Non-suicide self-injury identification and prediction method and system

Abstract

The invention relates to a non-suicide self-injury identification and prediction method. The method comprises the following steps: carrying out behavior data acquisition and electroencephalogram signal acquisition on a subject; the collected electroencephalogram signals are preprocessed; the preprocessed electroencephalogram signals are analyzed; according to the collected behavior data and the electroencephalogram signals obtained through analysis, a classification model is constructed, and non-suicide self-injury is recognized and predicted. The invention also relates to a non-suicide self-injury identification and prediction system. According to the invention, an objective and accurate detection means can be provided, early recognition and illness state monitoring of the NSSI are realized, and favorable conditions are created for timely intervention and treatment.

Description

Technical Field

[0001] The present invention relates to a method and system for identifying and predicting nonsuicidal self-injury. Background Art

[0002] Nonsuicidal self-injury (NSSI) refers to the repeated behavior of an individual intentionally harming their own tissues and organs without the intention of suicide. It mostly occurs in the adolescent group aged 10 to 19, with an incidence rate as high as 20% to 30%, posing a serious threat to the physical and mental health of adolescents. Effective identification and prediction can promote early diagnosis and treatment, improve the quality of life of patients, and avoid adverse consequences. Therefore, objective and accurate detection means are very necessary. Scalp electroencephalogram has the characteristics of being cheap, convenient, and non-invasive, and has good operability in children, adolescents, and clinical patients. Existing research shows that NSSI is closely related to impulsive cognitive control. The Go / No-Go task is a classic experimental paradigm commonly used to study response inhibition, impulse control, and attention. In the task, the subject must respond or inhibit the response according to the presented stimulus type. Previous studies have found that patients with depression accompanied by suicidal ideation or suicidal behavior show a decrease in the P3 component of the electroencephalogram in this task. However, the existing technology has not yet solved the core problem of how to construct an identification model for NSSI using the electroencephalogram activity characteristics related to the Go / No-Go task.

[0003] The existing diagnosis of nonsuicidal self-injury behavior (NSSI) mainly relies on subjective clinical interviews and patient self-report, lacking objective and quantifiable biomarkers, resulting in the diagnostic results being easily affected by individual expression biases, making it difficult to accurately evaluate the severity of symptoms, leading to inaccurate and untimely diagnoses. The existing methods based on skin conductance detection and video behavior analysis in the prior art, although they can partially reflect autonomic arousal or external action characteristics, cannot locate the brain function abnormalities directly related to the pathological mechanism of NSSI (such as prefrontal inhibitory control defects and limbic system emotion regulation abnormalities), and cannot effectively identify and monitor the condition of NSSI patients, making it difficult to meet the clinical needs. Summary of the Invention

[0004] In view of this, it is necessary to provide a method and system for identifying and predicting nonsuicidal self-injury, which can provide objective and accurate detection means, make up for the defect of the existing technology lacking biomarkers, realize the early identification and condition monitoring of NSSI, and create favorable conditions for timely intervention and treatment.

[0005] The present invention provides a method for identifying and predicting non-suicidal self-injury, and the method includes the following steps: a. Collecting behavioral data and electroencephalogram (EEG) signals of the subject; b. Preprocessing the collected EEG signals; c. Analyzing the preprocessed EEG signals; d. Constructing a classification model based on the collected behavioral data and the analyzed EEG signals, and identifying and predicting non-suicidal self-injury.

[0006] Preferably, the step a includes: The behavioral data collection includes: Recording the correct hit rate, false alarm rate, and reaction time of correct responses of the subject, and evaluating the attention control and inhibitory bias of the subject to emotional stimuli; The EEG signal collection includes: Collecting the task-state EEG signals of the subject during the completion of the emotional task and 10 minutes of resting-state EEG signals.

[0007] Preferably, the step b includes: For all EEG signals: Using offline analysis to perform filtering in turn to remove high-frequency noise and baseline drift; Identifying bad leads through visual inspection, and using the data of surrounding leads for interpolation replacement, and finally re-referencing the data to the whole-brain average reference; For the task-state EEG signals: Centering on the stimulus event, segmenting into 2.5-second time periods, performing independent component analysis, identifying and removing eye movement, blink, and myoelectric artifact components; After segmentation, performing correction and removing abnormal segments; For the resting-state EEG signals: After removing the first and last 20 seconds of data, segmenting the filtered and re-referenced continuous signals into 4-second time periods, performing independent component analysis, identifying and removing eye movement, blink, and myoelectric artifact components, and then subtracting the mean value of the whole segment of data to eliminate the DC offset, and also removing the time periods with amplitude exceeding the limit to ensure that the data quality meets the requirements of subsequent analysis.

