CN-122025028-A - Closed loop self-adaptive psychological intervention method and system based on multi-mode brain-computer fusion

CN122025028ACN 122025028 ACN122025028 ACN 122025028ACN-122025028-A

Abstract

The invention discloses a closed-loop self-adaptive psychological intervention method and a closed-loop self-adaptive psychological intervention system based on multi-modal brain-computer fusion, which are characterized in that real-time multi-modal physiological and behavioral data are analyzed through fusion, a current state vector of a user is generated, parallel strategy simulation is further carried out through a personalized digital twin brain network model based on the vector and an intervention target, a simulation decision result comprising an optimal intervention strategy package and an expected effect curve is generated, multi-modal intervention equipment is dynamically scheduled to execute cooperative intervention according to the simulation decision result, real-time feedback data are collected and compared with the expected effect curve, an intervention effect evaluation result and model correction parameters are generated to update the personalized digital twin brain network model, closed-loop optimization is formed until the intervention effect evaluation result meets a termination condition, and a report is output. The invention realizes prospective simulation and optimization of the intervention strategy through the personalized digital twin brain network model, and remarkably improves the accuracy, the adaptability and the long-term effect of psychological intervention based on a closed-loop mechanism of real-time feedback and model update.

Inventors

Sun jiangnan
LIN ZHANGYA
Zhuang Zisen
SHEN NAN
HUANG JINQIAO
CHEN YING
ZHANG XINRAN

Assignees

福建医科大学附属第一医院

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (10)

