Individualized intervention generation system and updating method based on multi-modal health state
By generating a multimodal health status individualized intervention system, the problem of existing systems being unable to dynamically adjust intervention plans has been solved, realizing multimodal-driven individualized intervention and providing intelligent health management solutions for multiple scenarios.
Patent Information
- Application Number
- CN202610127836.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-29
- Publication Date
- 2026-05-15
AI Technical Summary
Existing health monitoring systems lack multimodal driven closed-loop capabilities and cannot dynamically adjust intervention plans based on various physiological signals such as EEG, EDA, HRV, voice, and facial movements, resulting in poor static intervention effects.
Design a personalized intervention generation system based on multimodal health status, including a health status input module, an intervention rule base module, an intervention matrix generation module, an intervention execution module, an intervention feedback module, and a learning module. The system constructs an intervention matrix through multimodal signals, achieves adaptive learning and predicts future trends, and generates personalized intervention plans.
It enables multimodal-driven individualized intervention, which can adjust the intensity, frequency and timing of intervention in real time, and provides intelligent solutions for multiple scenarios such as stress management, sleep regulation, emotional balance and cognitive enhancement, and has adaptive learning capabilities.
Smart Images
Figure CN122050815A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of biomedical engineering technology, and in particular to a personalized intervention generation system and update method based on multimodal health status. Background Technology
[0002] With the development of wearable devices and artificial intelligence, the use of multimodal physiological signals for health monitoring has become a trend. However, most existing systems remain at the "detection and recognition" stage, lacking the closed-loop capability of "recognition → prediction → intervention → self-learning". Especially in scenarios such as stress management, sleep regulation, emotional stabilization, and cognitive enhancement, existing technologies mainly adopt static interventions, such as fixed meditation content and fixed frequency sound stimulation, which cannot dynamically adjust the intervention plan according to the individual's real-time state.
[0003] Existing technologies lack multimodal driven intervention systems: most intervention systems rely on only a single modality and cannot construct accurate interventions based on multimodal information such as EEG, EDA, HRV, voice, and facial movements. Summary of the Invention
[0004] The purpose of this invention is to provide a personalized intervention generation system and update method based on multimodal health status, aiming to construct a health status vector that can drive intervention generation based on multimodal signals.
[0005] To achieve the above objectives, in a first aspect, the present invention provides a personalized intervention generation system based on multimodal health status, including a health status input module, an intervention rule base module, an intervention matrix generation module, an intervention execution module, an intervention feedback module, a learning module, and a digital twin prediction module; The health status input module is used to receive the health status vector generated by the system; The intervention rule base module is used to store the intervention rule base; The intervention matrix generation module generates an intervention matrix based on the state vector and the intervention rule base. The intervention execution module executes corresponding interventions based on the intervention matrix; The intervention feedback module is used to collect changes in state before and after the intervention and generate feedback data; The learning module updates the intervention sensitivity coefficient, intervention rule base, and intervention matrix generation parameters based on the feedback data to achieve individualized intervention updates. The digital twin prediction module predicts future trends and generates preventative interventions based on multimodal time series, individual characteristics, and historical intervention effects.
[0006] The multimodal signals include electroencephalography (EEG), heart rate variability, skin conductance, blood oxygenation, speech features, respiratory rhythm, motor behavior, and facial movement signals.
[0007] The intervention rule base includes a TCM syndrome differentiation rule layer, a neural regulation rule layer, a physiological regulation rule layer, a nutritional intervention rule layer, and a behavioral adjustment rule layer. The TCM syndrome differentiation rule layer includes syndrome differentiation parameters and intervention mappings based on Qi, blood, body fluids, deficiency and excess, cold and heat, and meridian status. The neural regulation rule layer includes transcranial ultrasound frequency, duty cycle, power, and point selection rules.
[0008] The intervention matrix includes intervention type, intensity parameter, frequency parameter, duration, execution sequence, and priority.
[0009] Secondly, the present invention also provides a method for updating individualized interventions based on multimodal health status, applied to the individualized intervention generation system based on multimodal health status as described in the first aspect above, comprising the following steps: Obtain the multimodal health state vector; Construct an intervention candidate set based on the intervention rule base; An intervention matrix is generated based on the multimodal health state vector, and each intervention parameter is determined; Execute the interventions in the intervention matrix, collect the state changes before and after the intervention, and obtain feedback data; The intervention rule base, intervention sensitivity coefficient, and intervention matrix generation parameters are updated based on the feedback data to achieve adaptive intervention learning. Based on multimodal time series, individual characteristics, and historical intervention effects, future trends are predicted, and preventive interventions are generated.
