Clinical Patient Emotion Recognition Intervention Method and System Based on Multimodal Action Analysis

By constructing a personalized emotion coupling baseline spectrum and a targeted decoupling intervention strategy, the problem of the inherent physiological-behavioral synergistic disorder of emotional abnormalities in patients with aphasia, autism, and major depression that cannot be identified and addressed in existing technologies has been solved, achieving precise personalized emotion intervention and closed-loop regulation.

CN123058729APending Publication Date: 2026-11-10HUNAN UNIV OF SCI & ENG +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611128555.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-28
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing technologies are unable to effectively identify and address the underlying physiological-behavioral synergistic disorders of emotional abnormalities in patients with language expression disorders such as aphasia, autism, and major depressive disorder. This results in intervention methods lacking mechanistic targeting, unstable effects, and low personalization.

Method used

By constructing a personalized emotional coupling baseline spectrum, abnormal feature coupling pairs are diagnosed in real time, and targeted decoupling intervention strategies are matched to form a mechanism-driven closed-loop regulation, including collecting action and physiological data, calculating coupling strength, predicting evolutionary trends, identifying abnormal feature coupling pairs, executing targeted decoupling intervention strategies, and evaluating decoupling effectiveness.

Benefits of technology

It enables accurate diagnosis and personalized intervention for emotional abnormalities, forming a quantifiable, interpretable, and predictable intelligent closed loop, thereby improving the effectiveness of clinical emotional intervention.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123058729A_ABST
    Figure CN123058729A_ABST
Patent Text Reader

Abstract

This invention relates to the fields of artificial intelligence and clinical medical technology, and specifically discloses a clinical patient emotion recognition intervention method and system based on multimodal action analysis. The method establishes a personalized emotion coupling baseline spectrum for patients, including a baseline coupling strength and an emotion dynamics model; monitors and compares the baseline in real time, and identifies anomalous coupling pairs based on model prediction; matches and executes targeted intervention strategies aimed at reducing the strength of these anomalous couplings from a pre-constructed decoupling strategy mapping table; and finally evaluates the decoupling effectiveness and confirms the completion of the intervention based on the degree of decrease in coupling strength after intervention. This invention addresses the fundamental shortcomings of existing technologies, such as insufficient quantification of the inherent synergistic disorder patterns of nonverbal emotions and a lack of mechanistic targeting in interventions. It achieves a leap from emotion label management to precise regulation of emotional synergistic relationships, providing clinical practice with an objective, personalized, and interpretable emotion intervention tool.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and clinical medical technology, and in particular to a clinical patient emotion recognition and intervention method and system based on multimodal action analysis. Background Technology

[0002] In the field of clinical psychology and mental health care, especially for patients with language expression disorders such as aphasia, autism, and major depressive disorder, objective, non-invasive identification and timely, effective intervention of their emotional states are core needs for improving the quality of diagnosis and treatment and patient well-being. Currently, mainstream technologies in this field rely on two main approaches: one is motion pattern analysis based on vision or wearable sensors, attempting to statically correlate specific postures and gestures with emotional labels (such as anxiety and calmness); the other is multimodal emotion recognition combined with physiological signals (such as heart rate and skin conductance). While these methods have made some progress, they are essentially still classifications or ratings of patients' external behavioral performance.

[0003] The fundamental flaw in existing technologies lies in their treatment of emotion as a state definable by discrete features, failing to address its dynamic nature as a complex psychophysiological process. This leads to two major clinical challenges: First, it cannot reveal the underlying mechanisms of patients' emotional abnormalities, i.e., it cannot answer the fundamental question of which physiological and behavioral systems have experienced individualized disruptions in their synergistic relationships. Second, the resulting interventions lack mechanistic specificity, often applying generic solutions based on emotion labels rather than precisely regulating the disrupted synergistic relationships. This results in unstable intervention effects, low personalization, and difficulty in forming a closed-loop clinical value from identification to intervention.

[0004] Therefore, this invention proposes a clinical patient emotion recognition intervention method and system based on multimodal action analysis. Summary of the Invention

[0005] This invention provides a clinical patient emotion recognition intervention method and system based on multimodal action analysis. By constructing an individual coupling baseline spectrum, diagnosing abnormal feature coupling pairs in real time, and matching and executing targeted intervention strategies aimed at decoupling this specific disorder, a mechanism-driven closed-loop regulation is formed, which effectively solves the technical defects of existing methods that cannot quantify and address individualized multimodal collaborative disorder patterns in clinical patient emotion expression.

[0006] This invention provides a clinical patient emotion recognition intervention method based on multimodal action analysis, comprising: Collect action and physiological data of the target patient in a baseline emotional state, calculate the baseline coupling strength of at least one set of preset emotional feature pairs, and combine them with emotional state spatial mapping data to form a personalized emotional coupling baseline spectrum; wherein the emotional state spatial mapping data includes baseline state points and fitted personalized emotional dynamics models. Real-time acquisition of action and physiological data of target patients, extraction of identical preset emotional feature pairs and calculation of real-time coupling strength, comparison of real-time coupling strength with baseline coupling strength, prediction of evolutionary trends by combining personalized emotion dynamics model, and identification of abnormal feature coupling pairs; Based on the type of the identified anomalous feature coupling pairs, a targeted decoupling intervention strategy is matched from a pre-built decoupling strategy mapping table and executed; wherein the targeted decoupling intervention strategy is used to reduce the coupling strength of the target anomalous feature coupling pairs; After implementing the targeted decoupling intervention strategy, the patient's action data and physiological data were collected again. The post-intervention coupling strength of the target abnormal feature coupling pair was calculated. The decoupling effectiveness was evaluated based on the degree of decrease in the post-intervention coupling strength relative to the real-time coupling strength. When the decoupling efficiency exceeds a preset threshold, it is determined that the target abnormal feature coupling has effectively intervened in the corresponding emotional state, and an emotional intervention completion signal is output.

[0007] Preferred, preset emotion feature pairs include: The rhythm-respiration coupling pair consists of the characteristics of movement rhythm disorder and the characteristics of respiratory waveform abnormality; Electromyography-skin electrocoagulation pairs are formed by the amplitude characteristics of surface electromyography of specific muscle groups and the level characteristics of skin electrical conductivity response. The visual-cardiac coupling pair consists of the entropy value characteristics of the gaze saccade path and the low-frequency power characteristics of heart rate variability; Gait-blood oxygen coupling pair, consisting of gait stability parameter characteristics and blood oxygen saturation fluctuation characteristics; The posture-endocrine coupling pair consists of trunk posture tilt angle characteristics and cortisol concentration change rate characteristics.

[0008] Preferably, the real-time coupling strength or baseline coupling strength of the preset emotional feature pairs is calculated using a method for quantifying time series synchronicity or a causal analysis method for quantifying information flow. Among them, the methods used to quantify the synchronization of time series include the phase-locked value method or the windowed time-delay mutual information method; Causal analysis methods used to quantify information flow include Granger causality analysis or transfer entropy analysis.

[0009] Preferably, the identification of anomalous feature coupling pairs employs a dynamic threshold method, combined with a personalized emotional dynamics model to refine the judgment logic: The initial judgment threshold is the product of the benchmark coupling strength and the benchmark standard deviation. The real-time coupling strength and the feature sequence used to calculate the real-time coupling strength are input into the personalized emotion dynamics model to obtain a short-term predicted value of the coupling strength. When the short-term forecast value shows a continuous upward trend, the initial judgment threshold is dynamically lowered.

[0010] Preferably, the decoupling strategy mapping table is pre-built in the following way: Define a basic unit library containing various basic intervention movement units; among which, the basic intervention movement units include breathing guidance units, progressive muscle relaxation guidance units, symbolic gesture imitation units, and slow limb extension dance units; Through controlled experiments, the coupling strength influence coefficient of each basic intervention action unit on each type of preset emotional feature pair was measured, and the invasiveness cost and patient acceptance scores were recorded simultaneously. Based on causal analysis, a set of candidate basic intervention action units is selected for each type of abnormal feature coupling pair; Based on the candidate unit set, generate multiple candidate unit combination strategies; Based on the coupling strength influence coefficient, invasive cost, and patient acceptance score of all basic intervention action units included in each candidate unit combination strategy, calculate the comprehensive coupling strength influence coefficient, comprehensive invasive cost, and comprehensive patient acceptance score of the corresponding candidate unit combination strategy. From all candidate unit combination strategies, the basic intervention action unit combination with the negative and largest absolute value of coupling strength influence coefficient, the lowest invasive cost, and the highest patient acceptance score is selected to form a targeted decoupling intervention strategy; The clinical utility value of the targeted decoupling intervention strategy is labeled; where the clinical utility value is equal to the absolute value of the impact coefficient multiplied by the patient acceptance score and then divided by the invasive cost. Based on the type of each abnormal feature coupling pair, the corresponding targeted decoupling intervention strategy, and the corresponding clinical utility value, a mapping relationship is established and stored to form a decoupling strategy mapping table.