1. A closed loop adaptive psychological intervention method based on multi-modal brain-computer fusion, comprising: receiving real-time multi-mode physiological data and behavior data of a user acquired by a wearable device cluster; Performing fusion analysis on the real-time multi-mode physiological data and the behavior data to generate a user current state vector containing a multi-dimensional psychological state index and a signal quality index; Based on the current state vector of the user and a preset intervention target, carrying out parallel strategy simulation through a personalized digital twin brain network model, and generating a simulation decision result comprising at least one optimal intervention strategy packet and an expected effect curve; Dynamically scheduling and executing the multi-mode intervention equipment appointed in the optimal intervention strategy package according to the simulation decision result so as to apply cooperative intervention to a user and collect real-time feedback data in the intervention process; comparing the real-time feedback data with the expected effect curve to generate an intervention effect evaluation result and a model correction parameter; Updating the personalized digital twin brain network model according to the intervention effect evaluation result and the model correction parameters, re-executing steps from generating a user current state vector to generating a simulation decision result based on the updated personalized digital twin brain network model until the intervention effect evaluation result meets a preset closed-loop termination condition, and outputting a final psychological state optimization report.
2. The closed-loop adaptive psychological intervention method based on multi-modal brain-computer fusion according to claim 1, wherein performing fusion analysis on the real-time multi-modal physiological data and behavior data to generate a user current state vector containing a multi-dimensional psychological state index and a signal quality index comprises: Performing adaptive mixed filtering based on independent component analysis and wavelet transformation on the electroencephalogram signals in the real-time multi-mode physiological data to filter motion artifacts and calculate the real-time signal quality index of the electroencephalogram signals; Performing time domain analysis and frequency domain analysis on the heart rate variability signals in the real-time multi-mode physiological data to extract heart rate variability characteristics; extracting features of eye movement signals in the real-time multi-mode physiological data to obtain pupil diameter change features and glance speed features; carrying out emotion voice analysis on voice signals in the behavior data, and extracting fundamental frequency characteristics and speech speed characteristics; based on a preset task type, distributing dynamic fusion weights to the electroencephalogram signals, the heart rate variability signals, the eye movement signals and the voice signals through a dynamic weight self-adaptive algorithm; Adopting a cross-modal feature extraction framework to perform cross-modal feature fusion based on a multi-head attention mechanism on the electroencephalogram signals, the heart rate variability features, the pupil diameter change features and glance speed features, and the fundamental frequency features and speech speed features after motion artifacts are filtered; based on the cross-modal feature fusion result, calculating and generating the multidimensional psychological state index comprising the anxiety index, the depression tendency index, the cognitive load index and the emotion valence index, and forming the real-time signal quality index and the multidimensional psychological state index into the current state vector of the user.
3. The closed-loop adaptive psychological intervention method based on multi-modal brain-computer fusion according to claim 2, wherein assigning dynamic fusion weights to the electroencephalogram signal, the heart rate variability signal, the eye movement signal and the voice signal by a dynamic weight adaptive algorithm based on a preset task type comprises: According to the preset task type, basic weight configuration corresponding to the task type is obtained from a preset weight configuration library, wherein the basic weight configuration is an initial weight value respectively distributed to the electroencephalogram signal, the heart rate variability signal, the eye movement signal and the voice signal; Acquiring a real-time signal quality index of the electroencephalogram signal, a signal-to-noise ratio of the heart rate variability signal, sampling integrity of the eye movement signal and definition of the voice signal, and taking the acquired real-time signal quality index and the acquired signal quality index as real-time quality indexes of all mode signals; Based on the real-time quality index of each modal signal, calculating to obtain respective quality adjustment coefficients for the electroencephalogram signal, the heart rate variability signal, the eye movement signal and the voice signal through a preset quality-weight mapping function; Multiplying the initial weight values corresponding to the electroencephalogram signal, the heart rate variability signal, the eye movement signal and the voice signal by the quality adjustment coefficient to obtain adjusted weights of the electroencephalogram signal, the heart rate variability signal, the eye movement signal and the voice signal; And normalizing the adjusted weights of the electroencephalogram signal, the heart rate variability signal, the eye movement signal and the voice signal to obtain the dynamic fusion weights of the electroencephalogram signal, the heart rate variability signal, the eye movement signal and the voice signal.
4. The closed-loop adaptive psychological intervention method based on multi-modal brain-computer fusion according to claim 1, wherein based on the user current state vector and a preset intervention target, parallel strategy simulation is performed through a personalized digital twin brain network model, and a simulation decision result comprising at least one optimal intervention strategy package and an expected effect curve is generated, comprising: Inputting the current state vector of the user and the preset intervention target into the personalized digital twin brain network model, wherein the personalized digital twin brain network model is a calculation model which is constructed based on user historical multi-mode physiological data and behavior data and simulates the dynamic coupling relation of a user brain key network; Generating a plurality of candidate intervention strategies based on the personalized digital twin brain network model, wherein the candidate intervention strategies comprise a combination of at least one intervention means of transcranial electrical stimulation parameters, nerve feedback game parameters, virtual reality scene parameters and audio guidance parameters; In the personalized digital twin brain network model, carrying out parallel simulation on the candidate intervention strategies, simulating the evolution track of the simulated neural state and psychological state indexes in the personalized digital twin brain network model after each candidate intervention strategy is applied, and generating a plurality of candidate expected effect curves; Evaluating the plurality of candidate expected effect curves based on a multi-objective optimization algorithm, wherein the evaluated optimization objectives comprise an expected improvement amplitude of an intervention effect, an expected neural efficiency of an intervention process and an expected user acceptance of an intervention strategy; And selecting at least one candidate intervention strategy from the plurality of candidate intervention strategies as the optimal intervention strategy package according to the evaluation result of the multi-objective optimization algorithm, and taking the candidate expected effect curve corresponding to the optimal intervention strategy package as the expected effect curve to jointly form the simulation decision result.
5. The closed loop adaptive psychological intervention method based on multi-modal brain-computer fusion according to claim 4, wherein generating a plurality of candidate intervention strategies based on the personalized digital twin brain network model comprises: Determining a search space of an intervention strategy according to a preset intervention target and the current state vector of the user, wherein the search space is defined by a frequency and intensity parameter range of transcranial electric stimulation, a difficulty and feedback threshold parameter range of a nerve feedback game, an emotion valence and complexity parameter range of a virtual reality scene and a type and rhythm parameter range of audio guidance; Adopting a strategy generation network based on reinforcement learning, taking the simulation state of the personalized digital twin brain network model as environment input, taking the preset intervention target as a reward function guide, and exploring in the search space to generate a preliminary intervention strategy sequence; Performing diversity enhancement processing on the preliminary intervention strategy sequence, wherein the diversity enhancement processing comprises random disturbance and combined variation on the transcranial electrical stimulation parameters, the nerve feedback game parameters, the virtual reality scene parameters and the audio guidance parameters; And fusing the preliminary intervention strategy sequence subjected to the diversity enhancement processing with a strategy template retrieved from a history successful intervention case based on case reasoning to generate the candidate intervention strategies.
