An agent-based brain state perception-regulation integrated method

Abstract

The application provides an integrated brain state perception-regulation method based on an agent, comprising: constructing a closed-loop graph state machine; outputting brain state representation and trusted state by a brain state perception node; performing trusted evaluation by the agent and executing transfer gating from the perception node to a subsequent node or a self-loop; outputting a regulation window, a regulation strategy and candidate parameters by a regulation decision-making node; executing individualized path planning and safety constraint checking by the agent; executing regulation by a brain state regulation node and supervising the execution process by the agent; collecting subjective and objective feedback by a brain state feedback node; and determining the next transfer direction according to the feedback summary by the agent. The application realizes the upgrade from a fixed process closed loop to an adaptive intelligent closed loop, and improves the flexibility, safety and efficiency of brain state closed loop regulation.

Description

Technical Field

[0001] This invention belongs to the field of brain-computer interface and artificial intelligence interdisciplinary technology, specifically involving an integrated method for brain state perception and regulation based on intelligent agents. Background Technology

[0002] Brain-computer interface (BCI) technology acquires and analyzes EEG, near-infrared spectroscopy, magnetoencephalography (MEG), or other brain signals to identify, interpret, and regulate brain states. It has significant application value in neurorehabilitation, sleep management, emotion regulation, attention training, fatigue intervention, and cognitive enhancement. With the development of brain signal decoding models and neuromodulation techniques, closed-loop systems that execute regulation based on real-time brain state estimation results are gradually becoming an important direction in the field of BCI.

[0003] Traditional closed-loop brain state regulation methods, such as those in [Neurostimulation CL. Creating the Feedback Loop[J]. Neurosurg Clin N Am, 2014, 25: 187-204] and [Parastarfeizabadi M, Kouzani A Z. Advances in closed-loop deep brainstimulation devices[J]. Journal of neuroengineering and rehabilitation, 2017,14(1): 79], typically follow a fixed path: first, brain state perception is performed; then, regulatory decisions are made; subsequently, brain state regulation is executed; finally, brain state feedback is collected; and after the feedback, brain state perception is re-entered. This process can form a basic closed loop, but the order of its nodes is relatively fixed, mainly relying on preset rules or manually set thresholds for judgment. It lacks unified intelligent decision-making regarding the credibility of brain state representation, individual differences, regulatory safety boundaries, and feedback trends.

[0004] In real-world applications, brain signals are susceptible to noise, artifacts, individual differences, environmental changes, and task state fluctuations, potentially leading to low reliability, conflict, or incomplete brain state representations. If the system mechanically proceeds to subsequent regulatory decision-making and execution stages, it may result in inaccurate selection of the regulatory window, unreasonable parameter settings, or unstable regulatory effects. Furthermore, when feedback indicates the regulatory direction is correct but the intensity is insufficient, the feedback result is uncertain, or the subject experiences discomfort, the fixed closed-loop process struggles to promptly select differentiated paths such as reverting to the original perception, re-decision, continuing the same strategy, or terminating the closed loop.

[0005] Therefore, existing closed-loop brain state regulation methods suffer from problems such as rigid processes, insufficient intelligent decision-making capabilities, inflexible handling of low-reliability perception results and abnormal feedback, and inadequate security constraints and execution supervision. There is an urgent need for a method that can uniformly control the dynamic transfer logic between the four stages—brain state perception, regulation decision-making, brain state regulation, and brain state feedback—without altering the internal processes, in order to improve the flexibility, security, and efficiency of the closed-loop regulation system. Summary of the Invention

[0006] In view of the above, the present invention provides an integrated brain state perception-regulation method based on intelligent agents, which can upgrade from a fixed process closed loop to an adaptive intelligent closed loop, and improve the flexibility, security and efficiency of brain state closed loop regulation.