[0011] Preferably, through controlled experiments, the coupling strength influence coefficient of each basic intervention action unit on each pre-defined pair of emotional characteristics is measured, including: The subjects were stratified according to their clinical characteristics, and an intervention group that performed targeted basic intervention action units and a control group that performed simulated actions were set up for each stratum; During the experiment, real-time coupling strength data of preset emotion feature pairs were collected simultaneously; Based on the intervention group data, the static difference between the mean coupling strength after intervention and the baseline coupling strength before intervention was calculated, and the coupling strength decay rate was obtained by curve fitting of the real-time coupling strength data. By weighting and fusing the static difference with the coupling strength decay rate, the coupling strength influence coefficient of the corresponding basic intervention action unit on the corresponding preset emotional feature pair is obtained.

[0012] Preferably, the process for determining the overall patient acceptance score includes: Define multidimensional attribute features for each basic intervention action unit; the multidimensional attribute features include unit execution duration, sensory stimulation intensity, required level of interpersonal interaction, and historical execution records; Based on the real-time physiological and motor data of the target patients, predict the immediate acceptance tendency of the target patients to intervention actions with different multidimensional attributes. By fusing historical static patient acceptance scores with immediate acceptance propensity, dynamic acceptance scores for corresponding basic intervention action units are generated. The overall patient acceptance score of the corresponding candidate unit combination strategy is obtained by weighted averaging the dynamic acceptance scores of all basic intervention action units included in each candidate unit combination strategy.

[0013] Preferably, a targeted decoupling intervention strategy is constructed by selecting basic intervention action units with a negative and largest absolute value of coupling strength influence coefficient, the lowest invasive cost, and the highest patient acceptance score, including: The urgency of intervention is determined based on the prediction of future trends in the coupling strength of current abnormal features using a personalized emotion dynamics model. Based on the urgency of the intervention, the weight coefficients of coupling strength influence coefficient, invasive cost, and patient acceptance score in the clinical utility decision function are dynamically configured. When the urgency level is high, increase the weight of the coupling strength influence coefficient and decrease the weight of invasive cost; when the urgency level is low, perform reverse configuration. Multiple candidate unit combination strategies are generated based on the combination of all basic intervention action units; The decision function with dynamically configured weights is used to calculate the utility value of the combined strategies of each candidate unit, and the final targeted decoupling intervention strategy is selected based on the utility value.

[0014] Preferably, when labeling the clinical utility value of targeted decoupling intervention strategies, personalized adaptation factors based on patient characteristics are introduced for optimization calculation, including: Personalized adaptation factors are generated based on the target patient's disease type, baseline emotional level, and historical intervention response data. These personalized adaptation factors are used to dynamically adjust the relative weights of the coupling strength influence coefficient, invasive cost, and patient acceptance score when calculating clinical utility. The coupling strength influence coefficient and patient acceptance score were corrected using personalized adaptation factors, and then combined with invasive costs to calculate the optimized personalized clinical utility value.

[0015] This invention provides a clinical patient emotion recognition and intervention system based on multimodal action analysis, comprising: The baseline spectrum construction module is used to collect action and physiological data of the target patient in a baseline emotional state, calculate the baseline coupling strength of at least one set of preset emotional feature pairs, and form a personalized emotional coupling baseline spectrum by combining the emotional state spatial mapping data; wherein the emotional state spatial mapping data includes the baseline state points and the fitted personalized emotional dynamics model. The anomaly identification and prediction module is used to collect the target patient's action data and physiological data in real time, extract the same preset emotional feature pairs and calculate the real-time coupling strength, compare the real-time coupling strength with the benchmark coupling strength, combine the personalized emotional dynamics model to predict the evolution trend, and identify abnormal feature coupling pairs. The strategy matching and execution module is used to match and execute targeted decoupling intervention strategies from a pre-built decoupling strategy mapping table based on the type of the identified abnormal feature coupling pairs; wherein the targeted decoupling intervention strategy is used to reduce the coupling strength of the target abnormal feature coupling pairs. The efficacy assessment module is used to collect the patient's action and physiological data again after the targeted decoupling intervention strategy is executed, calculate the post-intervention coupling strength of the target abnormal feature coupling pair, and evaluate the decoupling efficacy based on the degree of decrease in the post-intervention coupling strength relative to the real-time coupling strength. The intervention result determination and output module is used to determine that the target abnormal feature coupling has been effectively intervened in the corresponding emotional state when the decoupling efficiency exceeds a preset threshold, and outputs an emotion intervention completion signal.

[0016] The beneficial effects of this invention compared to existing technologies are as follows: it overcomes the fundamental deficiency of existing clinical emotion recognition and intervention technologies, which can only perform superficial emotion labeling and cannot penetrate the underlying physiological-behavioral synergistic disorder mechanism of emotions. By establishing a digital twin of the emotion system that includes a personalized coupling baseline and dynamic model, this method achieves three major technical effects: 1) it upgrades from identifying what emotion is to diagnosing which system synergistic relationships are disordered; 2) the intervention target is precisely positioned from the general symptom relief to decoupling specific abnormal synergistic pairs; 3) it forms a complete intelligent closed loop of monitoring-diagnosis-targeted intervention-quantitative verification, enabling clinical emotion intervention to for the first time possess quantifiable, interpretable, and predictable engineering characteristics.

[0017] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a clinical patient emotion recognition and intervention method based on multimodal action analysis in an embodiment of the present invention. Figure 2 This is an architecture diagram of a clinical patient emotion recognition and intervention system based on multimodal action analysis, as described in an embodiment of the present invention. Detailed Implementation

[0020] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] like Figure 1 As shown, this invention provides an implementation method for a clinical patient emotion recognition intervention method based on multimodal action analysis, comprising: Collect action and physiological data of the target patient in a baseline emotional state, calculate the baseline coupling strength of at least one set of preset emotional feature pairs, and combine them with emotional state spatial mapping data to form a personalized emotional coupling baseline spectrum; wherein the emotional state spatial mapping data includes baseline state points and fitted personalized emotional dynamics models. Real-time acquisition of action and physiological data of target patients, extraction of identical preset emotional feature pairs and calculation of real-time coupling strength, comparison of real-time coupling strength with baseline coupling strength, prediction of evolutionary trends by combining personalized emotion dynamics model, and identification of abnormal feature coupling pairs; Based on the type of the identified anomalous feature coupling pairs, a targeted decoupling intervention strategy is matched from a pre-built decoupling strategy mapping table and executed; wherein the targeted decoupling intervention strategy is used to reduce the coupling strength of the target anomalous feature coupling pairs; After implementing the targeted decoupling intervention strategy, the patient's action data and physiological data were collected again. The post-intervention coupling strength of the target abnormal feature coupling pair was calculated. The decoupling effectiveness was evaluated based on the degree of decrease in the post-intervention coupling strength relative to the real-time coupling strength. When the decoupling efficiency exceeds a preset threshold, it is determined that the target abnormal feature coupling has effectively intervened in the corresponding emotional state, and an emotional intervention completion signal is output.

[0022] In this embodiment, the target patients specifically refer to the clinical patient group who are unable to effectively express their emotions through language due to diseases such as aphasia, autism, or major depression, and who need to rely on non-verbal behavior for analysis.

[0023] In this embodiment, the baseline emotional state is a stable state in which the patient is calm, relaxed, and without significant negative emotions, as confirmed in advance through clinical assessment, and serves as a reference point for establishing a personalized emotional measurement baseline.

[0024] In this embodiment, motion data and physiological data refer to kinematic information such as body joint movement, posture, and gait captured by visual sensors, and signals reflecting autonomic nerve activity such as heart rate, skin conductance, and electromyography collected by biosensors.

[0025] In this embodiment, the preset emotion feature pairs are predefined cross-modal feature combinations in pairs, each pair containing an action feature and a physiological feature, used to quantify specific emotion-related mind-body coordination patterns.

[0026] In this embodiment, calculating the baseline coupling strength of at least one set of preset emotional feature pairs refers to calculating the numerical value of the degree of correlation between two feature sequences in each set of preset feature pairs using a specific mathematical method under the baseline emotional state.

[0027] In this embodiment, the baseline coupling strength of the preset emotional feature pair refers to the value calculated above, which represents the normal intensity level of a specific mind-body coordination pattern of the patient in a calm state.

[0028] In this embodiment, the personalized emotion coupling baseline spectrum is a set composed of the baseline coupling strength values ​​and statistical distribution characteristics of all preset emotion feature pairs of the patient, which characterizes the personalized features of the patient's emotional system under normal conditions.

[0029] In this embodiment, the baseline state point is a data point in a multidimensional mathematical space used to comprehensively represent emotions, representing the location of the patient's baseline emotional state.