[0010] This invention discloses a personalized intervention generation system based on multimodal health status. The system comprises a health status input module receiving a health status vector generated by the system; an intervention rule base module storing the intervention rule base; an intervention matrix generation module generating an intervention matrix based on the status vector and the intervention rule base; an intervention execution module executing the corresponding intervention based on the intervention matrix; an intervention feedback module collecting status changes before and after the intervention and generating feedback data; and a learning module updating the intervention sensitivity coefficient, intervention rule base, and intervention matrix generation parameters based on the feedback data to achieve personalized intervention updates. The learning module employs reinforcement learning, Bayesian update, Bandit parameter selection, or meta-learning models. The digital twin prediction module predicts future trends and generates preventative interventions based on multimodal time series, individual characteristics, and historical intervention effects. Through multimodal input, intervention matrix generation, and adaptive self-learning modules, this system achieves personalized interventions for multiple scenarios such as stress management, sleep regulation, emotional balance, and cognitive enhancement, realizing a closed-loop process of "status recognition → prediction → dynamic intervention generation → effect feedback → adaptive parameter update," providing a closed-loop intelligent solution for digital health management. Compared with existing technologies, this system has the following significant advantages: I. Achieving true multimodal driven intervention: The intervention is driven in real time by multimodal inputs such as electroencephalography (EEG), heart rate variability (HRV), electrodermatology (EDA), respiration, and speech.
[0011] II. Implement a multi-pathway combined intervention program: covering multiple pathways such as light, sound, smell, respiration, transcranial ultrasound (tFUS), traditional Chinese medicine diagnosis, and nutrition.
[0012] 3. Real-time individualization of intervention plans: automatic adjustment of intensity, frequency, dosage, and timing.
[0013] IV. Achieving a closed loop of intervention effects: feedback is completed through state changes ΔS.
[0014] V. Achieving Adaptive Learning: The more interventions, the more accurate the model.
[0015] VI. It can be extended to various health scenarios: stress management, sleep regulation, cognitive enhancement, chronic disease management, etc. Attached Figure Description
[0016] To make the technical solution and advantages of this application clearer, the embodiments of this application will be described in further detail below.
[0017] The terms "first," "second," "third," and "fourth," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.
[0018] It should be noted that all information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals involved in this application have been authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the strain data, acceleration data, displacement data, pressure data, and video data involved in this application were all obtained with full authorization.
[0019] Figure 1 This is a schematic diagram of the overall closed-loop framework of the system of the present invention.
[0020] Figure 2 Structure diagram of the health status input module and intervention rule base.
[0021] Figure 3 This is a schematic diagram of the intervention matrix generation module IM(t).
[0022] Figure 4 A schematic diagram showing the formula for calculating intervention intensity and its parameter input.
[0023] Figure 5 The flowchart for the policy update of the adaptive update module.
[0024] Figure 6 A flowchart for prediction by the digital twin prediction module.
[0025] Figure 7 A detailed structural diagram of the TCM syndrome differentiation rule layer and the tFUS neural regulation rule layer.
[0026] Figure 8 A timeline flowchart for system intervention execution (status acquisition, intervention, feedback, update).
[0027] Figure 9 This is a schematic diagram of the individualized intervention generation system based on multimodal health status provided by the present invention.
[0028] Figure 10This is a flowchart of the individualized intervention update method based on multimodal health status provided by the present invention.
[0029] In the diagram: 1-Health status input module, 2-Intervention rule base module, 3-Intervention matrix generation module, 4-Intervention execution module, 5-Intervention feedback module, 6-Learning module, 7-Digital twin prediction module. Detailed Implementation
[0030] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, but should not be construed as limiting the present invention.
