Drug addict relapse risk monitoring and early warning method based on heart and brain behavior intelligent fusion

Abstract

The invention discloses a relapse risk monitoring and early warning method based on heart and brain behavior intelligent fusion, and belongs to the technical field of drug banning and health monitoring. Three-mode data of HRV sample entropy (electrocardio), prefrontal lobe theta wave coherence (electroencephalogram) and motion fractal dimension (behavior) of a patient abstained from drugs are collected in a non-stimulation mode through a wearable device; motion interference is suppressed by adopting a motion-driven cascade filtering technology (kurtosis detection and self-adaptive band elimination filtering); inputting the data into a space-time diagram convolution-federated learning hybrid model (ST-FedGCN), and outputting a relapse risk level (1-5 levels) and expected relapse time; when the risk probability value is Rgt; and when the risk level is 0.82 or the risk level is greater than or equal to 4, the portable tDCS equipment is automatically triggered, and 1.5 mA regulation and control current is applied to the front frontal lobe on the back and outer side of the right side. Laboratory dependence is broken through, community dynamic monitoring is achieved, the prediction accuracy is larger than 90%, the relapse rate is reduced, and powerful support is provided for drug banning work.

Description

Technical Field

[0001] This invention belongs to the field of drug control and health monitoring technology, specifically involving a method for monitoring and early warning of relapse risk in drug addicts based on the integration of mind, brain, and behavioral intelligence, which is mainly used in drug rehabilitation centers, community rehabilitation centers, and home environments. Background Technology

[0002] Monitoring and early warning of relapse risk among drug addicts can not only help them consolidate their rehabilitation achievements (such as alleviating psychological cravings through cognitive training), but also optimize the allocation of social resources (focusing on the management of high-risk groups), and at the same time, it has important preventive value for maintaining social order and reducing drug crimes.

[0003] However, the relapse rate remains high, and current methods for monitoring relapse among drug addicts all have significant shortcomings:

[0004] Asynchronous signal: Existing technologies (such as CN116070120B) only support bimodal alignment and cannot solve the real-time synchronization of ECG-EEG-behavioral trimodalities.

[0005] Stimulus dependence (laboratory dependence): Methods such as those in patent CN118072949A require bringing drug addicts back to the drug rehabilitation center and monitoring them by presenting drug-related pictures or videos to induce their physiological and psychological reactions (visual stimulus-induced response). This method cannot be applied to dynamic monitoring in daily community life scenarios.

[0006] Insufficient mobility adaptability: The EEG+NIRS device of Shanghai Health Medical College CN114224297A relies on fixed laboratory deployment and has a motion artifact suppression rate of only 65%; Wangli Technology's VR assessment system requires induced environment support and cannot achieve daily mobile monitoring.

[0007] Real-time performance defects of early warning models: The LightGBM model of Hubei University of Technology (CN110826852A) has a response delay of >5 minutes (tested with 100,000 data points).

[0008] Missing EEG indicators: Methods such as patent CN110623650A only monitor physiological indicators such as electrocardiogram / cortical conductance, ignoring the core mechanism of addiction "wanting to take drugs" - abnormal brain nerve function.

[0009] Defects of open-loop systems: Methods such as US20220192625A1 only achieve the monitoring function and lack the "early warning-intervention" closed-loop mechanism. Although they can detect problems and issue warnings, they lack automatic and immediate intervention measures, making it easy to miss the best intervention time and unable to stop relapse behavior in time when the risk occurs (only early warning without immediate intervention).

[0010] Insufficient prediction accuracy: assessments based on traditional scales are highly subjective and prone to large errors; intelligent models based on a single modality (such as ECG or EEG only) have limited accuracy, typically less than 80%.

[0011] Therefore, there is an urgent need for a relapse risk monitoring and early warning method that can be effectively monitored without special stimulation during the normal life of drug addicts (in the community or at home) and can accurately predict relapse risk by integrating cardiac, brain, and behavioral signals. Summary of the Invention

[0012] In view of this, the purpose of this invention is to provide a method for monitoring and early warning of relapse risk in drug addicts based on the integration of mind, brain, and behavioral intelligence, in order to solve the problems of existing methods, such as limited applicability, long response delay, lack of intervention measures, and low prediction accuracy.