[0030] In this embodiment, the fitted personalized emotion dynamics model is a predictive model built based on multiple observation data of the patient near the baseline state, which can describe how the patient's emotional state changes and evolves over time in that space.

[0031] The core of the personalized emotion dynamics model in this invention is a recurrent neural network prediction model that can predict changes in the coupling strength of emotions in the next minute based on the patient's emotional coupling strength data from the past few minutes. Specifically, it is preferably constructed using a long short-term memory network.

[0032] How to prepare training data (to build a model for a specific patient): In the initial stage, in addition to recording baseline data when the patient is calm, medical staff will guide the patient to engage in some safe activities that can induce mild emotional fluctuations (such as watching short videos, listening to different styles of music, or having simple conversations).

[0033] During this process, the system continuously records the coupling strength values ​​of all preset emotional characteristic pairs of the patient, forming a curve that fluctuates over time.

[0034] The curve is divided into many continuous small segments. For example, each segment is 60 consecutive seconds of data (one point per second, for a total of 60 points), and the value of the second immediately following this segment is the target answer that the model is expected to learn to predict.

[0035] In this way, we obtain hundreds or thousands of pairs of data from the past 60 seconds and the answer for the next second, which can be used as material to train the model.

[0036] How to train the model: All the data prepared in the previous step are paired and input into an initialized long short-term memory network.

[0037] The network's task is to learn a pattern: when it sees a pattern of changes in coupling strength over the past 60 seconds, it can guess the value for the next second as accurately as possible.

[0038] The training process involves continuously adjusting the network's internal parameters to minimize the average error between its predictions and the actual answers. This process is automated by a computer using an optimization algorithm called Adam.

[0039] When the prediction error no longer decreases significantly, the training is complete, and a personalized emotion dynamics model specific to the patient is obtained.

[0040] How to use the model for prediction: During actual monitoring, the system will always save the patient's coupling strength data for the most recent 60 seconds.

[0041] Every second, the system inputs the latest 60 seconds of data into the trained personal model.

[0042] The model will immediately output a prediction of the coupling strength value for the next second.

[0043] By continuously calling the model, a predicted future trend line can be obtained, which can determine whether the coupling strength is about to rise, fall, or remain stable.

[0044] In this embodiment, real-time acquisition of the target patient's action and physiological data refers to continuously or synchronously acquiring the patient's current action and physiological signal stream in daily or clinical observation scenarios.

[0045] In this embodiment, extracting the same preset emotion feature pairs and calculating the real-time coupling strength means extracting feature values ​​from the real-time data stream according to the same definition and calculating the correlation strength of these features at the current moment.

[0046] In this embodiment, the real-time coupling strength is compared with the baseline coupling strength, and the evolution trend is predicted by combining the personalized emotion dynamics model to identify abnormal feature coupling pairs. This means first determining whether the real-time value deviates significantly from the individual baseline, and then using the dynamics model to predict whether the deviation will intensify. The results of both are combined to determine which feature pairs have experienced personalized abnormalities in their synergistic relationship.

[0047] In this embodiment, the personalized emotion dynamics model predicts the evolution trend by inputting the patient's current real-time characteristic sequence into their personal dynamics model, and the model outputs the possible direction and rate of change of the key coupling strength in the near future.

[0048] In this embodiment, the pre-built decoupling strategy mapping table is a database established through prior systematic experiments. It stores the correspondence between different types of abnormal feature coupling pairs and the optimal intervention strategy that has been verified to effectively reduce the abnormality.

[0049] In this embodiment, the targeted decoupling intervention strategy is a specific intervention plan retrieved from the above mapping table and composed of a series of standardized basic intervention action units.

[0050] In this embodiment, the targeted decoupling intervention strategy is used to reduce the coupling strength of the target anomalous feature coupling pair, which means that the design goal and verification effect of this strategy are specifically to weaken the excessive correlation between the identified specific anomalous feature pair.

[0051] In this embodiment, calculating the post-intervention coupling strength of the target anomalous feature coupling pair refers to re-collecting data and calculating the association strength of the same anomalous feature pair during an evaluation period after the intervention strategy is implemented.

[0052] In this embodiment, evaluating decoupling effectiveness based on the degree of decrease in coupling strength after intervention relative to real-time coupling strength means comparing the coupling strength value after intervention with the real-time value before intervention, and quantifying the direct effect of the intervention by calculating the magnitude of the decrease.

[0053] In this embodiment, the preset threshold is a pre-set numerical limit used to determine whether the decoupling efficiency meets the clinically effective standard.

[0054] In this embodiment, determining that the target abnormal feature coupling has been effectively intervened in the corresponding emotional state and outputting an emotional intervention completion signal means that when the decoupling effectiveness evaluation result exceeds a preset threshold, the system makes a technical judgment that the intervention for the abnormal emotional pattern has been successful and generates a clear signal that can be recorded or notified.

[0055] To accurately capture key collaborative disorder patterns across physiological and behavioral systems closely associated with different emotional states, pre-defined pairs of emotional characteristics are proposed, including: The rhythm-respiration coupling pair consists of the characteristics of movement rhythm disorder and the characteristics of respiratory waveform abnormality; Electromyography-skin electrocoagulation pairs are formed by the amplitude characteristics of surface electromyography of specific muscle groups and the level characteristics of skin electrical conductivity response. The visual-cardiac coupling pair consists of the entropy value characteristics of the gaze saccade path and the low-frequency power characteristics of heart rate variability; Gait-blood oxygen coupling pair, consisting of gait stability parameter characteristics and blood oxygen saturation fluctuation characteristics; The posture-endocrine coupling pair consists of trunk posture tilt angle characteristics and cortisol concentration change rate characteristics.

[0056] In this embodiment, the movement rhythm disorder characteristic refers to the quantitative index of irregular and unstable movement rhythm calculated by analyzing the time series data of the patient's body movements (such as hand tremors and pacing frequency). The higher the value, the more chaotic the movement.

[0057] In this embodiment, the abnormality characteristic of respiratory waveform refers to the quantitative indicators of respiratory depth and rhythm uniformity that deviate from the normal pattern by analyzing the patient's respiratory signals (usually obtained by chest and abdominal movement or airflow sensors), reflecting the degree of rapid, shallow or irregular breathing.

[0058] In this embodiment, the rhythm-respiration coupling pair, composed of the characteristics of motor rhythm disorder and the characteristics of respiratory waveform abnormality, is used to quantify the synergistic change relationship between the patient's motor system anxiety and respiratory system disorder. The synchronous enhancement of both is often a typical manifestation of anxiety or agitation.

[0059] In this embodiment, the surface electromyography amplitude characteristics of a specific muscle group refer to the average intensity or root mean square value of the electromyography signal collected by electrodes attached to the surface of a specific muscle group (such as the trapezius or masseter). An increase in this characteristic value reflects continuous muscle tension or stiffness.

[0060] In this embodiment, the skin conductance response level characteristic refers to the baseline level of skin conductivity measured by a skin conductance sensor. An increase in this characteristic value reflects sympathetic nerve excitation and increased sweating, which is a direct indicator of physiological arousal or tension.

[0061] In this embodiment, the electromyography-skin electrical coupling pair, which consists of the surface electromyography amplitude characteristics of specific muscle groups and the skin electrical conductivity response level characteristics, is used to quantify the correlation between somatic muscle tension and the degree of excitation of the autonomic nervous system. Strong coupling between the two often corresponds to a high stress state or somatization symptoms.

[0062] In this embodiment, the entropy value of the gaze saccade path refers to a mathematical index that tracks the patient's gaze movement trajectory using an eye tracker and calculates the disorder or unpredictability of the path. The higher the entropy value, the more wandering the gaze and the less focused the attention.

[0063] In this embodiment, the low-frequency power characteristic of heart rate variability refers to the power intensity of heart rate interval fluctuations extracted from electrocardiogram signals in the low-frequency band (approximately 0.04-0.15Hz). This characteristic is related to sympathetic nerve regulation and emotional load, and its abnormal changes can reflect emotional stress or decreased regulatory capacity.

[0064] In this embodiment, the visual-cardiac coupling pair, which consists of the entropy value of the gaze saccade path and the low-frequency power value of heart rate variability, is used to quantify the intrinsic link between the degree of visual attention distraction and cardiac autonomic nervous system regulation disorder. Abnormal correlation between the two often occurs during hypervigilance or cognitive-emotional disorder.

[0065] In this embodiment, the gait stability parameter characteristic refers to the walking smoothness and coordination index calculated by analyzing parameters such as stride length, stride speed, stride width, and body swing during walking. A decrease in the value indicates a slow, dragging, or unstable gait.

[0066] In this embodiment, the blood oxygen saturation fluctuation characteristics refer to the fluctuation amplitude or frequency of the blood oxygen saturation value monitored by the pulse oximeter over a certain period of time. Abnormal fluctuations may reflect respiratory rhythm disorders or cardiovascular regulation abnormalities.