[0031] Please see Figures 1 to 10 In a first aspect, the present invention provides an individualized intervention generation system based on multimodal health status, including a health status input module 1, an intervention rule base module 2, an intervention matrix generation module 3, an intervention execution module 4, an intervention feedback module 5, a learning module 6, and a digital twin prediction module 7. The health status input module 1 is used to receive the health status vector generated by the system; The intervention rule base module 2 is used to store the intervention rule base; The intervention matrix generation module 3 generates an intervention matrix based on the state vector and the intervention rule base. The intervention execution module 4 executes the corresponding intervention based on the intervention matrix; The intervention feedback module 5 is used to collect changes in state before and after intervention and generate feedback data; The learning module 6 updates the intervention sensitivity coefficient, intervention rule base, and intervention matrix generation parameters based on the feedback data to achieve individualized intervention updates. The digital twin prediction module 7 predicts future trends and generates preventative interventions based on multimodal time series, individual characteristics, and historical intervention effects.
[0032] In this embodiment of the invention, the health status input module 1 receives the health status vector S(t) generated by the system, the intervention rule base module 2 stores the intervention rule base, the intervention matrix generation module 3 generates the intervention matrix IM(t) based on the status vector and the intervention rule base, the intervention execution module 4 executes the corresponding intervention based on the intervention matrix, the intervention feedback module 5 collects the status changes before and after the intervention and generates feedback data, the learning module 6 updates the intervention sensitivity coefficient, the intervention rule base and the intervention matrix generation parameters based on the feedback data to achieve individualized intervention updates, the learning module 6 adopts a reinforcement learning model, a Bayesian update model, a Bandit parameter selection model or a meta-learning model, and the digital twin prediction module 7 predicts future trends and generates preventive interventions based on multimodal time series, individual characteristics and historical intervention effects, with the evaluation time window for intervention effects being 10 seconds to 10 minutes. This system, through modules such as multimodal input, intervention matrix generation, and adaptive self-learning, enables individualized interventions in multiple scenarios, including stress management, sleep regulation, emotional balance, and cognitive enhancement. It achieves a closed-loop process of "state recognition → prediction → dynamic intervention generation → effect feedback → adaptive parameter update," providing a closed-loop intelligent solution for digital health management.
[0033] Secondly, the present invention also provides a method for updating individualized interventions based on multimodal health status, applied to the individualized intervention generation system based on multimodal health status as described in the first aspect above, comprising the following steps: S1 obtains the multimodal health state vector; In this embodiment of the invention, the multimodal signals include electroencephalography (EEG), heart rate variability, skin conductance, blood oxygenation, speech features, respiratory rhythm, motor behavior, and facial movement signals. The health state vector S(t) is derived from the multimodal signal recognition system and includes stress index, sleep stage, and neurological function indicators.
[0034] S2 constructs an intervention candidate set based on the intervention rule base; In this embodiment of the invention, the intervention rule base includes a TCM syndrome differentiation rule layer, a neural regulation rule layer, a physiological regulation rule layer, a nutritional intervention rule layer, and a behavioral adjustment rule layer for classified storage, ensuring the feasibility and diversity of interventions. The TCM syndrome differentiation rule layer includes syndrome differentiation parameters and intervention mappings based on Qi, blood, body fluids, deficiency and excess, cold and heat, and meridian status. The neural regulation rule layer includes transcranial ultrasound frequency, duty cycle, power, and point selection rules.
[0035] S3 generates an intervention matrix based on the multimodal health state vector and determines each intervention parameter; In this embodiment of the invention, a multi-objective decision-making algorithm based on S(t) evaluates the offset of the current state S(t) and then queries the rule base to obtain all valid intervention candidate sets. Subsequently, specific parameters are determined through intervention intensity calculation, and finally a combined intervention matrix IM(t) containing multiple intervention types, their respective parameters, and priorities is generated. The priority of the intervention matrix is calculated based on the state offset ΔS, the user sensitivity coefficient k_user (updated by the adaptive module), and the tolerance threshold T_tolerance.
[0036] The formula for calculating intervention intensity I is as follows: I = f(ΔS, k_user, dS / dt, T_tolerance) Where ΔS represents the deviation between the current health status S(t) and the individual's baseline status; k_user represents the individual's intervention sensitivity coefficient, which reflects the individual's response to different intervention methods; dS / dt represents the real-time trend of health status changes, which is used to judge the urgency of status changes; T_tolerance represents the individual's tolerance threshold to intervention intensity; and I represents the intervention intensity parameter.