[0007] A brain state perception-regulation integrated method based on intelligent agents includes the following steps: (1) Construct a closed-loop graph state machine including brain state perception nodes, regulation decision-making nodes, brain state regulation nodes and brain state feedback nodes, and configure an intelligent agent flow controller located on the outer layer of the closed-loop graph state machine. (2) First, the closed-loop graph state machine receives brain signal input at the brain state perception node, performs brain state feature extraction and brain signal decoding, and the intelligent agent flow controller performs a reliable evaluation of the brain state representation output by the brain state perception node. (3) When the closed-loop graph state machine is transferred to the regulation decision-making node, the node determines the regulation window, regulation strategy and candidate regulation parameter range based on the current brain state representation and the target brain state; the agent flow controller performs regulation path planning and safety constraint checks on this information; (4) The brain state regulation node performs brain state regulation according to the regulation window, regulation strategy and candidate regulation parameter range that have passed the safety constraint check. During the regulation execution process, the intelligent agent flow controller continuously monitors the execution status, parameter boundaries and risk markings. (5) When the regulation process is completed according to the constraints or reaches the preset stage observation point, the intelligent agent flow controller will transfer the closed-loop diagram state machine to the brain state feedback node; the brain state feedback node collects the objective brain signal changes, behavioral performance, physiological indicators and subjective and objective feedback after regulation, and forms a feedback summary. (6) The intelligent agent flow controller performs multi-branch transfer based on the feedback summary. The transfer is jointly determined by the effectiveness of feedback, the degree of goal achievement, the consistency of the control direction, the subjective discomfort mark and the objective brain state change trend.

[0008] Furthermore, in step (2), the brain state perception node is used to complete brain signal acquisition, signal preprocessing, brain state feature extraction, brain signal decoding, and brain state representation output; the intelligent agent flow controller does not change the internal process of the brain state perception node, but only performs a reliability assessment based on its output brain state representation. When the reliability score and signal quality meet the preset conditions, the closed-loop graph state machine is transferred to the control decision-making node; when the reliability score or signal quality does not meet the preset conditions, the closed-loop graph state machine is kept at the brain state perception node to trigger retesting, re-acquisition, or re-sensing. The credibility assessment includes a comprehensive evaluation of the model confidence, signal quality, key feature integrity, decoding result consistency, and historical state continuity of the brain state representation, and generates a credibility state for closed-loop transfer gating.

[0009] Further, in step (3), the intelligent agent flow controller performs control path planning and safety constraint checks on the control window, control strategy, and candidate control parameter range output by the control decision-making node based on the individualized control target, historical flow trajectory, and safety boundary. It then decides whether to enter the brain state control node, return to the brain state perception node, or remain at the control decision-making node for strategy switching through closed-loop transfer gating. Control path planning involves determining the candidate transfer path from the current node to the next node based on the individual's baseline brain state, target brain state, historical control response, current risk state, and feedback trend, and setting priorities for candidate control strategies. Safety constraint checks include boundary checks on control intensity, frequency, duration, stimulation location, stimulation sequence, cumulative dose, and subject risk markers. When parameters exceed the limits, the closed-loop graph state machine is prohibited from entering the brain state control node.

[0010] Furthermore, the specific execution criteria for the closed-loop transfer gating are as follows: When the confidence score reaches the threshold, the signal quality meets the requirements, and the brain state category or continuous representation is sufficient to support the regulatory decision, the brain state perception node is transferred to the regulatory decision-making node; otherwise, the brain state perception node is triggered to self-loop in order to perform retesting, re-sampling, or re-sensing. Once the regulation window, regulation strategy, and candidate regulation parameter range have been determined and passed the safety constraint check, the regulation decision-making node is transferred to the brain state regulation node. When decision-making information is insufficient, individualized goals are unclear, or strategy selection relies on additional sensory results, return to the brain state perception node; when candidate strategies do not meet safety, adaptability, or efficiency requirements, reselect the control window or switch control strategies within the control decision-making node.

[0011] Furthermore, the closed-loop graph state machine is used to maintain the global closed-loop state, which includes brain state representation, credibility score, current regulation target, candidate regulation strategy, parameter safety boundary, feedback summary, historical flow trajectory and abnormal event record; the historical flow trajectory is used to record the path, triggering conditions, strategy parameters and feedback results of the closed-loop graph state machine from one node to another, thereby providing a basis for subsequent individualized path planning.

[0012] Furthermore, in step (4), when parameters are detected to be out of bounds, there is a safety risk, strategy mismatch, or abnormal trend of control effect before or during the control process, the intelligent agent flow controller triggers a pause, downgrade, or returns the closed-loop diagram state machine to the control decision-making node for replanning.