[0067] In this embodiment, the gait-blood oxygen coupling pair, which consists of gait stability parameter characteristics and blood oxygen saturation fluctuation characteristics, is used to quantify the synergistic degradation between decreased exercise vitality, decreased posture control, and unstable body oxygenation status, which is commonly seen in states of depression, fatigue, or overall functional decline.

[0068] In this embodiment, the trunk posture tilt angle feature refers to the tilt angle of the patient's trunk relative to the vertical direction and its changes obtained by posture sensors or visual analysis. Persistent abnormal forward tilting, side tilting or curled-up postures may reflect low mood, avoidance or physical discomfort.

[0069] In this embodiment, the cortisol concentration change rate characteristic refers to the rate of change of the stress hormone cortisol level over time obtained by detecting saliva, sweat or blood. This characteristic reflects the degree of activation of the hypothalamus-pituitary-adrenal axis, and an abnormally high change rate suggests chronic or acute stress.

[0070] In this embodiment, the posture-endocrine coupling pair, which consists of trunk posture tilt angle characteristics and cortisol concentration change rate characteristics, is used to quantify the deep association between withdrawal, defensive body posture and the dynamics of stress hormone secretion in the body. It is an important cross-system biological behavioral marker of psychosomatic stress load and depressive tendency.

[0071] To objectively and reliably quantify the dynamic correlation strength between different modal features and accurately reflect the intrinsic synergistic state of the emotional system, this paper proposes to calculate the real-time coupling strength or benchmark coupling strength of preset emotional feature pairs, and adopts methods for quantifying time series synchronicity or causal analysis methods for quantifying information flow. Among them, the methods used to quantify the synchronization of time series include the phase-locked value method or the windowed time-delay mutual information method; Causal analysis methods used to quantify information flow include Granger causality analysis or transfer entropy analysis.

[0072] In this embodiment, the calculation of the real-time coupling strength or baseline coupling strength of the preset emotional feature pairs, using a method for quantifying time series synchronicity or a causal analysis method for quantifying information flow, means that for any set of preset emotional feature pairs (such as action rhythm and breathing waveform), one of two types of mathematical tools is selected to accurately calculate the correlation strength between them: one type is to assess the consistency (synchronicity) of the two signals in the fluctuation rhythm, and the other type is to determine whether the change of one signal precedes and affects the other signal in time sequence (causality).

[0073] In this embodiment, the methods for quantifying the synchronization of time series include the phase-locked value method or the windowed time-delay mutual information method. The phase-locked value method measures the degree of rhythmic synchronization by calculating the stability of the instantaneous phase difference between two signals; the more stable the phase difference, the higher the phase-locked value, indicating stronger synchronization. The windowed time-delay mutual information method calculates the amount of shared information between two signals at different time delays within a sliding time window, taking the maximum value as the synchronization strength. This method can capture nonlinear synchronization relationships.

[0074] In this embodiment, the causal analysis methods used to quantify information flow include Granger causality analysis or the transfer entropy method. Granger causality analysis is based on a linear prediction model, determining that if adding historical information of signal A can better predict the future value of signal B, then A is considered a cause of B in a Granger sense. The transfer entropy method is a nonlinear causal measure based on information theory; it directly calculates the amount of information transferred from signal A to signal B, effectively revealing the causal driving relationships of nonlinear systems.

[0075] In this embodiment, how to select a suitable coupling strength calculation method for different feature pairs: For signals with obvious fluctuations in rhythm, such as movement rhythm and breathing, it is recommended to calculate their phase synchronization (e.g., using phase-locked loop method) to see if their peaks and troughs are synchronized.

[0076] For signals with potentially more complex relationships, such as electromyography (EMG) and electrodermatology (EDS), it is recommended to calculate their mutual information at different time delays to capture nonlinear correlations.

[0077] For signals like body posture and cortisol that you want to explore which influences which, Granger causality analysis is recommended to statistically determine whether changes in posture precede changes in hormones over time.

[0078] To achieve early warning and dynamic, accurate identification of abnormal emotional evolution, and to avoid missed or false judgments caused by a single static threshold, this paper proposes a dynamic threshold method for identifying abnormal feature coupling pairs, combined with a personalized emotional dynamics model to correct the judgment logic. The initial judgment threshold is the product of the benchmark coupling strength and the benchmark standard deviation. The real-time coupling strength and the feature sequence used to calculate the real-time coupling strength are input into the personalized emotion dynamics model to obtain a short-term predicted value of the coupling strength. When the short-term forecast value shows a continuous upward trend, the initial judgment threshold is dynamically lowered.

[0079] In this embodiment, the personalized emotion dynamics model is a mathematical prediction system built based on the patient's historical emotion data. By analyzing the evolution trajectory of the patient's past emotional states in a multidimensional feature space, the model learns the inherent laws and patterns of their emotional fluctuations, thereby enabling it to predict the possible direction and trend of changes in their emotional state (especially the coupling strength of key emotional feature pairs) in the near future based on current observation data.

[0080] In this embodiment, using the product of the baseline coupling strength and the baseline standard deviation as the initial judgment threshold means that when judging the abnormality of a specific emotional feature pair, a judgment standard is first set. This standard is centered on the normal correlation strength (baseline coupling strength) of the feature pair in the patient's calm state, plus a certain multiple (such as 1.5 times or 2 times) of its normal fluctuation range (baseline standard deviation) as the boundary line. If the real-time coupling strength exceeds this boundary line, it is initially judged as abnormal.

[0081] In this embodiment, the baseline standard deviation refers to the statistical dispersion index calculated from the coupling strength data of a specific emotional characteristic pair obtained from multiple measurements of the patient under a baseline emotional state (calm period). It represents the natural fluctuation range of the patient's coordination pattern under normal conditions and is a personalized benchmark for measuring how far a deviation is considered significantly abnormal.

[0082] In this embodiment, the feature sequence used to calculate the real-time coupling strength refers to the sequence of observations of two (or more) emotional features arranged in chronological order within a recent time window, which is the basis for calculating the real-time coupling strength. For example, calculating the real-time coupling strength between the degree of motor rhythm disorder and the degree of respiratory waveform abnormality requires the specific numerical sequences of these two features over a continuous period of time prior to the current moment.

[0083] In this embodiment, the real-time coupling strength and the feature sequence used to calculate the real-time coupling strength are input into a personalized emotion dynamics model to obtain a short-term predicted value of the coupling strength. This means that the correlation strength value of the feature pair observed at the current moment, as well as the recent original data sequence of each feature that forms the basis of the strength calculation, are input together into the prediction model established for the patient. The model comprehensively analyzes the current correlation state and the dynamics of its constituent features, and outputs an estimate of the value that the coupling strength may reach in the next few seconds or minutes.

[0084] In this embodiment, when the short-term predicted value shows a continuous upward trend, the initial judgment threshold is dynamically lowered. This means that if the prediction model's multiple consecutive predictions show that the coupling strength of the abnormal feature pair will continue to increase and deteriorate significantly in the short term, the system will automatically adopt a more conservative (i.e., more sensitive) early warning strategy. Specifically, the initial judgment threshold is temporarily lowered based on the original threshold, allowing the system to identify and issue an alarm before the coupling strength rises to the originally predetermined significant abnormal level, thereby achieving early warning and intervention.

[0085] In this embodiment, how is the dynamic threshold adjusted? Set an initial alarm line: Using the average coupling strength when the patient is at rest as the center, draw an initial alarm line two standard deviations upwards. Exceeding this line is considered an initial abnormality.

[0086] Determining a worsening trend: The system uses the aforementioned personalized emotion dynamics model for real-time prediction. If the model's predictions are higher in each of the last three predictions, it is determined that there is a continuously rising worsening trend.

[0087] Lowering the alert threshold: Once a deteriorating trend is detected, the system will immediately lower the alert threshold, for example, from the average plus two standard deviations to the average plus 1.5 standard deviations. This allows the system to trigger warnings and interventions earlier, before the situation becomes very serious.