[0008] Preferably, the step c includes: The EEG signal analysis includes: Event-related potential analysis, time-frequency analysis, and resting-state EEG microstate analysis; The event-related potential analysis includes: Averaging the segmented EEG data under each condition after preprocessing to obtain the event-related potential waveform; The time-frequency analysis includes: Subtracting the event-related potential component from the single-trial EEG signal to obtain the induced power, calculating the spectral power in the 1-40 Hz frequency band using a sliding Hanning window, and normalizing it with the mean power of the baseline period; Subsequently, using a cluster permutation test to evaluate the power differences between non-suicidal self-injury patients and the control group in the delta, theta, alpha, and beta frequency bands; Finally, extracting the average power of the theta frequency band within the 300-600 ms time window after stimulation of the central midline electrode as the core index; The resting-state EEG microstate analysis includes: band-pass filtering the resting-state EEG signals, identifying the global field power peaks and constructing individual microstate maps; after matching the microstate maps to the standardized templates, back-fitting them to the individual data to extract the parameters for quantifying the dynamic characteristics of the resting-state brain state.

[0009] Preferably, step d described above includes: Constructing a classification model based on the behavioral data and EEG signal characteristics; By means of the forward selection strategy, gradually optimizing the classification model in descending order of feature importance, evaluating the performance using three-fold cross-validation, and finally selecting the classification model with the highest accuracy; Evaluating the discrimination efficacy of the classification model for non-suicidal self-injury patients and healthy control groups, responders and non-responders.

[0010] The present invention provides a non-suicidal self-injury identification and prediction system, which includes an acquisition module, a preprocessing module, an analysis module, and an identification and prediction module. Among them: the acquisition module is used for collecting behavioral data and EEG signals of the subject; the preprocessing module is used for preprocessing the collected EEG signals; the analysis module is used for analyzing the preprocessed EEG signals; the identification and prediction module is used for constructing a classification model based on the collected behavioral data and the analyzed EEG signals, and identifying and predicting non-suicidal self-injury.

[0011] Preferably, the acquisition module specifically is used for: The behavioral data acquisition includes: recording the correct hit rate, false alarm rate and reaction time of correct responses of the subject, and evaluating the attention control and inhibition bias of the subject to emotional stimuli; The EEG signal acquisition includes: acquiring the task-state EEG signals of the subject during the completion of the emotional task and 10 minutes of resting-state EEG signals.

[0012] Preferably, the preprocessing module specifically is used for: For all EEG signals: using offline analysis to perform filtering in turn to remove high-frequency noise and baseline drift; identifying bad leads through visual inspection, and using the data of surrounding leads for interpolation replacement, and finally re-referencing the data to the whole-brain average reference; For task-state EEG signals: centering on the stimulus event, segmenting into 2.5-second time periods, performing independent component analysis, identifying and removing eye movement, blink and electromyogram artifact components; performing correction after segmentation, and removing abnormal segments; For resting-state EEG signals: After removing the first and last 20 seconds of data, the filtered and rereferenced continuous signals are segmented into 4-second epochs, independent component analysis is performed to identify and remove eye movement, blink, and electromyogram artifact components. Subsequently, the entire data segment is subtracted by the mean to eliminate the DC offset, and epochs with amplitude exceeding the limit are also removed to ensure that the data quality meets the requirements of subsequent analysis.

