrained by a medical knowledge base and safety rules. Its function is positioned as explanatory and auxiliary support, rather than a replacement for diagnostic or treatment decisions. Based on the user's current health status, medication history, and risk warnings, the system can provide clear and understandable explanations and consultation suggestions. This approach effectively alleviates the problem of patients having "no one to ask and no way to understand" in out-of-hospital settings, reduces the probability of errors due to information asymmetry, and significantly improves the accessibility of healthcare services and the user experience, while avoiding medical liability risks.

[0041] 6. The system forms a closed loop of "monitoring-analysis-intervention-feedback." It constructs a data-driven intelligent medical management model, significantly improving patient compliance and management efficiency. Through the organic combination of multimodal perception, intelligent analysis, proactive intervention, and behavioral feedback mechanisms, this invention constructs a complete data-driven closed-loop health management system. The system can transform continuous monitoring data into actionable intervention strategies and continuously refine these strategies through medication check-ins, risk alerts, and health feedback. Compared to existing fragmented, one-way output health management systems, this closed-loop model significantly improves patients' understanding and adherence to management plans, effectively enhancing medication compliance and health management efficiency, and possesses significant long-term application value. Attached Figure Description

[0042] Figure 1 This is a flowchart illustrating a smart healthcare method based on multimodal wearable sensing and artificial intelligence, as an example.

[0043] Figure 2 This is a flowchart illustrating AI-based health risk prediction and tiered early warning.

[0044] Figure 3 This is a closed-loop logic diagram of intelligent medication safety management and intervention for an example. Detailed Implementation

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] like Figures 1-3 As shown in this embodiment, a smart healthcare method based on multimodal wearable sensing and artificial intelligence includes the following steps: 1. Health status modeling using multimodal wearable physiological signals, including: (1) Unified representation of multi-source physiological signals Assume the user is in a time window within, from the first The first data collected by wearable devices Physiological signals are denoted as:

[0047] in, Indicates the device type. Indicates the type of physiological signal. Indicates the specific equipment type number, This indicates the total number of categories of wearable devices connected to the system; This indicates the specific type of physiological signal number. This indicates the total number of types of physiological signals collected by the system; Indicates time window The first i A specific sampling time point, This indicates the last sampling time point within the time window (i.e., the end time of the time window). Indicates the sampling time point At that time, by the first k The first data collected by wearable devices Raw observation values ​​of physiological signals.

[0048] In one embodiment, a smartwatch is used to collect heart rate / blood oxygen / motion signals; a smart ring is used to collect body temperature / sleep / hemodynamic signals. Signals; smart glasses / necklaces collect attitude / environment perception signals.

[0049] To eliminate dimensional differences between different devices and sensors, the system standardizes the raw signal:

[0050] in , These are the historical statistical mean and standard deviation, respectively. This represents the dimensionless physiological signal characteristic value obtained after standardization.

[0051] (2) Extracting and fusing time series features.

[0052] For each type of physiological signal, within a sliding time window Internal extraction of statistical and dynamic features, including but not limited to: Mean characteristics, variance characteristics, first-order difference trend, volatility index.

[0053] Taking heart rate signals as an example, their trend characteristics can be expressed as follows:

[0054] in, Indicates that the user is Real-time heart rate measurement; Indicates that the user is The first-order difference of heart rate at any given moment (i.e., the change in heart rate between the current moment and the previous moment) is used to quantify the instantaneous fluctuation trend of heart rate.

[0055] The system concatenates the features of each signal to form a unified multimodal feature vector. :

[0056] in This represents the total number of dimensions (i.e., the total number of features) of the extracted multimodal physiological features. Indicates in Multimodal feature vectors constructed at each time step The first in The last (i.e., the last) specific feature component value.

[0057] (3) Construct individualized health status vectors.

[0058] To reflect a user's overall health status at a specific point in time, a health status vector is introduced. Generated through linear mapping and nonlinear activation functions:

[0059] in For learnable weight matrix, For bias terms, It is a non-linear activation function.