[0037] The intervention matrix includes one or more of the following: light stimulation, sound stimulation, odor stimulation, breathing training, tFUS, traditional Chinese medicine syndrome differentiation intervention, nutritional intervention, and behavioral intervention. The light stimulation intervention includes a wavelength range of 450-650nm and a PWM parameter range of 1-60%. The sound stimulation intervention includes a rhythmic frequency of 1-100Hz and a cyclic stimulation sequence of 10-60 seconds. The breathing training intervention includes 4-7-8 rhythms, 4-6 rhythms, or individualized rhythm modeling.
[0038] S4 executes the interventions in the intervention matrix, collects state changes before and after the intervention, and obtains feedback data; In this embodiment of the invention, the state change ΔS is calculated based on the EEG α / θ ratio (dS / dt), HRV index, EDA response frequency, breathing depth, speech stability and / or behavioral fluctuation parameters.
[0039] S5 updates the intervention rule base, intervention sensitivity coefficient, and intervention matrix generation parameters based on the feedback data to achieve adaptive intervention learning.
[0040] In this embodiment of the invention, the intervention matrix IM(t) is updated on a rolling basis with a period of 5-300 seconds, and the intervention intensity is automatically reduced or the intervention is paused when an adverse reaction is detected.
[0041] S6 predicts future trends and generates preventative interventions based on multimodal time series, individual characteristics, and historical intervention effects. In this embodiment of the invention, the digital twin prediction module 7 estimates the future state S(t+Δt). If S(t+Δt) indicates a poor state, the system generates a preventative intervention matrix in advance. For example, if it predicts that sleep quality will decline 45 minutes before bedtime, the system generates an acoustic stimulation transition plan in advance.
[0042] To better understand this technical solution, the following embodiments are provided for further explanation: Example 1: Stress Intervention Generation Based on EEG+HRV EEG data was collected (alpha wave decreased by 35%), HRV (LF / HF = 3.2); the state vector S(t) was identified as "high stress level"; the system retrieved the following from the rule base: Breathing rhythm training (4-6 rhythms) Generating a soothing photostimulation (560nm, low PWM) intervention matrix: IM = {breathing intensity 0.7, frequency 5 breaths / min, duration 6 minutes, ...} Light stimulation (intensity 0.3, wavelength 560nm, duration 200 seconds) After intervention, alpha wave recovery was 18%, and LF / HF decreased to 2.1; the intervention sensitivity coefficient k_user was updated.
[0043] Example 2: Pre-sleep intervention based on EEG delta wave enhancement EEG collected before sleep: delta wave is low (40% below baseline); Digital twin prediction: Sleep quality will decline after 45 minutes; System generated: γ→α acoustic stimulation transition scheme Micro-light decompression lamp (450–550nm) intervention matrix: Sound stimulation: linear decrease from 40Hz to 12Hz (5 minutes) Photostimulation: 500–550 nm decreases over time After the intervention, delta waves increased by 18%, predicting improved sleep quality.
[0044] Example 3: tFUS intervention generation (mild anxiety) Data collection includes EDA (increased SCR frequency) and increased speech fundamental frequency jitter; state recognition: mild anxiety; system calls neural modulation rules: tFUS left prefrontal cortex (F3) 400kHz Air volume 20% After 4 rounds of intervention with 30 seconds per round, EDA decreased by 25% and voice jitter decreased by 18%.
[0045] Example 4: Traditional Chinese Medicine (TCM) syndrome differentiation and intervention to generate (deficiency syndrome) HRV collected: SDNN low; decreased speech energy; low facial vitality; Traditional Chinese Medicine diagnosis module determined: Qi deficiency; System generated: Sound stimulation of the Ren meridian (CV6) Meridian photostimulation (620nm) intervention matrix: CV6 high-frequency 20Hz acoustic stimulation for 200 seconds, 620nm light stimulation for 300 seconds SDNN increased by 12% after intervention.
[0046] Example 5: Emotion Intervention Generation Based on Breathing Rhythm Breathing data collected: shallow breathing, frequency 16 breaths / min; The state is identified as "tense"; Generate breathing training: 4-7-8 rhythm After an intervention of intensity 0.5, breathing depth increased by 30%.
[0047] Example 6: Light stimulation intervention to generate (attention deficit) EEG: Increased θ / β ratio = decreased attention; system selection: blue light stimulation (460nm); short pulse (PWM 30%) + intermittent stimulation; after 3 minutes of intervention, the θ / β ratio decreased by 20%.