6. The closed-loop adaptive psychological intervention method based on multi-modal brain-computer fusion according to claim 1, wherein dynamically scheduling and executing the multi-modal intervention device specified in the optimal intervention policy package according to the simulation decision result to apply collaborative intervention to the user and collect real-time feedback data in the intervention process comprises: Analyzing the optimal intervention strategy packet in the simulation decision result, and acquiring an intervention means type, an intervention parameter sequence and an intervention time sequence logic appointed in the optimal intervention strategy packet; Generating cooperative control instructions for at least two devices of transcranial electrical stimulation equipment, nerve feedback presentation and interaction equipment, virtual reality equipment and bone conduction audio equipment according to the intervention means type and the intervention parameter sequence; according to the intervention time sequence logic, issuing the cooperative control instruction to corresponding equipment in the transcranial electric stimulation equipment, the nerve feedback presentation and interaction equipment, the virtual reality equipment and the bone conduction audio equipment so as to start cooperative intervention on a user; In the execution process of the collaborative intervention, synchronously acquiring intervention process physiological data of the user through the wearable equipment cluster, wherein the intervention process physiological data at least comprises an electroencephalogram signal, a heart rate variability signal and an eye movement signal in the intervention process; And the physiological data of the intervention process and the execution state information of the collaborative intervention are used together as the real-time feedback data.
7. The method of closed loop adaptive psychological intervention based on multi-modal brain-computer fusion according to claim 6, wherein issuing the cooperative control instructions to the corresponding devices of the transcranial electrical stimulation device, the neural feedback presentation and interaction device, the virtual reality device, and the bone conduction audio device in accordance with the intervention timing logic to initiate the cooperative intervention to the user comprises: constructing a collaborative intervention execution plan taking an absolute time stamp and a relative event triggering condition as nodes according to the intervention time sequence logic; converting the absolute timestamp node in the collaborative intervention execution plan into an accurate scheduling moment for transmitting the collaborative control instruction to the transcranial electrical stimulation device, the nerve feedback presentation and interaction device, the virtual reality device and the bone conduction audio device; Converting the relative event triggering condition in the collaborative intervention execution plan into a monitoring rule and a judging threshold value of the intervention process physiological data acquired by the wearable equipment cluster in real time; After the collaborative intervention is started, sending the collaborative control instruction to the corresponding equipment at the precise scheduling moment, and simultaneously starting monitoring of physiological data of the intervention process; And when the monitored physiological data of the intervention process meets the judgment threshold value corresponding to the relative event triggering condition, immediately sending the corresponding cooperative control instruction to the next device appointed in the cooperative intervention execution plan so as to realize event-driven device scheduling based on physiological state feedback.
8. The closed-loop adaptive psychological intervention method based on multi-modal brain-computer fusion according to claim 1, wherein comparing the real-time feedback data with the expected effect curve to generate an intervention effect evaluation result and a model correction parameter comprises: Extracting intervention process physiological data from the real-time feedback data, and performing fusion analysis on the intervention process physiological data to generate a multidimensional psychological state index sequence in the intervention process; Comparing the multidimensional psychological state index sequence in the intervention process with the psychological state index prediction sequence corresponding to the expected effect curve point by point, and calculating the deviation measure between the multidimensional psychological state index sequence and the psychological state index prediction sequence in the intervention process; Based on the deviation measure, whether the actual effect of the collaborative intervention reaches a target threshold set by the expected effect curve is evaluated, and the intervention effect evaluation result is generated; And when the intervention effect evaluation result indicates that the actual effect of the collaborative intervention does not reach the target threshold, calculating the difference between the posterior distribution and the prior distribution of the parameters of the personalized digital twin brain network model through a Bayesian updating framework based on a deviation mode between the multidimensional psychological state index sequence and the psychological state index prediction sequence in the intervention process, and quantifying the difference into the model correction parameters.
9. The closed-loop adaptive psychological intervention method based on multi-modal brain-computer fusion according to claim 8, wherein calculating the difference between the posterior distribution and the prior distribution of the parameters of the personalized digital twin brain network model by a bayesian update framework based on the deviation pattern between the multi-dimensional psychological state index sequence and the psychological state index prediction sequence in the intervention process, and quantifying the difference as the model correction parameter, comprises: taking the multidimensional psychological state index sequence in the intervention process as observation data in a Bayesian updating framework; Taking the prior probability distribution of the current parameter set of the personalized digital twin brain network model as the prior distribution in a Bayesian updating framework; based on the structure of the personalized digital twin brain network model, constructing a likelihood function of the observed data under the condition of giving the current parameter set; Calculating the posterior probability distribution of the current parameter set by using the observation data, the prior distribution and the likelihood function according to a Bayesian theorem; Calculating a statistical distance between the posterior probability distribution and the prior distribution, the statistical distance comprising a KL divergence or a gasstat distance; And jointly encoding the statistical distance and key deviation characteristics between the observed data formed by driving the posterior probability distribution and the psychological state index prediction sequence into the model correction parameters.
10. A closed loop adaptive psychological intervention system based on multi-modal brain-computer fusion, adapted for use in the method of any of claims 1 to 9, the system comprising: the data acquisition module is configured to acquire real-time multi-mode physiological data and behavior data of a user through the wearable device cluster; the state evaluation module is configured to perform fusion analysis on the real-time multi-mode physiological data and the behavior data to generate a user current state vector containing a multi-dimensional psychological state index and a signal quality index; The strategy simulation module is configured to perform parallel strategy simulation through a personalized digital twin brain network model based on the current state vector of the user and a preset intervention target, and generate a simulation decision result comprising at least one optimal intervention strategy packet and an expected effect curve; The intervention execution module is configured to dynamically schedule and execute the multi-mode intervention equipment appointed in the optimal intervention strategy package according to the simulation decision result so as to apply cooperative intervention to a user and collect real-time feedback data in the intervention process; the effect evaluation module is configured to compare the real-time feedback data with the expected effect curve and generate an intervention effect evaluation result and a model correction parameter; The model updating and closed-loop control module is configured to update the personalized digital twin brain network model according to the intervention effect evaluation result and the model correction parameters, and to trigger the state evaluation module, the strategy simulation module, the intervention execution module and the effect evaluation module to work cooperatively again based on the updated personalized digital twin brain network model until the intervention effect evaluation result meets the preset closed-loop termination condition, and to output a final psychological state optimization report.