[0013] Furthermore, the specific execution criteria for the multi-branch transfer in step (6) are as follows: when the feedback indicates that the brain state has changed or the result is uncertain, the agent flow controller returns the closed-loop graph state machine to the brain state perception node; when the feedback shows that the control direction is correct but the effect is insufficient, the agent flow controller returns the closed-loop graph state machine to the control decision-making node for parameter adjustment; when the feedback meets the conditions for continuing the same strategy and the control parameters are still within the safe boundary, the agent flow controller transfers the closed-loop graph state machine to the brain state control node; when the brain state reaches the target range and the feedback is stable, the agent flow controller terminates the current closed loop or enters the low-frequency monitoring and maintenance state.

[0014] A computer device includes a memory and a processor, wherein the memory stores a computer program and the processor executes the computer program to implement the above-described agent-based brain state perception-regulation integrated method.

[0015] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described agent-based brain state perception-regulation integrated method.

[0016] The method of this invention abstracts the four existing links of brain state perception, regulation decision making, brain state regulation, and brain state feedback into four types of nodes in a closed-loop graph state machine, and sets an intelligent agent flow controller on the outer layer of the four types of nodes. This controller does not replace the internal algorithm flow of each node, but dynamically determines the next transition direction of the closed-loop graph state machine based on the state summary, credibility score, safety mark and feedback index output by each node.

[0017] The core of this invention lies in upgrading the traditional fixed closed-loop control path to an adaptive intelligent closed loop by using an intelligent agent to achieve reliable assessment of brain representations, individualized control path planning, closed-loop transfer gating, safety constraints on control parameters, and supervision of the execution process. The traditional path is a fixed loop of brain state perception, control decision-making, brain state control, brain state feedback, and then returning to brain state perception. This invention, however, allows for retesting or supplementary sampling when reliability is insufficient, regression to perception when decision information is insufficient, re-decision when the strategy is incompatible, pausing and adjusting the strategy when execution risks occur, and choosing to continue control, re-perceive, re-decision, terminate, or enter a low-frequency monitoring and maintenance state based on different feedback results.

[0018] Compared with traditional closed-loop intervention techniques, the present invention has at least the following beneficial technical effects: 1. This invention extends the fixed linear closed loop into a dynamically jumpable closed-loop diagram state machine, improving process flexibility.

[0019] 2. This invention introduces a reliable assessment of brain representations before entering the regulatory decision-making process, thereby reducing the risk of erroneous regulation caused by low-quality perceptual results.

[0020] 3. This invention plans and controls the path based on the individual's historical state and feedback trends, thereby improving the individualized adaptation capability.

[0021] 4. This invention introduces parameter safety constraints and execution supervision before and during regulation, thereby improving safety.

[0022] 5. After feedback, the present invention can choose to continue regulation, re-sensing, re-decision, or terminate the closed loop based on the degree of goal achievement and the certainty of the result, thereby improving the efficiency of the closed loop. Attached Figure Description

[0023] Figure 1 This is a system block diagram of the brain state perception-regulation integrated method based on intelligent agents according to the present invention.

[0024] Figure 2 This is a schematic diagram of the dynamic transition logic of the closed-loop diagram state machine of the present invention. Detailed Implementation

[0025] To describe the present invention in more detail, the technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] like Figure 1As shown, this example provides a specific implementation system for an agent-based integrated brain state perception and regulation method. The system includes a brain state perception node, a regulation decision-making node, a brain state regulation node, a brain state feedback node, and an agent flow controller located outside the four nodes. The four nodes correspond to the four functional stages in the closed-loop regulation, and the agent flow controller is responsible for reading the node outputs and controlling the transfer between nodes.

[0027] In this embodiment, the brain state perception node is used to complete brain signal acquisition, preprocessing, brain state feature extraction, brain signal decoding, and brain state representation output. This node can output discrete brain state categories, continuous brain state vectors, brain state change trends, credibility scores, signal quality labels, key feature integrity labels, and anomaly labels. The agent flow controller does not intervene in the feature extraction or decoding model within this node; it only uses the state summary output by this node as the basis for transition judgment.