[0013] This invention provides a method for monitoring and early warning of relapse risk in drug addicts based on the integration of mind-brain behavioral intelligence, comprising the following steps:

[0014] S1. Non-stimulatory data collection: Using wearable devices to simultaneously collect electrocardiogram (ECG) data, electroencephalogram (EEG) data, and daily behavior data of drug addicts in a natural living state;

[0015] S2. Dynamic noise suppression: Based on behavioral motion data-driven cascaded filtering, motion artifact suppression is performed on EEG and ECG data;

[0016] S3. Input the noise-reduced EEG data, ECG data, and daily behavior data into the spatiotemporal graph convolution-federated learning hybrid model (ST-FedGCN), and output the relapse risk level of 1-5, the expected relapse time, and the relapse risk probability value R.

[0017] S4. Early warning and intervention;

[0018] When the relapse risk probability value R>0.82 or the relapse risk level ≥4, the portable tDCS device is automatically triggered to apply a 1.5mA regulating current to the right dorsolateral prefrontal cortex.

[0019] Furthermore, the wearable device includes an ECG patch, a dry electrode EEG patch, and an inertial measurement unit;

[0020] ECG patch: Used to measure the HRV sample entropy of drug addicts, i.e., ECG data;

[0021] Dry electrode EEG patch: used to collect the theta wave coherence of the prefrontal cortex of drug addicts, i.e., EEG data;

[0022] Inertial measurement unit: used to calculate the motion fractal dimension of drug addicts, i.e., daily behavior data.

[0023] Furthermore, in step S1, the coherence of the prefrontal theta wave is calculated using the FP1-FP2 ultra-simplified lead scheme;

[0024] The calculation expression is as follows:

[0025] FC ij =|corr(θ) i ,θ j )|

[0026] In the formula, FC ij Represents the coherence matrix of FP1-FP2 channels; θ i This represents the signal measured on the FP1 channel; θ j This indicates the signal measured on the FP2 channel.

[0027] Furthermore, in step S1, the motion fractal dimension is calculated using the "box counting method";

[0028] The calculation expression for the "box counting method" is as follows:

[0029]

[0030] In the formula, D b denoted by fractal dimension; ò represents the coverage scale of the data points; N(ò) represents the minimum number of boxes that can completely cover the data points at the given scale.

[0031] Furthermore, step S2 includes the following sub-steps:

[0032] S2.1 Generate dynamic filtering parameters f based on the daily behavior data collected in step S1. c ;

[0033]

[0034] In the formula, f c Indicates the cutoff frequency; a rms It represents the root mean square acceleration, used to reflect the vibration amplitude of the wearer;

[0035] S2.2 uses kurtosis detection to locate EEG / ECG signal segments severely contaminated by motion artifacts;

[0036] S2.3 employs an adaptive band-stop filter to selectively attenuate interference in the corresponding frequency band based on the type and intensity of actions in daily behavior data.

[0037] Furthermore, the spatiotemporal graph convolution-federated learning hybrid model in step S3 is constructed by fusing the spatiotemporal graph convolution module and the federated learning framework;

[0038] The spatiotemporal graph convolution module is used to map ECG data, EEG data, and daily behavior data into a spatiotemporal graph structure, and extract node features and spatiotemporal dependencies; the federated learning framework is used to protect data privacy.

[0039] Furthermore, in step S3, the criteria for classifying relapse risk levels 1-5 are as follows:

[0040] Level 1, a safe state, corresponds to the following three signal characteristic ranges: "sample entropy > 1.2, coherence > 0.5, fractal dimension < 1.0";

[0041] Level 2, representing the state of interest, corresponds to the following three signal characteristic ranges: "0.9 < sample entropy ≤ 1.2, 0.3 < coherence ≤ 0.5";

[0042] Level 3 represents moderate risk, corresponding to the following three signal characteristic ranges: "sample entropy < 0.9, fractal dimension > 1.2";

[0043] Level 4 is considered high risk, corresponding to the three signal characteristic ranges: "sample entropy < 0.8, coherence < 0.3";

[0044] Level 5 represents an emergency risk, with the corresponding three-signal characteristic range: "sample entropy < 0.7, coherence < 0.2, fractal dimension > 1.8".