[0060] This health status vector serves as a unified input for subsequent crisis prediction, medication risk assessment, and disease management.

[0061] (4) Health status stability and abnormality measurement.

[0062] Define a health status change rate index to measure the stability of physiological state:

[0063] in, Indicates in The rate of change in health status is calculated at any given time. This indicator is obtained by calculating the health status vector at the current time. Compared with the health state vector at the previous time step The Euclidean distance in the multidimensional feature space is used to quantify the fluctuation and stability of a user's physiological state at adjacent time points. The higher the value, the more severe the abnormal fluctuations in physiological state.

[0064] when Exceeding the individualization threshold When the system detects abnormal fluctuations in health status, it provides a priori triggering conditions for crisis state prediction and proactive early warning in the subsequent step 2.

[0065] 2. Identifying crisis states and providing proactive early warnings based on time-series predictions, including the following steps: (21) Define a crisis state.

[0066] In this invention, a crisis state refers to a user's key physiological indicators showing an abnormal upward trend or exceeding the individual's safe range within a short period of time, including but not limited to: abnormally high systolic / diastolic blood pressure, and sustained rise or rapid fluctuation in blood sugar levels.

[0067] Unlike traditional methods based on single-point thresholds, this invention emphasizes a trend-driven predictive recognition mechanism.

[0068] (22) Time series modeling of physiological indicators.

[0069] Suppose the time series of a key physiological indicator (such as blood pressure or blood glucose) is as follows:

[0070] in, This represents the actual measured value of a key physiological indicator (such as blood pressure or blood sugar) at the nth time step (i.e., the last observation time of the time series). Generally, This represents the measured value of the physiological indicator at time t.

[0071] Based on sliding time window Constructing feature sequences :

[0072] A Long Short-Term Memory (LSTM) network was used as the prediction model to model this time series, capturing long-term dependencies and short-term fluctuations. The LSTM state update process is as follows:

[0073] in, and These represent the hidden state vectors at the current time step t and the previous time step t-1, respectively. and These represent the memory cell state vectors of the current and previous time steps, respectively; This represents the physiological indicator input feature vector at time step t. , , These represent the activation and output vectors of the Forget Gate, Input Gate, and Output Gate, respectively. This represents the state vector of the candidate memory cells at the current time step. , , These represent the weight matrices corresponding to the forget gate, input gate, output gate, and candidate memory units, respectively. , , , These represent the corresponding bias term vectors. This represents the Sigmoid activation function. This represents the hyperbolic tangent activation function. This represents the element-wise multiplication operator (Hadamard product).

[0074] Long Short-Term Memory (LSTM) networks output predictions for future time steps. :

[0075] in, This represents the output function that maps the hidden layer state to the predicted value of a specific physiological indicator (e.g., a mapping network composed of fully connected layers). Let represent the hidden state vector computed by the Long Short-Term Memory (LSTM) network at the i-th time step, which encapsulates the historical temporal features up to the current time step.

[0076] (23) Calculate the crisis risk score.

[0077] A risk scoring function is constructed by comparing the predicted values ​​with an individual's safety baseline.

[0078] in , For the user's historical safe range parameters, Indicates the rate of change of the trend. These are the weighting coefficients. This indicates the system's prediction of future time steps. The calculated comprehensive crisis risk score quantifies the extent to which the predicted value deviates from an individual's safety baseline and the rate of its deterioration.

[0079] when:

[0080] in, This represents the system's preset risk trigger threshold for determining a crisis state. When the overall score reaches or exceeds this threshold, the system determines that a potential crisis state exists.

[0081] (24) Tiered early warning and proactive intervention mechanism.

[0082] A tiered response strategy is adopted based on the risk score: Level 1 Warning: Reminds users to pay attention to changes in indicators.

[0083] Level 2 warning: Suggests medication or lifestyle interventions.

[0084] Level 3 warning: It is recommended to contact family members or medical personnel.