[0028] The regulation decision-making node is used to determine the regulation window, regulation strategy, and candidate regulation parameter range based on the target brain state and the current brain state representation. The regulation window can include a time window, frequency band window, brain region window, or task window; the regulation strategy can include neural feedback strategies, stimulation strategies, training strategies, or other brain state regulation strategies; candidate regulation parameters can include intensity, frequency, duration, stimulation location, stimulation sequence, and cumulative dose, etc. The agent flow controller does not change the policy generation logic within this node, but instead performs path planning and safety checks on its output.

[0029] The brain state regulation node is used to execute brain state regulation according to the regulation window, regulation strategy, and regulation parameters output by the regulation decision-making node and which have passed the safety constraint check. The brain state feedback node is used to collect objective brain signal changes, behavioral performance, physiological indicators, and subjective and objective feedback after regulation, and to form a feedback summary. Subjective feedback may include comfort level, fatigue level, emotional feelings, or other self-rated indicators; objective feedback may include changes in brain state indicators, changes in task performance, changes in physiological indicators, or adverse risk markers.

[0030] A closed-loop graph state machine is used to maintain the global closed-loop state, which includes brain state representation, credibility score, current regulatory target, candidate regulatory strategies, parameter safety boundaries, feedback summary, historical flow trajectory, and abnormal event records. The historical flow trajectory records the path, triggering conditions, policy parameters, and feedback results of the closed-loop graph state machine transitioning from one node to another, thus providing a basis for subsequent individualized path planning.

[0031] The agent flow controller first performs a brain representation credibility assessment. Specifically, the agent generates a credible state based on the model confidence, signal quality, key feature integrity, decoding result consistency, and historical state continuity output by the brain state perception node. When the credible state meets preset conditions, the closed-loop graph state machine transitions from the brain state perception node to the control decision-making node; when the credible state does not meet preset conditions, the closed-loop graph state machine remains at the brain state perception node, triggering retesting, re-sampling, re-preprocessing, or re-decoding.

[0032] The agent flow controller then executes individualized regulation path planning. Specifically, the agent determines whether to enter regulation execution, supplement sensory information, reselect strategy, or terminate the closed loop based on the individual's baseline brain state, target brain state, historical regulation response, current risk state, feedback trend, and clinical or application scenario constraints. This path planning does not require the four nodes to cycle in a fixed order, but rather treats the four nodes as a state diagram that can be dynamically jumped to according to conditions.

[0033] like Figure 2 As shown, the closed-loop graph state machine includes at least the following transition logic: when the confidence level of the brain state perception result reaches the threshold, the signal quality meets the requirements, and the brain state representation is sufficient to support the control decision, the transition from the brain state perception node to the control decision making node is executed; when the brain signal quality is insufficient, the confidence level of the brain representation is low, the decoding result is conflicting, or the key features are missing, the self-loop of the brain state perception node is executed to trigger retesting, re-sampling, or re-sensing.

[0034] When the control window, control strategy, and parameter range are all determined and pass the safety constraint check, the transfer from the control decision-making node to the brain state control node is executed; when the brain state information required for decision-making is insufficient, the individualized goal is unclear, or the strategy selection depends on additional sensory results, the transfer from the control decision-making node back to the brain state sensing node is executed; when the candidate strategy does not meet the safety, adaptability, or efficiency requirements, the closed-loop graph state machine remains at the control decision-making node to reselect the control window or switch the control strategy.

[0035] When the regulation process is completed according to constraints or reaches the preset stage observation point, the transfer from the brain state regulation node to the brain state feedback node is executed; when parameters are out of bounds, adverse risks, strategy mismatch, or abnormal trend of regulation effect is found before or during execution, the agent suspends or downgrades execution and executes the transfer from the brain state regulation node back to the regulation decision making node.

[0036] After the brain state feedback node, the agent performs multi-branch transitions based on the feedback summary: when the feedback indicates that the brain state has changed, the result is uncertain, the current brain state needs to be re-estimated, or the next complete closed loop needs to be entered, the closed-loop graph state machine returns to the brain state perception node; when the feedback shows that the regulation direction is correct but the effect is insufficient, or only the policy parameters such as window, intensity, frequency, and duration need to be adjusted, the agent returns to the regulation decision-making node; when the feedback meets the conditions for continuing the same policy and the parameters are still within the safety boundary, the agent enters the brain state regulation node; when the brain state reaches the target range, the feedback is stable, and there is no safety risk, the current closed loop is terminated or the agent enters the low-frequency monitoring and maintenance state.