[0048] Example 7: Odor Intervention Generation (Fatigue Recovery) Behavioral data collected: decreased gait stability, slowed eye movements; state identified as "mild fatigue"; system-generated odor stimuli: Citrus volatile components Peak stimulation for 10 seconds + platform for 90 seconds resulted in a 15% recovery of behavioral indicators.
[0049] Example 8: Nutritional Intervention Generates (Cognitive Decline Risk) EEGα power decreased by 25%, speech rate decreased; digital twin prediction MCI risk increased; system generation: Omega-3 B vitamins combination The nighttime light suppression strategy reduced the predicted risk by 18%.
[0050] Example 9: Hybrid Intervention Matrix (High Stress + Sleep Instability) Status: HRV anomaly + EEG alpha wave suppression + EDA high system generation combination matrix: Intervention type parameter Breathing training 5-5 rhythm Sound stimulation 10Hz 10 Hertz Light stimulation Warm color 580nm odor Low concentration of lavender tFUS Do not enable Sleep improved for two consecutive nights after the intervention.
[0051] Example 10: Adaptive Learning Process The initial intervention matrix strength was I=0.5; the recorded improvement was ΔS=+12%; the system improved k_user=k_user+0.2; the new intervention strength was increased to I=0.6; after 10 days, the model completed 7 self-learning updates, and the intervention effect steadily improved by 30%.
[0052] The above-disclosed embodiments are merely preferred embodiments of the individualized intervention generation system and update method based on multimodal health status in this application, and should not be construed as limiting the scope of this application. Those skilled in the art can understand that all or part of the processes of the above embodiments can be implemented, and equivalent changes made in accordance with the claims of this application still fall within the scope of this application.
Claims
1. A personalized intervention generation system based on multimodal health status, characterized in that, It includes a health status input module, an intervention rule base module, an intervention matrix generation module, an intervention execution module, an intervention feedback module, a learning module, and a digital twin prediction module; The health status input module is used to receive the health status vector generated by the system; The intervention rule base module is used to store the intervention rule base; The intervention matrix generation module generates an intervention matrix based on the state vector and the intervention rule base. The intervention execution module executes the corresponding intervention based on the intervention matrix; The intervention feedback module is used to collect changes in state before and after the intervention and generate feedback data; The learning module updates the intervention sensitivity coefficient, intervention rule base, and intervention matrix generation parameters based on the feedback data to achieve individualized intervention updates. The digital twin prediction module predicts future trends and generates preventative interventions based on multimodal time series, individual characteristics, and historical intervention effects.
2. The personalized intervention generation system based on multimodal health status as described in claim 1, characterized in that, The multimodal signals include electroencephalography (EEG), heart rate variability, skin conductance, blood oxygenation, speech features, respiratory rhythm, motor behavior, and facial movement signals.
3. The personalized intervention generation system based on multimodal health status as described in claim 1, characterized in that, The intervention rule base includes a TCM syndrome differentiation rule layer, a neural regulation rule layer, a physiological regulation rule layer, a nutritional intervention rule layer, and a behavioral adjustment rule layer. The TCM syndrome differentiation rule layer includes syndrome differentiation parameters and intervention mappings based on Qi, blood, body fluids, deficiency and excess, cold and heat, and meridian status. The neural regulation rule layer includes transcranial ultrasound frequency, duty cycle, power, and point selection rules.
4. The personalized intervention generation system based on multimodal health status as described in claim 1, characterized in that, The intervention matrix includes intervention type, intensity parameter, frequency parameter, duration, execution sequence, and priority.
5. A method for updating individualized interventions based on multimodal health status, applied to the individualized intervention generation system based on multimodal health status as described in any one of claims 1-4, characterized in that, Includes the following steps: Obtain the multimodal health state vector; Construct an intervention candidate set based on the intervention rule base; An intervention matrix is generated based on the multimodal health state vector, and each intervention parameter is determined; Execute the interventions in the intervention matrix, collect the state changes before and after the intervention, and obtain feedback data; The intervention rule base, intervention sensitivity coefficient, and intervention matrix generation parameters are updated based on the feedback data to achieve adaptive intervention learning. Based on multimodal time series, individual characteristics, and historical intervention effects, future trends are predicted, and preventive interventions are generated.