[0085] The specific evaluation logic for the Level 1, Level 2, and Level 3 early warnings is as follows: Three incrementally increasing risk thresholds are preset, defined sequentially as the first risk threshold. Second risk threshold and the third risk threshold And satisfy < < Based on the calculated crisis risk score The interval in which it is located is classified into levels: (1) Level 1 Warning (Low Risk / Attention Level): When the following conditions are met < < At this point, a Level 1 warning is issued. When physiological indicators show only slight fluctuations, the system sends a notification message via the mobile device, prompting the user to pay attention to changes in these indicators. (2) Level II Early Warning (Medium Risk / Intervention Level): When the following conditions are met < < At this point, it is determined to be a Level II warning. When physiological indicators show a clear trend of abnormal evolution, the system automatically loads the corresponding disease management strategy and pushes specific medication adjustment suggestions or lifestyle intervention guidance to the user. (3) Level 3 Early Warning (High Risk / Crisis Level): When the conditions are met > At this point, the system determines that the user is in an acute risk state and immediately triggers the proactive warning mechanism, automatically sending a request for help and location data to preset emergency contacts, family members, or the medical monitoring backend.

[0086] This tiered strategy effectively reduces the false alarm rate while improving the timeliness of intervention.

[0087] 3. Intelligent judgment of consistency between medication safety management and doctor's orders, such as... Figure 3 As shown, it includes the following steps: (31) Structured representation of medication information.

[0088] Patients can upload medical orders and medication information via photos. After Optical Character Recognition (OCR) and information extraction, each medication is represented as a structured object:

[0089] in, Represents the structured feature object of the j-th drug. Indicates the name of the drug. Indicates a single dose. Indicates the frequency of use. Indicates the duration of administration.

[0090] (32) Determine the total dosage of medication and whether it exceeds the limit.

[0091] Calculate the total actual dosage of medication taken per unit time. :

[0092] Let the drug The safe maximum dose is ,when:

[0093] The system has identified this as a risk of overdose.

[0094] (33) Identify whether drugs are mixed or have conflicting ingredients.

[0095] Introducing a collection of drug components express:

[0096] When two drugs are present ,satisfy:

[0097] This is then identified as a risk of duplicate drug use.

[0098] Meanwhile, based on the pharmacological action rule base, if the following exists:

[0099] This will trigger a drug interaction warning. Among them, , represents a drug interaction discrimination function based on a pharmacological action rule base. This function is used to evaluate drug components. Are there any pharmacological conflicts or contraindications between them?

[0100] (34) Use the medical order consistency verification model to determine whether there is a deviation in medical order execution.

[0101] The system represents the user's actual medication use behavior as a time function:

[0102] This represents the user's actual medication behavior status indicator for drug j at time t; specifically, when the user performs the action of taking medication at time t, The value is 1; if the user has not taken the medication at time t, then... The value is 0.

[0103] The prescribed medication time is as follows ,when:

[0104] in If the compliance threshold is set, the system will determine that there is a deviation in the execution of medical orders.

[0105] (35) Comprehensive risk score of combined drug use Multiple risk categories are combined into a unified score:

[0106] in For the first Risk indicators For the corresponding weights, It represents a comprehensive risk score for combined drug use and sets... This indicates the system's preset threshold for triggering medication risk alerts. For the first Risk indicators For the corresponding weights.

[0107] when At that time, the system will automatically trigger a medication risk alert.

[0108] 4. Automatically classify chronic diseases based on the disease management database.

[0109] The disease management database is constructed as follows: A standardized and scalable disease management library will be constructed to store the core knowledge elements of different diseases in the long-term management process. Each disease category will be represented as a structured object:

[0110] in, Indicates the first in the disease management database i Structured feature objects of chronic diseases Indicates the disease name and code. It represents a set of key physiological indicators (such as blood pressure, blood sugar, etc.). Indicates the risk indicator threshold range, This represents a set of recommended management strategies.

[0111] This disease management database supports dynamic expansion, allowing the addition of new disease types without affecting the overall system structure.