[0045] Furthermore, the formula for calculating the expected relapse time is as follows:

[0046]

[0047] In the formula, T represents the expected relapse time; τd represents the neural adaptation delay time; r1 represents the relapse triggering event rate; and r2 represents the effectiveness of the intervention.

[0048] Furthermore, the formula for calculating the relapse risk probability value R is as follows:

[0049]

[0050] In the formula, R(t) represents the predicted relapse risk, i.e., the probability value of relapse risk; β0 represents the intercept term; β1, β2, and β3 are the weights obtained from training the ST-FedGCN model, corresponding to the coefficients of each input feature; HRV SampBn Sample entropy representing heart rate variability, i.e., autonomic neural complexity; EEG Coh This indicates the coherence of the electroencephalogram (EEG), i.e., the prefrontal cortex's control ability; D b Indicates the level of anxiety in behavior.

[0051] This invention also provides a relapse risk monitoring and early warning system for drug addicts based on the integration of mind-brain behavioral intelligence, comprising:

[0052] Wearable EEG acquisition module: used to collect EEG data from the wearer;

[0053] Wearable ECG acquisition module: used to collect the wearer's ECG data;

[0054] Multi-axis inertial measurement unit: used to collect data on the wearer's daily behavior;

[0055] Edge processing unit: Built-in cascaded filtering algorithm and ST-FedGCN model, used to obtain the wearer's relapse risk level, relapse probability value R, and expected relapse time in real time;

[0056] Portable tDCS devices: used for interventions on wearers;

[0057] Communication module: Used to transmit early warning information in real time and trigger tDCS control.

[0058] Beneficial effects:

[0059] 1. **Convenient "Seamless Monitoring":** The world's first effective monitoring solution for natural living conditions in communities and homes, eliminating "laboratory dependence," with lightweight and comfortable equipment. 2. **Accurate "Three-in-One":** Integrating heart, brain, and movement signal analysis, it is more accurate than single-dimensional analysis, and the intelligent model excels at uncovering complex relationships. 3. **Reliable "Motion Noise Reduction":** "Motion Navigation Noise Reduction" technology (kurtosis + adaptive filtering) effectively eliminates interference from daily activities, resulting in clearer signals. 4. **Privacy-Friendly and "Smarter with Use":** Federated learning technology enables model sharing and improvement without leaking personal data, ensuring security and compliance. 5. **Realistic and Effective:** The ST-FedGCN model has a prediction accuracy of 92.3%, outperforming existing technologies, and can provide accurate early warnings up to 2.8 days in advance. 6. **Real-Time Performance:** Edge computing reduces response latency to <150ms.

[0060] In summary, this invention, through multimodal data fusion, advanced artificial intelligence analysis, and an innovative closed-loop intervention system, achieves more accurate and timely prediction of the risk of relapse in drug addicts, significantly improves intervention effectiveness and management efficiency, and breaks through many limitations of existing technologies, demonstrating significant clinical value and technological innovation.

[0061] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0062] Figure 1 This is a diagram illustrating the architecture of the method for monitoring and early warning of relapse risk in drug addicts according to the present invention.

[0063] Figure 2 This is a schematic diagram of the early warning triggering logic of the method for monitoring and early warning of relapse risk in drug addicts according to the present invention;

[0064] Figure 3 The flowchart for the ST-FedGCN model;

[0065] Figure 4 This is a flowchart for encrypted data transmission in the ST-FedGCN model. Detailed Implementation

[0066] To make the technical solutions, advantages, and objectives of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the protection scope of this application.

[0067] like Figure 1 and Figure 2 As shown, this invention provides a method for monitoring and early warning of relapse risk in drug addicts based on the fusion of mind-brain behavioral intelligence, comprising the following steps:

[0068] S1. Non-stimulatory data collection: Using wearable devices to simultaneously collect electrocardiogram (ECG) data, electroencephalogram (EEG) data, and daily behavior data of drug addicts in a natural living state;

[0069] Wearable devices include ECG patches, dry electrode EEG patches, and inertial measurement units;

[0070] ECG patch: Used to measure the HRV sample entropy (focus: autonomic nervous state) of drug addicts, i.e., ECG data;

[0071] Dry electrode EEG patch: used to collect the theta wave coherence (focus: neural synchrony) of the prefrontal cortex of drug addicts, i.e., EEG data;

[0072] Inertial measurement unit: used to calculate the motion fractal dimension of drug addicts (focus: behavioral complexity, referring to the linear correlation between fractal dimension and muscle activity), i.e. daily behavioral data;

[0073] The three key signal sources are shown in Table 1.