Description

Closed loop self-adaptive psychological intervention method and system based on multi-mode brain-computer fusion Technical Field The invention relates to the technical field of intelligent psychological intervention, in particular to a closed-loop self-adaptive psychological intervention method and system based on multi-mode brain-computer fusion. Background Mental health interventions are evolving gradually from relying on subjective scale to combining objective physiological and behavioral data in an intelligent direction. The prior art can collect multi-mode data of a user by means of wearable equipment, voice, behavior analysis and the like, and generate preliminary intervention suggestions through an algorithm model based on the data, so that personalized pushing and dynamic adjustment of the intervention schemes are realized to a certain extent. However, the decision making process of such intelligent intervention systems remains limited. The core of the method generally relies on mining and learning of a 'behavioral-physiological-psychological' correlation pattern in historical data, and is essentially a 'posterior' optimization based on statistics or experience fitting. This approach makes it difficult to perform high-fidelity prospective simulations and comparisons of how various potential strategies affect the user's internal, dynamically changing physiological and psychological states before specific interventions are implemented. Therefore, the generation and selection of the strategy often lack a 'predictive' decision basis capable of reflecting the real-time state evolution of the individual, so that the optimization of the accuracy, timeliness and long-term effect of the intervention faces a bottleneck, and the self-adaptive closed-loop management which is truly adaptive to the depth of the dynamic change of the individual is difficult to realize. Disclosure of Invention In view of the above problems, the invention provides a closed-loop self-adaptive psychological intervention method and a closed-loop self-adaptive psychological intervention system based on multi-mode brain-computer fusion, which are used for carrying out parallel simulation and comparison on the neuropsychological effects of various strategies before intervention by constructing a personalized prediction model capable of simulating the dynamic response of a brain network, so as to realize accurate generation and dynamic closed-loop execution of an optimal intervention strategy based on prospective prediction. To achieve the above object, in a first aspect, the present application provides a closed-loop adaptive psychological intervention method based on multi-modal brain-computer fusion, including: receiving real-time multi-mode physiological data and behavior data of a user acquired by a wearable device cluster; performing fusion analysis on the real-time multi-mode physiological data and the behavior data to generate a user current state vector containing a multi-dimensional psychological state index and a signal quality index; Based on the current state vector of the user and a preset intervention target, carrying out parallel strategy simulation through a personalized digital twin brain network model, and generating a simulation decision result comprising at least one optimal intervention strategy packet and an expected effect curve; According to the simulation decision result, dynamically scheduling and executing the multi-mode intervention equipment appointed in the optimal intervention strategy package so as to apply cooperative intervention to the user and collect real-time feedback data in the intervention process; Comparing the real-time feedback data with an expected effect curve to generate an intervention effect evaluation result and a model correction parameter; Updating the personalized digital twin brain network model according to the intervention effect evaluation result and the model correction parameters, re-executing the steps from generating the current state vector of the user to generating the simulation decision result based on the updated personalized digital twin brain network model until the intervention effect evaluation result meets the preset closed-loop termination condition, and outputting a final psychological state optimization report. In some embodiments, performing fusion analysis on the real-time multi-modal physiological data and the behavioral data to generate a user current state vector comprising a multi-dimensional mental state index and a signal quality index, comprising: Performing adaptive mixed filtering based on independent component analysis and wavelet transformation on the electroencephalogram signals in the real-time multi-modal physiological data to filter motion artifacts and calculate the real-time signal quality index of the electroencephalogram signals; performing time domain analysis and frequency domain analysis on heart rate variability signals in the real-time mult