[0112] An automatic chronic disease patient classification model is used to automatically classify users into chronic disease categories: based on user health status vectors. Based on historical diagnostic information, users are automatically categorized for chronic diseases.

[0113] Define the disease matching function:

[0114] in The first in the health state vector One dimension, For disease The corresponding indicator range, For indicator functions, As the indicator weight, A vector representing the user's current health status. With the i-th type of chronic disease in the disease management database The feature matching score between the features is a comprehensive quantification of the degree to which the user's current physiological indicators conform to the pathological characteristics of the specific chronic disease.

[0115] Select the disease category with the highest matching score:

[0116] in, This represents the specific chronic disease category with the highest matching score calculated after comparing all candidate diseases (i.e., the most likely disease subtype initially matched by the system). This indicates the system's preset disease classification trigger score threshold.

[0117] when At that time, users will be categorized as the corresponding chronic disease management subjects.

[0118] After completing the disease classification, the corresponding set of management strategies is extracted from the disease management database. And combine user history and compliance level Generate personalized management suggestions:

[0119] in, This indicates that after calculation by this function, a set of personalized health management strategies is finally generated for the current user. This represents the personalized strategy generation and mapping function (or strategy adaptive update function), which is used to dynamically adjust the basic management logic of the specific disease based on the user's real-time health status. This represents the user's current comprehensive health status vector, which is continuously monitored and constructed by the system.

[0120] 5. Knowledge-constrained reasoning and medication consultation mechanism of AI medical assistants (51) Functional positioning of AI medical assistant The AI ​​medical assistant in this invention is not a general dialogue model, but a task-oriented reasoning system constrained by medical knowledge and rules. It is mainly used to: answer patients' questions about medication, explain the reasons for risk warnings, help patients understand medical orders, and provide non-diagnostic health management advice.

[0121] (52) Knowledge-constrained reasoning model The reasoning process of AI medical assistants uses structured knowledge as its core input, assuming the user's question is represented as a semantic vector. Search for relevant medical knowledge sets :

[0122] in, Represents a set of medical knowledge The first in k A specific knowledge entity (or knowledge item) serves as the prior background for logical reasoning in the medical big data model.

[0123] The inference results are given through the constraint generating function:

[0124] in, A set of medical safety constraint rules, The rules constrain the generation of probabilities based on conditions to ensure that the generated content does not violate existing medical knowledge or medication safety principles. This represents any candidate output result (such as a candidate health management action or candidate response strategy) that the medical big data model traverses during the inference generation phase. Indicates that given the current query input Knowledge Collection and rules and constraints Under the given conditions, the final optimal reasoning result (i.e., the medical decision or personalized strategy ultimately executed by the system) is selected by maximizing probability calculation.

[0125] The aforementioned set of medical safety constraint rules With a collection of personalized health management strategies It is not a static database fixed in existing technology, but rather obtained through structured parsing and dynamic mapping of authoritative medical literature using artificial intelligence technology. The specific acquisition method is as follows: The medical safety constraint rule set R: Instead of simply calling existing external pharmacopoeia libraries, the system utilizes the visual and textual deep understanding capabilities of the Medical LLM (Medical Large Language Model) to directly perform multimodal reading and feature extraction on massive amounts of authoritative clinical pharmacology documents, original drug instructions, and clinical treatment guidelines. The large model automatically extracts entity relationships such as drug incompatibilities, daily maximum dosage limits, and medication red lines under specific physiological indicators, and transforms them into a structured safety constraint rule graph that the system can directly compute.