[0037] In this embodiment, the agent flow controller also performs safety constraints on the modulation parameters. The safety constraint check may include boundary checks on modulation intensity, frequency, duration, stimulus location, stimulus sequence, cumulative dose, and subject risk markers. When the parameters exceed the safety boundaries, the agent prohibits the closed-loop graph state machine from entering the brain state modulation node; when a risk marker or trend abnormality is detected during the modulation execution, the agent triggers a pause, downgrade, or returns to the modulation decision-making node for replanning.

[0038] This implementation can also set different verification scenarios. For example, under high-confidence perception results, the closed-loop graph state machine runs normally according to the path of brain state perception, regulation decision making, brain state regulation, and brain state feedback; under low-confidence perception results, the brain state perception node self-loops; under insufficient decision information, the regulation decision making node reverts to the brain state perception node; under parameters with security risks, the regulation decision making node or the brain state regulation node returns to the regulation decision making node for replanning; under insufficient feedback effect but correct direction, the brain state feedback node returns to the regulation decision making node for parameter adjustment; under feedback indicating that the state has changed or the result is uncertain, the brain state feedback node returns to the brain state perception node; under the condition that the target brain state is reached and the feedback is stable, the system terminates the current closed loop or enters a low-frequency monitoring and maintenance state.

[0039] This invention is not limited to any particular brain signal modality and can be applied to EEG, functional near-infrared spectroscopy, magnetoencephalography (MEG), or multimodal brain signals; it is not limited to any specific brain state modulation method and can be applied to neurofeedback, acoustic-optical-electromagnetic stimulation, or other non-invasive or invasive modulation methods; it is not limited to any specific algorithm model used within the four nodes, and the internal algorithm can be replaced according to the application scenario. The key to this invention lies in utilizing an intelligent agent to control the dynamic transfer logic between the four stages of brain state perception, modulation decision-making, brain state modulation, and brain state feedback.

[0040] The above description of the embodiments is provided to enable those skilled in the art to understand and apply the present invention. Those skilled in the art can readily make various modifications to the above embodiments and apply the general principles described herein to other embodiments without creative effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made to the present invention by those skilled in the art based on the disclosure thereof should be within the scope of protection of the present invention.

Claims

1. A brain state perception-regulation integrated method based on intelligent agents, characterized in that, Includes the following steps: (1) Construct a closed-loop graph state machine including brain state perception nodes, regulation decision-making nodes, brain state regulation nodes and brain state feedback nodes, and configure an intelligent agent flow controller located on the outer layer of the closed-loop graph state machine. (2) First, the closed-loop graph state machine receives brain signal input at the brain state perception node, performs brain state feature extraction and brain signal decoding, and the intelligent agent flow controller performs a reliable evaluation of the brain state representation output by the brain state perception node. (3) When the closed-loop state machine is transferred to the regulation decision-making node, the node determines the regulation window, regulation strategy and candidate regulation parameter range based on the current brain state representation and the target brain state; The intelligent agent flow controller uses this information to regulate path planning and check safety constraints. (4) The brain state regulation node performs brain state regulation according to the regulation window, regulation strategy and candidate regulation parameter range that have passed the safety constraint check. During the regulation execution process, the intelligent agent flow controller continuously monitors the execution status, parameter boundaries and risk markings. (5) When the regulation process is completed according to the constraints or reaches the preset stage observation point, the intelligent agent flow controller will transfer the closed-loop diagram state machine to the brain state feedback node; the brain state feedback node collects the objective brain signal changes, behavioral performance, physiological indicators and subjective and objective feedback after regulation, and forms a feedback summary. (6) The intelligent agent flow controller performs multi-branch transfer based on the feedback summary. The transfer is jointly determined by the effectiveness of feedback, the degree of goal achievement, the consistency of the control direction, the subjective discomfort mark and the objective brain state change trend.