[0088] To systematically construct an interpretable, quantifiable, and clinically optimal abnormal pattern-intervention strategy precise mapping knowledge base, a decoupling strategy mapping table is proposed to be pre-constructed in the following way: Define a basic unit library containing various basic intervention movement units; among which, the basic intervention movement units integrate dance therapy and narrative nursing theories, including breathing guidance units, progressive muscle relaxation guidance units, symbolic gesture imitation units, and slow limb extension dance units; Through controlled experiments, the coupling strength influence coefficient of each basic intervention action unit on each type of preset emotional feature pair was measured, and the invasiveness cost and patient acceptance scores were recorded simultaneously. Based on causal analysis, a set of candidate basic intervention action units is selected for each type of abnormal feature coupling pair; Based on the candidate unit set, generate multiple candidate unit combination strategies; Based on the coupling strength influence coefficient, invasive cost, and patient acceptance score of all basic intervention action units included in each candidate unit combination strategy, calculate the comprehensive coupling strength influence coefficient, comprehensive invasive cost, and comprehensive patient acceptance score of the corresponding candidate unit combination strategy. From all candidate unit combination strategies, the basic intervention action unit combination with the negative and largest absolute value of coupling strength influence coefficient, the lowest invasive cost, and the highest patient acceptance score is selected to form a targeted decoupling intervention strategy; The clinical utility value of the targeted decoupling intervention strategy is labeled; where the clinical utility value is equal to the absolute value of the impact coefficient multiplied by the patient acceptance score and then divided by the invasive cost. Based on the type of each abnormal feature coupling pair, the corresponding targeted decoupling intervention strategy, and the corresponding clinical utility value, a mapping relationship is established and stored to form a decoupling strategy mapping table.

[0089] In this embodiment, the breathing guidance unit, the progressive muscle relaxation guidance unit, the symbolic gesture imitation unit, and the slow limb extension dance unit are standardized and executable action modules constituting a targeted intervention strategy. The breathing guidance unit refers to guiding patients to perform breathing exercises in specific patterns (such as diaphragmatic breathing, 4-7-8 breathing) through audiovisual cues. The progressive muscle relaxation guidance unit refers to systematically guiding patients to alternately tense and relax different muscle groups. The symbolic gesture imitation unit refers to guiding patients to imitate simplified gestures representing specific emotions or intentions (such as opening arms to represent acceptance). The slow limb extension dance unit refers to guiding patients to perform slow, fluid limb stretching and stroking movements to soothing music or rhythm.

[0090] In this embodiment, a controlled experiment refers to a systematic test conducted in a standardized laboratory or clinical setting, recruiting representative patient samples to systematically examine the impact of each basic intervention action unit on various pre-defined emotional characteristic pairs. The experiment typically includes a control group to strictly control variables, ensuring the scientific measurement and recording of effect data for each unit.

[0091] In this embodiment, the coupling strength influence coefficient of each basic intervention action unit on each type of preset emotional feature pair is a value obtained through controlled experiments. This coefficient represents the amount of change in the coupling strength of the target emotional feature pair relative to the baseline after executing the specific intervention unit. A negative value indicates that the unit can effectively weaken (decouple) the abnormal synergy, and the larger the absolute value, the stronger the decoupling effect.

[0092] In this embodiment, invasive cost is a comprehensive assessment value used to quantify the degree to which the implementation of a particular intervention unit or combination of strategies disrupts a patient's daily activities, autonomy, and clinical care processes. Lower costs indicate that the intervention is more convenient and less disruptive to the patient.

[0093] Invasive cost is a metric used to measure the disruption an intervention causes to patients and care processes.

[0094] It is scored mainly from three aspects: ① duration (the longer the time, the higher the cost), ② manpower required (whether one-on-one guidance from a professional therapist is required), and ③ equipment dependence (whether additional equipment is required).

[0095] Each aspect is scored from 0 to 1 (e.g., 0.2 points for less than 5 minutes and 0.8 points for more than 10 minutes), and then the scores are added together with a weighting of 5:3:2 (time is the most important factor, followed by manpower, and then equipment is the least important factor) to obtain a comprehensive cost score between 0 and 1. The higher the score, the greater the disruption.

[0096] In this embodiment, the patient acceptance score is a numerical value derived from historical implementation data, patient feedback, and a real-time predictive model. It is used to assess the target patient's willingness and comfort level in implementing a specific intervention unit or combination of strategies. The higher the score, the more likely the patient is to cooperate and complete the intervention well.

[0097] In this embodiment, based on causal analysis, a set of candidate basic intervention action units is selected for each type of anomalous feature coupling. This means using causal inference algorithms (such as Granger causality and transitive entropy) to identify those units from all basic units that have been proven to have a significant negative causal impact on a specific type of anomalous feature coupling, and listing them as effective candidate intervention options for that type of anomalousness.

[0098] In this embodiment, the strategy of generating multiple candidate unit combinations based on the candidate unit set refers to arranging and combining the selected candidate basic intervention action units in different ways to form a series of possible composite intervention schemes for further evaluation and selection.

[0099] In this embodiment, the comprehensive coupling strength influence coefficient, comprehensive invasive cost, and comprehensive patient acceptance score of the corresponding candidate unit combination strategy are calculated based on the coupling strength influence coefficient, invasive cost, and patient acceptance score of all basic intervention action units included in each candidate unit combination strategy. This means that by using specific aggregation rules (such as weighted average or synergistic effect model), the individual parameters of all units in the combination are merged to calculate three comprehensive indicators representing the overall performance of the entire combination strategy.

[0100] In this embodiment, the basic intervention action unit combination selected from all candidate unit combination strategies, which has the largest negative and absolute value of coupling strength influence coefficient, the lowest invasive cost, and the highest patient acceptance score, constitutes the targeted decoupling intervention strategy. This is a multi-objective optimization decision-making process, aiming to find the unique combination that achieves the best balance or optimal trade-off among all feasible options in terms of decoupling effect (maximizing the negative influence coefficient), implementation convenience (minimizing the invasive cost), and patient friendliness (maximizing the acceptance score).

[0101] In this embodiment, clinical utility values ​​are assigned to targeted decoupling intervention strategies. The clinical utility value is calculated by multiplying the absolute value of the impact coefficient by the patient acceptance score and then dividing by the invasive cost. This gives each ultimately selected optimal intervention strategy a comprehensive quantitative score. This formula reflects the core logic of clinical decision-making: pursuing high efficacy (large absolute value of the impact coefficient) and high acceptance (high score), while minimizing implementation costs (low invasive cost). A higher utility value indicates greater overall clinical value of the strategy.

[0102] In this embodiment, a mapping relationship is established and stored based on the type of each abnormal feature coupling pair, the corresponding targeted decoupling intervention strategy, and the corresponding clinical utility value, forming a decoupling strategy mapping table. This means that the final results of the above analysis and decision-making are structurally saved in a database. Using the abnormal feature coupling pair type as the query key, this database can directly retrieve the optimal targeted decoupling intervention strategy tailored to it, along with the expected clinical utility value of that strategy, thereby achieving rapid and accurate intervention matching in practical applications.

[0103] In this embodiment, the construction of the decoupling strategy mapping table and the core calculation logic are as follows: Grouped experiment: Patients with similar conditions were invited to participate in the study. Patients were randomly divided into two groups: Group A actually performed a specific intervention (such as guided breathing), while Group B performed a seemingly similar but ultimately ineffective placebo action.

[0104] Data collection results: During and after the action, the intensity changes of the target emotional feature pairs (such as electromyography-cortical coupling pairs) were continuously measured.

[0105] Calculate two key metrics: Static difference: This calculates how much the average coupling strength decreases over a period of time after the action is completed, compared to before the action began. The greater the decrease, the more significant the effect.

[0106] Attenuation rate: This analyzes how quickly the coupling strength decreases during the execution of the action. The faster the decrease, the more rapidly the effect takes hold.

[0107] Composite Effect Coefficient: The static difference and decay rate mentioned above are weighted and added together in a 70 / 30 ratio (effect accounts for 70%, speed accounts for 30%) to obtain a comprehensive effect coefficient. This coefficient is negative, and the larger the absolute value, the stronger and faster the effect of the action unit in weakening this specific emotional coupling pair.

[0108] Define movement attributes: Define several easily quantifiable attributes for each basic intervention movement unit (such as the slow limb extension dance unit), including: ① how long it takes to perform, ② how strong the stimulation to the eyes and ears is, and ③ whether it requires one-on-one guidance from a therapist.

[0109] Predicting Immediate Willingness: The system has a built-in small intelligent prediction module. This module analyzes the patient's current heart rate, skin sweating, and restlessness in real time, combining these signals with the attributes of the recommended action (such as duration and intensity) to predict the likelihood that the patient will cooperate with the action at that moment. This prediction result is a value between 0 and 100%.

[0110] In conjunction with historical performance: The system will also retrieve the patient's historical records to see what his / her average completion rate was for similar actions in the past (for example, if he / she completed 8 out of the past 10 times, the historical score would be 80%).

[0111] Calculate the dynamic score: Take half the weight of the static score based on historical records (e.g., 80%) and the immediate willingness predicted based on the current state (e.g., a predicted willingness of 60%), and average them to obtain the dynamic acceptance score of the action at the current moment (70% in this example).

[0112] Calculating the score of a combined strategy: If an intervention strategy consists of multiple small action units, the overall acceptability score of the strategy is calculated by weighting the dynamic scores of each small action according to the proportion of time it occupies.