[0126] Acquisition of Management Strategy Set: Using a medical big data language model, semantic analysis is performed on the Clinical Practice Guidelines and expert consensus for various chronic diseases to extract standard intervention pathways (including diet, exercise, and monitoring frequency benchmarks) for different diseases, forming a basic strategy. Personalized and Exclusive Health Management Strategy Set Generation: It is not statically existing, but dynamically generated based on the user's real-time health status. The generation method is as follows: using basic management strategies as a baseline, and the user's current multimodal health status vector H as the input variable, it is corrected through a strategy adaptive mapping function f. For example, when the system predicts a future risk of high blood pressure, function f automatically increases the monitoring frequency parameter of the basic management strategy set, and combines this with the user's current medication history and physiological characteristics to generate a personalized set of health management strategies, including specific medication adjustment reminders, targeted exercise contraindications, and follow-up suggestions. This strategy can be dynamically updated, enabling long-term follow-up management.

[0127] (53) Risk mitigation mechanisms in medication counseling The risk mitigation mechanism will be automatically activated when a user's inquiry involves one of the following situations: prescription drug replacement, dosage adjustment, or discontinuation of medication.

[0128] The system only provides risk warnings and consultation suggestions, without directly offering treatment decisions, to ensure medical compliance.

[0129] (54) AI Assistant and System Module Collaboration Mechanism AI medical assistants can access real-time information within the system, such as the current medication list, recent risk warning records, and current disease management status, thereby enabling context-aware intelligent interaction.

[0130] The intelligent health medical system (intelligent therapy system) of the present invention will be described below with reference to specific embodiments. These embodiments are only used to explain the technical solutions of the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0131] (a) Implementation Environment A patient with chronic diseases, who has long suffered from hypertension and is at risk of abnormal blood sugar, wears the following devices daily: a smartwatch (heart rate, blood pressure trend, body movement), a smart ring (blood oxygen, nighttime heart rate variability), and a physiological monitoring necklace (body temperature and posture). Patients install the smart treatment system application on their mobile devices.

[0132] (II) Multimodal physiological data acquisition and modeling Each wearable device continuously collects physiological signals and synchronizes them to the mobile terminal via low-power communication.

[0133] The system constructs a health status vector in the cloud:

[0134] Continuous calculation of the rate of change in health status:

[0135] The system detected a persistently rising trend in nighttime blood pressure, with the rate of change exceeding the individual threshold.

[0136] (III) Crisis Prediction and Early Warning The system invokes a time-series prediction model to predict blood pressure at future time steps:

[0137] Calculate the crisis risk score:

[0138] in The table represents the expected physiological characteristic vector of the user at the next time step t+1, derived by the system through an artificial intelligence prediction model.

[0139] When the risk score exceeds the threshold, the system triggers a level-two warning, prompting the patient to rest and monitor indicators.

[0140] (iv) Medication safety management and prescription verification Patients upload their medical orders and medication information by taking photos. The system automatically performs the following: structured analysis of the drugs, calculation of the total daily dose, detection of drug component conflicts, and generation of risk warnings when the system detects potential synergistic blood pressure increase risks between new and existing drugs. The AI ​​medical assistant then explains the reasons for the risks to the patient.

[0141] (v) Collaboration between disease management and AI assistant The system automatically categorizes patients as "people with long-term hypertension management," loads corresponding disease management strategies, and provides the following support through an AI medical assistant: explaining the causes of blood pressure fluctuations, reminding patients of medication precautions, and suggesting consultation with a doctor to adjust the treatment plan. The entire process does not involve diagnosis or prescription substitution and complies with medical compliance requirements.

[0142] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A health and intelligent medical method based on multimodal wearable sensing and artificial intelligence, characterized in that, Includes the following steps: S1. Collect multimodal physiological data and model to generate health status vectors, and measure the stability of physiological status; S2. Based on the health status vector, identify crisis states and proactively warn of potential emergencies using time-series prediction; S3. By uploading medical orders and medication information, determine whether the medication is consistent with the medical orders; S4. Automatically classify chronic diseases based on the disease management database.

2. The health and intelligent medical method based on multimodal wearable sensing and artificial intelligence according to claim 1, characterized in that, The modeling in step 1 generates a vector reflecting the user's overall health status at a certain point in time through linear mapping and nonlinear activation functions.