2. The agent-based brain state perception-regulation integrated method according to claim 1, characterized in that: In step (2), the brain state perception node is used to complete brain signal acquisition, signal preprocessing, brain state feature extraction, brain signal decoding, and brain state representation output. The intelligent agent flow controller does not change the internal process of the brain state perception node, but only performs a reliability assessment based on its output brain state representation. When the reliability score and signal quality meet the preset conditions, the closed-loop graph state machine is transferred to the control decision-making node. When the reliability score or signal quality does not meet the preset conditions, the closed-loop graph state machine is kept at the brain state perception node to trigger retesting, re-acquisition, or re-sensing. The credibility assessment includes a comprehensive evaluation of the model confidence, signal quality, key feature integrity, decoding result consistency, and historical state continuity of the brain state representation, and generates a credibility state for closed-loop transfer gating.

3. The agent-based brain state perception-regulation integrated method according to claim 1, characterized in that: In step (3), the intelligent agent flow controller performs control path planning and safety constraint checks on the control window, control strategy and candidate control parameter range output by the control decision-making node according to the individualized control target, historical flow trajectory and safety boundary, and decides to enter the brain state control node, return to the brain state perception node or stay at the control decision-making node to switch strategies through closed-loop transfer gating. Among them, the regulation path planning is to determine the candidate transfer path from the current node to the next node based on the individual's baseline brain state, target brain state, historical regulation response, current risk state and feedback trend, and set the priority of the candidate regulation strategies. Safety constraint checks include boundary checks on modulation intensity, frequency, duration, stimulus location, stimulus sequence, cumulative dose, and subject risk markers. When parameters exceed the limits, the closed-loop graph state machine is prohibited from entering the brain state modulation node.

4. The agent-based brain state perception-regulation integrated method according to claim 3, characterized in that, The specific implementation standards for the closed-loop transfer gating are as follows: When the credibility score reaches the threshold, the signal quality meets the requirements, and the brain state category or continuous representation is sufficient to support the regulation decision, the brain state perception node is transferred to the regulation decision making node. Otherwise, it triggers a self-loop in the brain state perception node to perform retesting, re-sampling, or re-sensing; Once the regulation window, regulation strategy, and candidate regulation parameter range have been determined and passed the safety constraint check, the regulation decision-making node is transferred to the brain state regulation node. When decision-making information is insufficient, individual goals are unclear, or strategy selection relies on additional sensory results, return to the brain state perception node; When a candidate strategy fails to meet the requirements of safety, adaptability, or efficiency, the control window is reselected or the control strategy is switched within the control decision-making node.

5. The agent-based brain state perception-regulation integrated method according to claim 1, characterized in that: The closed-loop graph state machine is used to maintain the global closed-loop state, which includes brain state representation, credibility score, current regulation target, candidate regulation strategy, parameter safety boundary, feedback summary, historical flow trajectory and abnormal event record; the historical flow trajectory is used to record the path, triggering conditions, strategy parameters and feedback results of the closed-loop graph state machine from one node to another, thereby providing a basis for subsequent individualized path planning.

6. The agent-based brain state perception-regulation integrated method according to claim 1, characterized in that: In step (4), when parameters are detected to be out of bounds, there is a safety risk, strategy mismatch, or abnormal trend of control effect before or during the control process, the intelligent agent flow controller triggers a pause, downgrade, or returns the closed-loop diagram state machine to the control decision-making node for replanning.

7. The agent-based brain state perception-regulation integrated method according to claim 1, characterized in that, The specific execution criteria for multi-branch transfer in step (6) are as follows: when the feedback indicates that the brain state has changed or the result is uncertain, the agent flow controller returns the closed-loop graph state machine to the brain state perception node; when the feedback shows that the control direction is correct but the effect is insufficient, the agent flow controller returns the closed-loop graph state machine to the control decision-making node for parameter adjustment; when the feedback meets the conditions for continuing the same strategy and the control parameters are still within the safety boundary, the agent flow controller transfers the closed-loop graph state machine to the brain state control node. When the brain state reaches the target range and the feedback is stable, the intelligent agent flow controller terminates the current closed loop or enters a low-frequency monitoring and maintenance state.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: The processor is used to execute the computer program to implement the agent-based brain state perception-regulation integrated method as described in any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, it implements the agent-based brain state perception-regulation integrated method as described in any one of claims 1 to 7.

Images