[0113] To obtain more reliable unit efficacy data that simultaneously reflects the intensity and timeliness of intervention effects through rigorous experimental design, this paper proposes a controlled experiment to measure the coupling strength influence coefficient of each basic intervention action unit on each pre-defined pair of emotional characteristics, including: The subjects were stratified according to their clinical characteristics, and an intervention group that performed targeted basic intervention action units and a control group that performed simulated actions were set up for each stratum; During the experiment, real-time coupling strength data of preset emotion feature pairs were collected simultaneously; Based on the intervention group data, the static difference between the mean coupling strength after intervention and the baseline coupling strength before intervention was calculated, and the coupling strength decay rate was obtained by curve fitting of the real-time coupling strength data. By weighting and fusing the static difference with the coupling strength decay rate, the coupling strength influence coefficient of the corresponding basic intervention action unit on the corresponding preset emotional feature pair is obtained.

[0114] In this embodiment, the experimental subjects are stratified according to clinical characteristics, and an intervention group performing the targeted basic intervention action unit and a control group performing the simulated action are set up for each stratum. This means that when conducting a controlled experiment, patients are first grouped according to key clinical characteristics such as disease type, severity, and age to ensure that the patients in each group have relatively consistent backgrounds. Within each characteristic stratum, patients are randomly assigned to two groups: the intervention group actually performs the target basic intervention action unit to be evaluated; the control group performs a simulated action that is similar to the intervention unit in appearance and duration, but has been identified as having no actual emotion regulation effect (i.e., placebo control), thereby isolating and quantifying the specific effects brought about by the intervention unit itself.

[0115] In this embodiment, during the experiment, the real-time coupling strength data of the preset emotional feature pairs are collected simultaneously. This means that during the entire process of the intervention group performing the target intervention unit and the control group performing the simulated action, the real-time correlation strength value of the specific emotional feature pairs (such as electromyography-dermal coupling pairs) of the target patients is continuously and synchronously recorded and calculated to form an intensity curve that changes over time, rather than only taking a single measurement before and after the intervention.

[0116] In this embodiment, based on the intervention group data, the static difference between the mean coupling strength after intervention and the baseline coupling strength before intervention is calculated, and the coupling strength decay rate is obtained by curve fitting of the real-time coupling strength data. This means performing two aspects of analysis on the intervention group data: First, the average value of the coupling strength over a period of time after the intervention ends is calculated, and the average value of the stable period before the intervention begins is subtracted to obtain a static difference reflecting the overall level change after intervention; Second, the real-time coupling strength curve collected during the intervention is fitted with a mathematical model (such as an exponential decay model), and the parameter characterizing the rate of intensity decline, i.e., the decay rate, is extracted from the fitted model, which reflects the dynamic process of the intervention taking effect.

[0117] In this embodiment, the static difference and the coupling strength decay rate are weighted and fused to obtain the coupling strength influence coefficient of the corresponding basic intervention action unit on the corresponding preset emotional feature pair. This means that, to comprehensively evaluate the effectiveness of the intervention unit, the two indicators—the static difference representing the magnitude of the effect and the decay rate representing the speed of the effect—are combined and calculated according to preset importance weights (e.g., influence coefficient = static difference × weight A + decay rate × weight B) to ultimately generate a single, comprehensive coupling strength influence coefficient. The larger the negative value of this coefficient, the better and faster the intervention unit's effect on reducing the abnormal coupling of that specific emotional feature pair.

[0118] To dynamically predict and quantify patients' immediate willingness to accept specific interventions, and to make strategy selection more aligned with patients' current psychophysiological state, a process for determining a comprehensive patient acceptance score is proposed, including: Define multidimensional attribute features for each basic intervention action unit; the multidimensional attribute features include unit execution duration, sensory stimulation intensity, required level of interpersonal interaction, and historical execution records; Based on the real-time physiological and motor data of the target patients, predict the immediate acceptance tendency of the target patients to intervention actions with different multidimensional attributes. By fusing historical static patient acceptance scores with immediate acceptance propensity, dynamic acceptance scores for corresponding basic intervention action units are generated. The overall patient acceptance score of the corresponding candidate unit combination strategy is obtained by weighted averaging the dynamic acceptance scores of all basic intervention action units included in each candidate unit combination strategy.

[0119] In this embodiment, multidimensional attribute features are defined for each basic intervention action unit. These multidimensional attribute features include unit execution time, sensory stimulation intensity, required level of interpersonal interaction, and historical execution records. This refers to establishing a quantitative description system for each unit to standardize the description and comparison of the characteristics of different intervention units. This system covers four key dimensions: the length of time required to execute the unit; the degree of stimulation of the patient's senses such as vision, hearing, or touch; whether and to what extent the implementation of the unit requires direct participation and interaction from medical staff or caregivers; and the success rate, completion rate, or compliance data of the unit in the past execution by this patient or similar patient groups.

[0120] In this embodiment, the unit execution time, sensory stimulation intensity, required level of interpersonal interaction, and historical execution records are the specific components of the multidimensional attribute features. Execution time directly affects the patient's patience and fatigue; sensory stimulation intensity relates to whether the patient feels uncomfortable or overloaded; the required level of interpersonal interaction reflects the degree of dependence on nursing resources and the patient's social pressure; and historical execution records provide objective data on how well the unit is accepted and completed by the patient in practice.

[0121] In this embodiment, predicting the target patient's immediate acceptance tendency for intervention actions with different multidimensional attributes based on the target patient's real-time physiological and motor data means that the system uses a machine learning model to analyze the patient's current physiological signals (such as heart rate variability and skin conductance) and motor state (such as activity level and posture) to infer in real time the patient's possible response to intervention actions with different attribute combinations. For example, when a patient exhibits high physiological arousal and agitated behavior, the model may predict a lower acceptance tendency for interventions with prolonged duration and high stimulation intensity.

[0122] In this embodiment, the historical static patient acceptance score refers to a baseline score calculated based on the patient's overall feedback and compliance data across all intervention units. It reflects the patient's relatively stable, long-term preferences and tolerance characteristics and is a background value that does not change with real-time status.

[0123] In this embodiment, a dynamic acceptance score for a corresponding basic intervention action unit is generated by fusing historical static patient acceptance scores with immediate acceptance propensity. This involves combining a static score representing long-term preferences with an immediate propensity prediction value reflecting the current state through weighting or other fusion algorithms to produce an updated acceptance assessment value for the patient at the current moment regarding that specific intervention unit. This allows the score to dynamically respond to real-time changes in the patient's state.

[0124] In this embodiment, a weighted average of the dynamic acceptance scores of all basic intervention action units included in each candidate unit combination strategy is used to obtain the comprehensive patient acceptance score for the corresponding candidate unit combination strategy. This means that for a candidate combination strategy composed of multiple basic units, the overall patient acceptance score is obtained by weighting the dynamic acceptance scores of each unit within the combination according to the relative importance or execution time percentage of each unit in the combination. This comprehensive score represents the system's prediction of the patient's overall willingness to cooperate with the complete combination strategy.

[0125] To intelligently adjust decision priorities based on the urgency of emotional crises and achieve a dynamic balance between maximizing effectiveness and optimizing experience, a targeted decoupling intervention strategy is proposed, which selects basic intervention action units with the largest negative and absolute coupling strength influence coefficient, the lowest invasive cost, and the highest patient acceptance score. This strategy includes: The urgency of intervention is determined based on the prediction of future trends in the coupling strength of current abnormal features using a personalized emotion dynamics model. Based on the urgency of the intervention, the weight coefficients of coupling strength influence coefficient, invasive cost, and patient acceptance score in the clinical utility decision function are dynamically configured. When the urgency level is high, increase the weight of the coupling strength influence coefficient and decrease the weight of invasive cost; when the urgency level is low, perform reverse configuration. The decision function with dynamically configured weights is used to calculate the utility value of the combined strategies of each candidate unit, and the final targeted decoupling intervention strategy is selected based on the utility value.

[0126] In this embodiment, the personalized emotion dynamics model's prediction of the future trend of the current abnormal feature coupling strength refers to inputting the patient's currently observed abnormal feature coupling strength data and its recent changes into their personal emotion dynamics model. The model then uses mathematical calculations to output an estimate of how this coupling strength will evolve (whether it will continue to rise, stabilize, or possibly decline) and the rate of evolution over a short period of time (e.g., the next few minutes).

[0127] In this embodiment, the urgency of intervention is determined based on the prediction of future trends in the coupling strength of current anomalous features by a personalized emotion dynamics model. This means that the system categorizes the urgency of intervention measures based on the prediction results. For example, if the prediction shows that the anomalous coupling strength will deteriorate rapidly and continuously, it is classified as high urgency, meaning that immediate and strong intervention is required; if the prediction shows that it will remain stable or change slowly, it is classified as low urgency, allowing more time to choose a gentler, more comfortable intervention.