[0074] Table 1

[0075]

[0076]

[0077] The three-modal indices are shown in Table 2.

[0078] Table 2

[0079]

[0080] In this embodiment, data acquisition is non-stimulating: data is collected using textile electrode ECG clothing, flexible dry electrode EEG patches, and IMU sensors, ensuring comfortable and lightweight wear that does not interfere with daily life.

[0081] Among them, HRV sample entropy (RR interval nonlinear complexity)

[0082] Algorithm principle:

[0083]

[0084] In the formula, H sample Let A represent the HRV sample entropy; A is the number of sequences matching m+1 points; B is the number of sequences matching m points (m=2, r=0.2σ);

[0085] Anomaly detection: Autonomic neuron disorder is marked when the sample entropy is <0.8.

[0086] The coherence of the theta wave in the prefrontal cortex was calculated using the FP1-FP2 ultra-simplified lead scheme, and the calculation expression is as follows:

[0087] FC ij =|corr(θ) i ,θ j )|

[0088] In the formula, FC ij Represents the coherence matrix of FP1-FP2 channels; θ i This represents the signal measured on the FP1 channel; θ j This indicates the signal measured on the FP2 channel.

[0089] The motion fractal dimension was calculated using the "box counting method" (100Hz sampling); the calculation expression is as follows:

[0090]

[0091] In the formula, D b denoted by fractal dimension; ò represents the coverage scale of the data points; N(ò) represents the minimum number of boxes that can completely cover the data points at the given scale.

[0092] S2. Dynamic noise suppression: Based on behavioral motion data-driven cascaded filtering, motion artifact suppression is performed on EEG and ECG data;

[0093] By using an inertial measurement unit (IMU sensor) to accurately acquire the wearer's movement patterns, and employing cascaded filtering technology, the signal segments severely interfered with by large movements are first quickly located through kurtosis detection. Then, based on the movement type and intensity information provided by the inertial measurement unit, adaptive band-stop filtering is used to dynamically adjust the filter parameters, accurately filtering the interference frequencies generated by specific movements. The dynamic noise suppression technology of this invention can achieve an artifact suppression rate of 92%, effectively eliminating unexpected interference information in the signal, which is superior to existing technologies.

[0094] The adaptive band-stop filter formula is as follows:

[0095]

[0096] In the formula, f c Indicates the cutoff frequency; a rms It represents the root mean square acceleration, used to reflect the vibration amplitude of the wearer.

[0097] S3.ST-FedGCN Model Analysis: The noise-reduced EEG data, ECG data, and daily behavior data are input into the spatiotemporal graph convolution-federated learning hybrid model (ST-FedGCN), which outputs a relapse risk level of 1-5, the expected relapse time, and the relapse risk probability value R.

[0098] Spatiotemporal Graph Convolution (ST-GCN): Constructs a spatiotemporal relational graph of heart, brain, and behavioral data (nodes: physiological indicators; edges: modal interactions). For example... Figure 3 and Figure 4 As shown, the spatiotemporal graph convolution-federated learning hybrid model (ST-FedGCN) is used. The spatiotemporal graph convolution part connects the three points of "heart", "brain" and "behavior" into a triangle to analyze the change patterns within each point and the mutual influence between points. It is good at capturing complex correlation patterns in time and space. The federated learning part enables drug rehabilitation centers / community centers in various places to train part of the model locally with their own data. Only the encrypted model update information is uploaded to the cloud. The cloud aggregates and forms a global model before it is distributed. This protects privacy and allows the model to share progress. Finally, it outputs the risk level (1-5, with level 5 being the highest risk), the expected relapse time (number of days) and the relapse risk probability value R.

[0099] Federated learning: Local drug rehabilitation centers train models locally, upload encrypted parameters to the cloud for aggregation (privacy protection, reducing data transmission by 80%), or directly upload to the cloud for data encryption processing by a remote cloud data center.