[0013] Preferably, the analysis module is specifically configured to: EEG signal analysis includes: event-related potential analysis, time-frequency analysis, and resting-state EEG microstate analysis; The event-related potential analysis includes: averaging the preprocessed EEG segmented data under each condition to obtain the event-related potential waveform; The time-frequency analysis includes: subtracting the event-related potential component from the single-trial EEG signal to obtain the induced power, calculating the spectral power in the 1-40 Hz frequency band using a sliding Hanning window, and normalizing it with the mean power of the baseline period; Subsequently, a cluster permutation test is used to evaluate the power differences between non-suicidal self-injurers and the control group in the delta, theta, alpha, and beta frequency bands; Finally, the average power of the theta frequency band within the 300-600 ms time window after stimulation at the central midline electrode is extracted as the core index; The resting-state EEG microstate analysis includes: band-pass filtering the resting-state EEG signal, identifying the global field power peak and constructing an individual microstate map; After matching the microstate map to a standardized template, it is back-fitted to the individual data to extract parameters for quantifying the dynamic characteristics of the resting-state brain state.

[0014] Preferably, the recognition and prediction module is specifically configured to: Construct a classification model based on behavioral data and EEG signal features; Through a forward selection strategy, the classification model is gradually optimized in descending order of feature importance, and a three-fold cross-validation is used to evaluate the performance. Finally, the classification model with the highest accuracy is selected; Evaluate the discrimination efficacy of the classification model for non-suicidal self-injurers and healthy controls, responders and non-responders.

[0015] This application constructs an objective detection model based on neural activity characteristics by fusing multi-modal EEG features (task-related P3 amplitude, theta wave power, and resting-state microstate dynamic parameters), realizing the precise early identification of NSSI and the quantitative evaluation of dynamic disease condition monitoring, providing a traceable and verifiable diagnosis and treatment basis for clinical practice. More specifically, the beneficial effects of this application include: First, precise neural markers: Based on electroencephalogram (EEG) during the Go / No-Go task to capture specific neural features of NSSI patients (such as reduced P3 amplitude and attenuated theta wave power), directly locate prefrontal cognitive control defects, and break through the limitations of subjective reports, electrodermal responses, and video detection that cannot reveal the pathological mechanism.

[0016] Second, dynamic diagnosis and prediction: Combine millisecond-level EEG analysis to real-time track the neural dynamics of impulse inhibition, and integrate resting-state microstate features to predict treatment response, realizing an integrated assessment of "detection - intervention - prognosis".

[0017] Third, efficient clinical application: The device is portable and non-invasive, suitable for the adolescent population; the algorithm is lightweight, and its computational efficiency is significantly better than video multi-model fusion, suitable for clinical real-time screening and long-term monitoring. Brief Description of the Drawings

[0018] Figure 1 It is a flowchart of the non-suicidal self-injury recognition and prediction method of the present invention; Figure 2 It is a schematic diagram of constructing a classification model according to behavioral data and EEG signal features provided by an embodiment of the present invention; Figure 3 It is a hardware architecture diagram of the non-suicidal self-injury recognition and prediction system of the present invention. Detailed Embodiments

[0019] The present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0020] Refer to Figure 1 As shown, it is an operation flowchart of a preferred embodiment of the non-suicidal self-injury recognition and prediction method of the present invention.

[0021] Step S1, collect behavioral data and EEG signals from the subject. Specifically: Behavioral data collection: In this embodiment, a visual presentation system is used: mainly for presenting visual stimuli for testing, including: a computer equipped with a monitor, and an emotional Go / No-go task presented on the computer screen.

[0022] Grayscale face images are used as stimulus materials, including three expression types: positive (happy), negative (sad), and neutral. The size, brightness, contrast, and temporal frequency are uniformly adjusted to ensure stimulus standardization. Each face stimulus is presented for 800 milliseconds, followed by a fixation point (lasting 800 - 1200 milliseconds).

[0023] Subjects are required to make button presses in response to specific types of face stimuli (Go trials), while not responding to another type of stimuli (NoGo trials), and to ensure the speed and accuracy of responses as much as possible. The task consists of 4 conditions, with 75% being Go trials and 25% being NoGo trials. The stimulus combinations for each group are as follows: Happy Go / Neutral NoGo, Neutral Go / Happy NoGo, Sad Go / Neutral NoGo, Neutral Go / Sad NoGo. The test order for different conditions is random and balanced among subjects. The correct hit rate, false alarm rate, and reaction time of correct responses of the subjects are recorded to evaluate the attentional control and inhibitory bias of the subjects towards emotional stimuli.