[0128] In this embodiment, the weighting coefficients of the coupling strength influence coefficient, invasive cost, and patient acceptance score in the clinical utility decision function are dynamically configured based on the urgency of the intervention. This means that the system's decision logic flexibly adjusts according to the urgency of the intervention. The clinical utility decision function is typically a mathematical formula that combines the three parameters: coupling strength influence coefficient (effect), invasive cost (cost), and patient acceptance score (experience). The system pre-sets a weighting table, assigning different importance weights to these three parameters based on different levels of urgency.

[0129] In this embodiment, when the urgency level is high, the weight of the coupling strength influence coefficient is increased, while the weight of invasive costs is decreased; when the urgency level is low, the reverse configuration is applied, reflecting the clinical wisdom of treating the symptoms in urgent situations and addressing the root cause in less urgent ones. In highly urgent situations, decision-making prioritizes rapid and powerful decoupling effects (therefore increasing the weight of effects), and can temporarily tolerate slightly greater disturbance or inconvenience to patients (therefore decreasing the weight of consideration for invasive costs). In less urgent situations, decision-making tends to choose interventions that are less disruptive to patients, provide a better experience, and are more acceptable to patients (therefore increasing the weight of invasive costs and acceptability scores), and the requirements for the strength of immediate effects can be relatively relaxed (therefore decreasing the weight of effects).

[0130] In this embodiment, the decision function with dynamically configured weights refers to the clinical utility calculation formula after updating the weight coefficients based on the currently determined intervention urgency. It is a variable, and its specific form (the weights of each parameter) changes with the context.

[0131] In this embodiment, the utility value of each candidate unit combination strategy is calculated using a decision function with dynamically configured weights. The final targeted decoupling intervention strategy is then selected based on the utility value. This means that the comprehensive parameters of each candidate intervention combination strategy (i.e., its comprehensive coupling strength influence coefficient, comprehensive invasiveness cost, and comprehensive patient acceptance score) are substituted into the decision function, whose weights are dynamically determined according to the current urgency level, to calculate the utility value of each strategy in the current situation. The system ultimately selects the candidate combination strategy with the highest calculated utility value as the targeted decoupling intervention strategy to be actually implemented.

[0132] To deeply integrate individual patient differences into the decision-making algorithm, and to ensure that the final strategy selection best matches the patient's unique disease characteristics and response history, this paper proposes to introduce personalized adaptation factors based on patient characteristics for optimization calculation when labeling the clinical utility values ​​of targeted decoupling intervention strategies. These factors include: Personalized adaptation factors are generated based on the target patient's disease type, baseline emotional level, and historical intervention response data. These personalized adaptation factors are used to dynamically adjust the relative weights of the coupling strength influence coefficient, invasive cost, and patient acceptance score when calculating clinical utility. The coupling strength influence coefficient and patient acceptance score were corrected using personalized adaptation factors, and then combined with invasive costs to calculate the optimized personalized clinical utility value.

[0133] In this embodiment, the target patient's disease type, baseline emotional level, and historical intervention response data are the core information set used to deeply characterize individual patient differences. Disease type refers to the patient's specific clinical diagnosis (e.g., autism, major depressive disorder); baseline emotional level refers to the patient's emotional stability and intensity under normal, uninterrupted conditions; historical intervention response data records the patient's actual reactions and feedback on the effects of different types and intensities of interventions in the past.

[0134] In this embodiment, generating personalized adaptation factors based on the target patient's disease type, baseline emotional level, and historical intervention response data refers to calculating correction coefficients for adjusting general decision-making logic by analyzing the aforementioned individualized information and applying preset rules or machine learning models. For example, the adaptation factor generated for a highly sensitive autistic patient may tend to reduce tolerance for sensory stimulus intensity parameters.

[0135] In this embodiment, the personalized adaptation factor is used to dynamically adjust the relative weights of the coupling strength influence coefficient, invasiveness cost, and patient acceptance score when calculating clinical utility. This means that the factor does not directly modify the values ​​of the three parameters, but rather acts on the weight coefficients of these three parameters in the final decision formula, making them more consistent with the patient's specific needs and response patterns. For example, for depressed patients with good intervention responses, the adaptation factor may increase the effect weight; for patients who are resistant to intervention, it may increase the acceptance weight.

[0136] In this embodiment, personalized adaptation factors are used to correct the coupling strength influence coefficient and patient acceptance score. Then, combined with invasive costs, an optimized personalized clinical utility value is calculated. This means that when actually calculating the utility of candidate strategies, the personalized adaptation factors are first used to numerically adjust (e.g., multiply or weight) the two parameters that are directly related to the individual patient's response: the coupling strength influence coefficient and the patient acceptance score. Then, the adjusted values ​​are substituted into the clinical utility calculation formula along with the invasive costs to finally obtain an optimized utility evaluation value that truly fits the individual characteristics of the patient, which is used to guide the final strategy selection.

[0137] In this embodiment, how to intelligently select the optimal intervention strategy using the above information: Candidate strategy generation: For the identified abnormal coupling pair types, the system will select several different actions or action combinations from the effective action library as candidate strategies.

[0138] Calculate the overall utility value of the strategy: The system calculates a final comprehensive utility value for each candidate strategy. The calculation logic for this value is: to pursue strong effectiveness, high acceptance, and low cost.

[0139] Its core calculation formula can be understood as: Overall utility value = (Effect intensity * Acceptability) / Cost.

[0140] Among them, the effect intensity is the absolute value of the effect coefficient (because the coefficient is negative, the larger the absolute value, the better the effect), the acceptability is the dynamic acceptability score, and the cost is the invasiveness cost score.

[0141] Dynamic adjustments based on urgency: If the model predicts that the patient's emotional coupling is rapidly deteriorating (high urgency), the system will automatically increase the weight of the effect intensity in the calculation formula while decreasing the weight of cost. This means that in critical situations, the system will prioritize the most effective intervention, even if it is slightly more disruptive to the patient.

[0142] Conversely, if the situation is mild (low urgency), the system will increase the weighting of acceptability and low cost, prioritizing intervention methods that are more comfortable and convenient for the patient.

[0143] Introducing personal preferences (personalized adaptation factors): The system also considers the individual characteristics of patients. For example, for patients with autism, the system's rule base has a pre-defined rule: such patients are sensitive to sensory stimulation, therefore, when making decisions, the importance of acceptability should be increased by 20%, and the part of the cost related to sensory stimulation should be considered a higher cost.

[0144] These rules constitute personalized adaptation factors. When calculating the final utility value, the system first uses these factors to adjust the relative importance of effectiveness, acceptability, and cost, and then uses the adjusted weights to calculate and select the best strategy that is truly suitable for the current patient.

[0145] like Figure 2 As shown, this invention provides an embodiment of a clinical patient emotion recognition and intervention system based on multimodal action analysis, comprising: The baseline spectrum construction module is used to collect action and physiological data of the target patient in a baseline emotional state, calculate the baseline coupling strength of at least one set of preset emotional feature pairs, and form a personalized emotional coupling baseline spectrum by combining the emotional state spatial mapping data; wherein the emotional state spatial mapping data includes the baseline state points and the fitted personalized emotional dynamics model. The anomaly identification and prediction module is used to collect the target patient's action data and physiological data in real time, extract the same preset emotional feature pairs and calculate the real-time coupling strength, compare the real-time coupling strength with the benchmark coupling strength, combine the personalized emotional dynamics model to predict the evolution trend, and identify abnormal feature coupling pairs. The strategy matching and execution module is used to match and execute targeted decoupling intervention strategies from a pre-built decoupling strategy mapping table based on the type of the identified abnormal feature coupling pairs; wherein the targeted decoupling intervention strategy is used to reduce the coupling strength of the target abnormal feature coupling pairs. The efficacy assessment module is used to collect the patient's action and physiological data again after the targeted decoupling intervention strategy is executed, calculate the post-intervention coupling strength of the target abnormal feature coupling pair, and evaluate the decoupling efficacy based on the degree of decrease in the post-intervention coupling strength relative to the real-time coupling strength. The intervention result determination and output module is used to determine that the target abnormal feature coupling has been effectively intervened in the corresponding emotional state when the decoupling efficiency exceeds a preset threshold, and outputs an emotion intervention completion signal.

[0146] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A clinical patient emotion recognition and intervention method based on multimodal action analysis, characterized in that, include: Collect action and physiological data of the target patient in a baseline emotional state, calculate the baseline coupling strength of at least one set of preset emotional feature pairs, and combine them with emotional state spatial mapping data to form a personalized emotional coupling baseline spectrum. The emotional state space mapping data includes baseline state points and fitted personalized emotional dynamics models. Real-time acquisition of action and physiological data of target patients, extraction of identical preset emotional feature pairs and calculation of real-time coupling strength, comparison of real-time coupling strength with baseline coupling strength, prediction of evolutionary trends by combining personalized emotion dynamics model, and identification of abnormal feature coupling pairs; Based on the type of the identified anomalous feature coupling pairs, a targeted decoupling intervention strategy is matched from a pre-built decoupling strategy mapping table and executed; wherein the targeted decoupling intervention strategy is used to reduce the coupling strength of the target anomalous feature coupling pairs. After implementing the targeted decoupling intervention strategy, the patient's action data and physiological data were collected again. The post-intervention coupling strength of the target abnormal feature coupling pair was calculated. The decoupling effectiveness was evaluated based on the degree of decrease in the post-intervention coupling strength relative to the real-time coupling strength. When the decoupling efficiency exceeds a preset threshold, it is determined that the target abnormal feature coupling has effectively intervened in the corresponding emotional state, and an emotional intervention completion signal is output.