3. The health and intelligent medical method based on multimodal wearable sensing and artificial intelligence according to claim 1, characterized in that, In step S1, the stability of physiological state is measured by the health status change rate index. When the health status change rate index exceeds the individualized threshold, it is determined that there is abnormal fluctuation in health status.

4. The health and intelligent medical method based on multimodal wearable sensing and artificial intelligence according to claim 1, characterized in that, In step S2, identifying a crisis state includes the following steps: First, a long short-term memory network is used to model the time series of physiological indicators in order to capture the characteristics of long-term dependence and short-term fluctuations and output the predicted values ​​of future time steps. Then, a crisis risk score is calculated: the predicted value is compared with the individual safety baseline to obtain a risk score; when the risk score is greater than or equal to the threshold, a potential crisis state is determined.

5. The health and intelligent medical method based on multimodal wearable sensing and artificial intelligence according to claim 1, characterized in that, Step S3 includes the following steps: (31) After OCR and information extraction, each drug is represented as a structured object: in, Represents the structured feature object of the j-th drug. Indicates the name of the drug. Indicates a single dose. Indicates the frequency of use. Indicates the duration of administration; (32) Determine the total dosage of medication and whether it is an overdose: Calculate the total actual dosage of medication taken per unit time. : Let the drug The safe maximum dose is ,when: The risk of overdose was identified. (33) Identify whether drugs are mixed or have conflicting ingredients: Introducing a collection of drug components express: When two drugs are present ,satisfy: This is identified as a risk of duplicate drug use; Meanwhile, based on the pharmacological action rule base, if the following exists: This triggers a drug interaction warning, in which, , represents a drug interaction discriminant function based on a pharmacological action rule base, used to evaluate drug components. Are there any pharmacological conflicts or contraindications between them? (34) Use the medical order consistency verification model to determine whether there is a deviation in medical order execution: Representing users' actual medication use behavior as a time function: In the formula Indicates that the user is Always be aware of the drug Actual medication use behavior status indicator value; when the user is... When performing the medication administration action at all times, The value is 1, if the user is... If medication is not taken for a period of time, then The value is 0; The prescribed medication time is as follows ,when: The system then determines that there is a deviation in the execution of medical orders, among which This is the compliance threshold; (35) Comprehensive risk score for combined drug use: Multiple risk categories are combined into a unified score: in, This indicates the overall risk score for combined drug use. For the first Risk indicators For the corresponding weights; when When a medication risk warning is triggered, This indicates the system's preset threshold for triggering medication risk alerts.

6. The health and intelligent medical method based on multimodal wearable sensing and artificial intelligence according to claim 1, characterized in that, In step S4, based on the user's health status vector and historical diagnostic information, the user is matched for chronic diseases using a disease matching function, and the highest matching score is selected as the disease category.

7. The health and intelligent medical method based on multimodal wearable sensing and artificial intelligence according to claim 1, characterized in that, The disease management database is constructed as follows: Each type of disease is represented as a structured object as follows: in, Indicates the first in the disease management database Structured feature objects of chronic diseases Indicates the disease name and code. This represents a set of key physiological indicators. Indicates the risk indicator threshold range, This represents a set of recommended management strategies.

8. A system for implementing the health and intelligent medical method based on multimodal wearable sensing and artificial intelligence as described in claim 1, characterized in that, include: The data acquisition module is used to collect multimodal physiological signals and generate individualized health status vectors; The crisis prediction module is used to identify crisis states and proactively issue warnings based on individualized health status vectors; The medication detection module is used to intelligently determine whether medication safety is consistent with the doctor's orders; The chronic disease classification module is used to automatically classify users' chronic diseases. The AI ​​medical assistant module is used to answer patients' medication-related questions, explain the reasons for risk warnings, help patients understand medical advice, and provide non-diagnostic health management advice.

9. A computer device comprising a memory and a processor, the memory being electrically connected to the processor, the memory storing a computer program, characterized in that: When the computer program is executed by the processor, it causes the processor to implement the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the processor implements the method as described in any one of claims 1 to 8.

Images