[0100] The performance of the ST-FedGCN model is shown in Table 3.

[0101] Table 3

[0102]

[0103] The criteria for classifying relapse risk levels 1-5 are shown in Table 4:

[0104] Table 4

[0105]

[0106] The formula or method for calculating the expected relapse time is as follows:

[0107]

[0108] In the formula, T represents the expected relapse time; τd represents the neural adaptation delay time; r1 represents the relapse triggering event rate; and r2 represents the effectiveness of the intervention. τd is measured by fMRI to detect the dopamine receptor recovery rate (default 7 days); the value range of τd is 7.0 ± 0.5 days. r1 is measured by wearable devices to count the average daily number of high-risk exposures (e.g., location of drug-related events + sudden heart rate changes), or GPS location to count the number of times the average daily stay in high-risk areas is >10 minutes (Gaode Map API fence function) combined with other vital signs and weight changes. The value range of r1 is 0.1-5.0 times / day. r2 is measured by: frequency of psychological counseling × effectiveness coefficient (e.g., twice a week × 0.7) + 0.3 × tDCS inhibition rate (41.3%). The value range of r2 is 0.05-0.8.

[0109] The formula or method for calculating the relapse risk probability value R is as follows:

[0110]

[0111] In the formula, R(t) represents the predicted relapse risk, i.e., the probability value of relapse risk; β0 represents the intercept term; β1, β2, and β3 are the weights obtained from training the ST-FedGCN model, corresponding to the coefficients of each input feature; HRV SampBn Sample entropy representing heart rate variability, i.e., autonomic neural complexity; EEG Coh This indicates the coherence of the electroencephalogram (EEG), i.e., the prefrontal cortex's control ability; D b This indicates behavioral anxiety level. HRV is one such indicator. SampBn Corresponding measurement equipment and algorithms: ECG patch / ECG garment sampling → Richman algorithm calculation (m=2, r=0.2σ); High-risk threshold: <0.8; Weighting basis: β1=-1.2±0.1. EEG Coh Corresponding measurement equipment and algorithm: Head loop leads (FP1-FP2) → 4-8Hz bandpass filtering → phase synchronization calculation; High-risk threshold: <0.3; Weighting basis: β2=-0.9±0.05. D bThe corresponding measurement equipment and algorithm are: six-axis sensor → OpenPose skeleton tracking → box counting fractal dimension calculation (ε = 0.5 mm); high-risk threshold: >1.5; weighting basis: β3 = 0.8 ± 0.05. The corresponding measurement equipment and algorithm for β0 are: federated learning global calibration value; high-risk threshold: 0.35 ± 0.1.

[0112] S4. Early warning and intervention;

[0113] Early warning response: The recipient (such as staff or organization) responds to the early warning in a timely manner;

[0114] Intervention Implementation: Initiate a personalized intervention plan, including:

[0115] Online neuromodulation: When the relapse risk probability value R>0.82 or the relapse risk level ≥4, the portable tDCS (transcranial direct current stimulation) device is automatically triggered to apply a 1.5mA modulating current to a specific area of ​​the brain (right dorsolateral prefrontal cortex) for 20 minutes.

[0116] In-person comprehensive intervention: combining psychological counseling, medication, community support, rehabilitation training, etc.

[0117] Effectiveness evaluation: Evaluate the effectiveness of the intervention in order to adjust the strategy.

[0118] Clinical basis for threshold 0.82: This threshold is set to meet the need for high-specificity early warning; when R>0.82, the model judges the probability of relapse risk to be over 82%; this threshold ensures a false positive rate of <5%.

[0119] The necessity of dual threshold logic design

[0120] Probability threshold (R>0.82): For sudden risks (such as acute cravings triggered by environmental factors), even if the current level is not at level 4, intervention is still required if the probability exceeds the threshold.

[0121] Level threshold (≥4 levels): Covers progressive risk accumulation (such as chronic neurological function decline) and avoids missed judgments by probability models due to data noise.