2. The clinical patient emotion recognition and intervention method based on multimodal action analysis according to claim 1, characterized in that, Preset emotional characteristic pairs, including: The rhythm-respiration coupling pair consists of the characteristics of movement rhythm disorder and the characteristics of respiratory waveform abnormality; Electromyography-skin electrocoagulation pairs are formed by the amplitude characteristics of surface electromyography of specific muscle groups and the level characteristics of skin electrical conductivity response. The visual-cardiac coupling pair consists of the entropy value characteristics of the gaze saccade path and the low-frequency power characteristics of heart rate variability; Gait-blood oxygen coupling pair, consisting of gait stability parameter characteristics and blood oxygen saturation fluctuation characteristics; The posture-endocrine coupling pair consists of trunk posture tilt angle characteristics and cortisol concentration change rate characteristics.

3. The clinical patient emotion recognition intervention method based on multimodal action analysis according to claim 1 or 2, characterized in that, Calculate the real-time coupling strength or baseline coupling strength of preset emotion feature pairs, using methods for quantifying time series synchronicity or causal analysis methods for quantifying information flow. Among them, the methods used to quantify the synchronization of time series include the phase-locked value method or the windowed time-delay mutual information method; Causal analysis methods used to quantify information flow include Granger causality analysis or transfer entropy analysis.

4. The clinical patient emotion recognition and intervention method based on multimodal action analysis according to claim 1, characterized in that, A dynamic threshold method is used to identify anomalous feature coupling pairs, and the judgment logic is corrected by combining a personalized emotion dynamics model. The initial judgment threshold is the product of the benchmark coupling strength and the benchmark standard deviation. The real-time coupling strength and the feature sequence used to calculate the real-time coupling strength are input into the personalized emotion dynamics model to obtain a short-term predicted value of the coupling strength. When the short-term forecast value shows a continuous upward trend, the initial judgment threshold is dynamically lowered.

5. The clinical patient emotion recognition and intervention method based on multimodal action analysis according to claim 1, characterized in that, The decoupling strategy mapping table is pre-built in the following way: Define a basic unit library containing various basic intervention movement units; among which, the basic intervention movement units include breathing guidance units, progressive muscle relaxation guidance units, symbolic gesture imitation units, and slow limb extension dance units; Through controlled experiments, the coupling strength influence coefficient of each basic intervention action unit on each type of preset emotional feature pair was measured, and the invasiveness cost and patient acceptance scores were recorded simultaneously. Based on causal analysis, a set of candidate basic intervention action units is selected for each type of abnormal feature coupling pair; Based on the candidate unit set, generate multiple candidate unit combination strategies; Based on the coupling strength influence coefficient, invasive cost, and patient acceptance score of all basic intervention action units included in each candidate unit combination strategy, calculate the comprehensive coupling strength influence coefficient, comprehensive invasive cost, and comprehensive patient acceptance score of the corresponding candidate unit combination strategy. From all candidate unit combination strategies, the basic intervention action unit combination with the negative and largest absolute value of coupling strength influence coefficient, the lowest invasive cost, and the highest patient acceptance score is selected to form a targeted decoupling intervention strategy; The clinical utility value of the targeted decoupling intervention strategy is labeled; where the clinical utility value is equal to the absolute value of the impact coefficient multiplied by the patient acceptance score and then divided by the invasive cost. Based on the type of each abnormal feature coupling pair, the corresponding targeted decoupling intervention strategy, and the corresponding clinical utility value, a mapping relationship is established and stored to form a decoupling strategy mapping table.

6. The clinical patient emotion recognition and intervention method based on multimodal action analysis according to claim 5, characterized in that, Through controlled experiments, the coupling strength influence coefficient of each basic intervention action unit on each pre-defined emotional feature pair was measured, including: The subjects were stratified according to their clinical characteristics, and an intervention group that performed targeted basic intervention action units and a control group that performed simulated actions were set up for each stratum; During the experiment, real-time coupling strength data of preset emotion feature pairs were collected simultaneously; Based on the intervention group data, the static difference between the mean coupling strength after intervention and the baseline coupling strength before intervention was calculated, and the coupling strength decay rate was obtained by curve fitting of the real-time coupling strength data. By weighting and fusing the static difference with the coupling strength decay rate, the coupling strength influence coefficient of the corresponding basic intervention action unit on the corresponding preset emotional feature pair is obtained.

7. The clinical patient emotion recognition and intervention method based on multimodal action analysis according to claim 5, characterized in that, The process of determining the overall patient acceptance score includes: Define multidimensional attribute features for each basic intervention action unit; the multidimensional attribute features include unit execution duration, sensory stimulation intensity, required level of interpersonal interaction, and historical execution records; Based on the real-time physiological and motor data of the target patients, predict the immediate acceptance tendency of the target patients to intervention actions with different multidimensional attributes. By fusing historical static patient acceptance scores with immediate acceptance propensity, dynamic acceptance scores for corresponding basic intervention action units are generated. The overall patient acceptance score of the corresponding candidate unit combination strategy is obtained by weighted averaging the dynamic acceptance scores of all basic intervention action units included in each candidate unit combination strategy.

8. The clinical patient emotion recognition and intervention method based on multimodal action analysis according to claim 5, characterized in that, A targeted decoupling intervention strategy is constructed by selecting basic intervention action units with the largest negative absolute value of coupling strength influence coefficient, the lowest invasive cost, and the highest patient acceptance score, including: The urgency of intervention is determined based on the prediction of future trends in the coupling strength of current abnormal features using a personalized emotion dynamics model. Based on the urgency of the intervention, the weight coefficients of coupling strength influence coefficient, invasive cost, and patient acceptance score in the clinical utility decision function are dynamically configured. When the urgency level is high, increase the weight of the coupling strength influence coefficient and decrease the weight of invasive cost; when the urgency level is low, perform reverse configuration. Multiple candidate unit combination strategies are generated based on the combination of all basic intervention action units; The decision function with dynamically configured weights is used to calculate the utility value of the combined strategies of each candidate unit, and the final targeted decoupling intervention strategy is selected based on the utility value.

9. The clinical patient emotion recognition and intervention method based on multimodal action analysis according to claim 5, characterized in that, When labeling the clinical utility values ​​of targeted decoupling intervention strategies, personalized adaptation factors based on patient characteristics are introduced for optimization calculations, including: Personalized adaptation factors are generated based on the target patient's disease type, baseline emotional level, and historical intervention response data. These personalized adaptation factors are used to dynamically adjust the relative weights of the coupling strength influence coefficient, invasive cost, and patient acceptance score when calculating clinical utility. The coupling strength influence coefficient and patient acceptance score were corrected using personalized adaptation factors, and then combined with invasive costs to calculate the optimized personalized clinical utility value.

10. A clinical patient emotion recognition and intervention system based on multimodal action analysis, characterized in that, include: The baseline spectrum construction module is used to collect action and physiological data of the target patient in a baseline emotional state, calculate the baseline coupling strength of at least one set of preset emotional feature pairs, and form a personalized emotional coupling baseline spectrum by combining the emotional state spatial mapping data. The emotional state space mapping data includes baseline state points and fitted personalized emotional dynamics models. The anomaly identification and prediction module is used to collect the target patient's action data and physiological data in real time, extract the same preset emotional feature pairs and calculate the real-time coupling strength, compare the real-time coupling strength with the benchmark coupling strength, combine the personalized emotional dynamics model to predict the evolution trend, and identify abnormal feature coupling pairs. The strategy matching and execution module is used to match and execute targeted decoupling intervention strategies from a pre-built decoupling strategy mapping table based on the type of the identified abnormal feature coupling pairs; wherein the targeted decoupling intervention strategy is used to reduce the coupling strength of the target abnormal feature coupling pairs. The efficacy assessment module is used to collect the patient's action and physiological data again after the targeted decoupling intervention strategy is executed, calculate the post-intervention coupling strength of the target abnormal feature coupling pair, and evaluate the decoupling efficacy based on the degree of decrease in the post-intervention coupling strength relative to the real-time coupling strength. The intervention result determination and output module is used to determine that the target abnormal feature coupling has been effectively intervened in the corresponding emotional state when the decoupling efficiency exceeds a preset threshold, and outputs an emotion intervention completion signal.