[0122] This invention also provides a relapse risk monitoring and early warning system for drug addicts based on the integration of mind-brain behavioral intelligence, comprising:

[0123] Wearable EEG acquisition module: used to collect EEG data from the wearer;

[0124] Wearable ECG acquisition module: used to collect the wearer's ECG data;

[0125] Multi-axis inertial measurement unit: used to collect data on the wearer's daily behavior;

[0126] Edge processing unit: Built-in cascaded filtering algorithm and ST-FedGCN model, used to obtain the wearer's relapse risk level, relapse probability value R, and expected relapse time in real time;

[0127] Portable tDCS devices: used for interventions on wearers;

[0128] Communication module: Used to transmit early warning information in real time and trigger tDCS control. Data is transmitted to the data analysis platform through wireless encryption (such as 4G / 5G, IoT) to ensure security.

[0129] Application scenarios

[0130] Taking the monitoring of a drug addict in a community rehabilitation center as an example, the addict wears an EEG patch (lightweight headband), an ECG vest, and an IMU (Induction Mutation Unit). The device begins collecting brainwaves, heartbeat signals, and the trajectory and movement details of daily activities from the addict's frontal region. "Motion navigation noise reduction" technology is used to process the collected data in real time to ensure signal clarity and accuracy. The processed clean data is transmitted to the cloud platform in real time for analysis by the ST-FedGCN model, which outputs the addict's current relapse risk level and expected relapse time. Assuming that after a period of monitoring, the model determines that the addict's risk level has reached level 4, the system immediately and automatically triggers an early warning mechanism, sending alerts to the addict (device vibration alert), family members, and community rehabilitation center management personnel (APP or platform alarm).

[0131] This invention discloses a method and system for monitoring and warning of relapse risk in drug addicts based on the integration of mind-brain behavioral intelligence. Starting with device wearing, it achieves accurate prediction and timely warning of relapse risk through real-time data acquisition, secure transmission, precise data processing, and deep learning analysis. Based on this, it initiates automated neuromodulation intervention and comprehensive offline intervention with human involvement, forming a closed-loop management process. This method not only improves the continuity and accuracy of monitoring but also effectively reduces the risk of relapse through timely intervention, making it applicable to various scenarios such as drug rehabilitation centers and communities.

[0132] It is hereby declared that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for monitoring and early warning of relapse risk in drug addicts based on the integration of mind-brain behavioral intelligence, characterized in that, Includes the following steps: S1. Non-stimulatory data collection: Using wearable devices to simultaneously collect electrocardiogram (ECG) data, electroencephalogram (EEG) data, and daily behavior data of drug addicts in a natural living state; S2. Dynamic noise suppression: Based on behavioral motion data-driven cascaded filtering, motion artifact suppression is performed on EEG and ECG data; S3. Input the noise-reduced EEG data, ECG data, and daily behavior data into the spatiotemporal graph convolution-federated learning hybrid model (ST-FedGCN), and output the relapse risk level of 1-5, the expected relapse time, and the relapse risk probability value R. S4. Early warning and intervention; When the relapse risk probability value R>0.82 or the relapse risk level ≥4, the portable tDCS device is automatically triggered to apply a 1.5mA regulating current to the right dorsolateral prefrontal cortex.

2. The method for monitoring and early warning of relapse risk in drug addicts based on the fusion of mind-brain behavioral intelligence according to claim 1, characterized in that, The wearable device includes an ECG patch, a dry electrode EEG patch, and an inertial measurement unit; ECG patch: Used to measure the HRV sample entropy of drug addicts, i.e., ECG data; Dry electrode EEG patch: used to collect the theta wave coherence of the prefrontal cortex of drug addicts, i.e., EEG data; Inertial measurement unit: used to calculate the motion fractal dimension of drug addicts, i.e., daily behavior data.

3. The method for monitoring and early warning of relapse risk in drug addicts based on the integration of mind-brain behavioral intelligence according to claim 2, characterized in that, In step S1, the coherence of the theta wave in the prefrontal cortex is calculated using the FP1-FP2 supersimplified lead scheme. The calculation expression is as follows: FC ij =|corr(θ i ,i j )| In the formula, FC ij Represents the coherence matrix of FP1-FP2 channels; θ i This represents the signal measured on the FP1 channel; θ j This indicates the signal measured on the FP2 channel.

4. The method for monitoring and early warning of relapse risk in drug addicts based on the integration of mind-brain behavioral intelligence according to claim 3, characterized in that, In step S1, the motion fractal dimension is calculated using the "box counting method"; The calculation expression for "box counting method" is as follows: In the formula, D b Represents the fractal dimension; Indicates the coverage scale of data points; This represents the minimum number of boxes that can completely cover the data points at a given scale.

5. A method for monitoring and early warning of relapse risk in drug addicts based on the fusion of mind-brain behavioral intelligence as described in claim 4, characterized in that, Step S2 includes the following sub-steps: S2.1 Generate dynamic filtering parameters f based on the daily behavior data collected in step S1. c ; In the formula, f c Indicates the cutoff frequency; a rms It represents the root mean square acceleration, used to reflect the vibration amplitude of the wearer; S2.2 uses kurtosis detection to locate EEG / ECG signal segments severely contaminated by motion artifacts; S2.3 employs an adaptive band-stop filter to selectively attenuate interference in the corresponding frequency band based on the type and intensity of actions in daily behavior data.

6. The method for monitoring and early warning of relapse risk in drug addicts based on the fusion of mind-brain behavioral intelligence according to claim 5, characterized in that, The spatiotemporal graph convolution-federated learning hybrid model in step S3 is constructed by fusing spatiotemporal graph convolution modules and federated learning frameworks. The spatiotemporal graph convolution module is used to map ECG data, EEG data, and daily behavior data into a spatiotemporal graph structure, and extract node features and spatiotemporal dependencies; the federated learning framework is used to protect data privacy.

7. A method for monitoring and early warning of relapse risk in drug addicts based on the fusion of mind-brain behavioral intelligence as described in claim 6, characterized in that, In step S3, the criteria for classifying relapse risk levels 1-5 are as follows: Level 1, a safe state, corresponds to the following three signal characteristic ranges: "sample entropy > 1.2, coherence > 0.5, fractal dimension < 1.0"; Level 2, representing the state of interest, corresponds to the following three signal characteristic ranges: "0.9 < sample entropy ≤ 1.2, 0.3 < coherence ≤ 0.5"; Level 3, representing moderate risk, corresponds to the following three-signal characteristic ranges: "sample entropy < 0.9, fractal dimension > 1.2"; Level 4 is considered high risk, corresponding to the following three signal characteristic ranges: "sample entropy < 0.8, coherence < 0.3"; Level 5 represents an emergency risk, with the corresponding three-signal characteristic ranges as follows: "sample entropy < 0.7, coherence < 0.2, fractal dimension > 1.8".

8. A method for monitoring and early warning of relapse risk in drug addicts based on the fusion of mind-brain behavioral intelligence according to claim 7, characterized in that, The formula for calculating the expected relapse time is as follows: In the formula, T represents the expected relapse time; τd represents the neural adaptation delay time; r1 represents the relapse triggering event rate; and r2 represents the effectiveness of the intervention.

9. A method for monitoring and early warning of relapse risk in drug addicts based on the fusion of mind-brain behavioral intelligence as described in claim 7, characterized in that, The formula for calculating the relapse risk probability value is as follows: In the formula, R(t) represents the predicted relapse risk, i.e., the probability value of relapse risk; β0 represents the intercept term; β1, β2, and β3 are the weights obtained from training the ST-FedGCN model, corresponding to the coefficients of each input feature; HRV SampBn Sample entropy representing heart rate variability, i.e., autonomic neural complexity; EEG Coh This indicates the coherence of the electroencephalogram (EEG), i.e., the prefrontal cortex's control ability; D b Indicates the level of anxiety in behavior.

10. A relapse risk monitoring and early warning system for drug addicts based on the fusion of mind-brain and behavioral intelligence, used to implement the relapse risk monitoring and early warning method for drug addicts based on the fusion of mind-brain and behavioral intelligence as described in any one of claims 1-9, characterized in that, include: Wearable EEG acquisition module: used to collect EEG data from the wearer; Wearable ECG acquisition module: used to collect the wearer's ECG data; Multi-axis inertial measurement unit: used to collect data on the wearer's daily behavior; Edge processing unit: Built-in cascaded filtering algorithm and ST-FedGCN model, used to obtain the wearer's relapse risk level, relapse probability value R, and expected relapse time in real time; Portable tDCS devices: used for interventions on wearers; Communication module: Used to transmit early warning information in real time and trigger tDCS control.

Images