[0024] EEG signal acquisition: The EEG data in this embodiment is acquired through a 32-channel EEG system (sampling rate 500 Hertz (Hz)). The electrode array is arranged based on the extended international 10-20 system. The online reference electrode is the central midline (Cz), and the ground electrode is the frontal midline (Fz). During the recording process, the electrode impedance is kept below 10 kiloohms (kΩ). The task-state EEG signals of the subjects during the completion of the emotional Go / No-go task and 10 minutes of resting-state EEG signals are acquired.

[0025] Step S2, preprocess the acquired EEG signals. Specifically: For all EEG signals: Use offline analysis to perform filtering in sequence to remove high-frequency noise and baseline drift. Identify bad leads through visual inspection, and use the data of surrounding leads for interpolation and replacement. Finally, re-reference the data to the whole-brain average reference.

[0026] For task-state EEG signals: Centered on the stimulus event, segment into 2.5-second time periods (1000 ms before the stimulus to 1500 ms after the stimulus), and perform independent component analysis (ICA) using the runica algorithm to identify and remove eye movement, blink, and electromyogram artifact components. After segmentation, correct with a 200-ms baseline before the stimulus, and remove abnormal segments with amplitudes exceeding ±100 µv.

[0027] For resting-state EEG signals: After removing the first and last 20 seconds of data, segment the filtered and re-referenced continuous signal into 4-second time periods. The ICA artifact removal process is the same as that of the task state. Subsequently, subtract the mean value from the entire segment of data to eliminate the DC offset, and also remove the time periods with amplitude overrun to ensure that the data quality meets the requirements of subsequent analysis.

[0028] Step S3, analyze the preprocessed EEG signals. Specifically: EEG signal analysis includes three parts: event-related potential (ERP) analysis, time-frequency analysis, and resting-state EEG microstate analysis.

[0029] For ERP analysis: First, the segmented EEG data under each condition after preprocessing were averaged to obtain the ERP waveform, and the P3 component at the Pz electrode point within the time window of 400 to 700 ms after the stimulus was focused on; the P3 component has been used in previous studies to evaluate cognitive control and impulsivity in clinical populations and is sensitive to emotional contexts.

[0030] For time-frequency analysis: The evoked power was obtained by subtracting the ERP component from the single-trial EEG signal. The spectral power in the frequency band of 1 - 40 Hz (1 Hz resolution) was calculated using a sliding Hanning window (frequency-dependent time window, 3 - 18 cycles per frequency), and was normalized by the mean power in the baseline period (-200 to 0 ms). Subsequently, a cluster permutation test was used to evaluate the power differences between NSSI patients and the control group in the delta (1 - 3 Hz), theta (4 - 7 Hz), alpha (8 - 12 Hz), and beta (13 - 30 Hz) frequency bands. Finally, the average power in the theta frequency band within the time window of 300 - 600 ms after the stimulus at the central midline electrodes (Cz and Fz) was extracted as the core index.

[0031] Resting-state EEG microstate analysis: The resting-state EEG signal was band-pass filtered at 2 - 20 Hz, and the peaks of the global field power (GFP) were identified and an individual microstate map was constructed: First, the individual data of the subjects were clustered at the first level by the k-means algorithm, and then the subjects in the same group were clustered at the second level (2 - 8 classes) based on the cross-validation criterion to distinguish four types of EEG microstate maps; after the microstate map was matched to the standardized template, it was back-fitted to the individual data, and parameters such as the average duration, occurrence frequency, time coverage rate, and average GFP of each type of microstate were extracted to quantify the dynamic characteristics of the resting-state brain state.

[0032] Step S4, according to the collected behavioral data and the analyzed EEG signals, a classification model was constructed to identify and predict non-suicidal self-injury. Specifically: In this embodiment, to achieve the identification and prediction of the intervention effect of non-suicidal self-injury (NSSI), a random forest classification model was constructed based on behavioral data (such as correct rejection rate, sensitivity) and EEG signal characteristics (task-state P3 amplitude, theta wave power, and resting-state EEG microstate parameters).

[0033] Using the NSSI occurrence frequencies at 1, 3, and 6 months after intervention as the core prediction indicators, patients were divided into a response group (no NSSI within 1 month) and a non-response group, and electroencephalogram signal features significantly related to the follow-up indicators were screened. Through a forward selection strategy, the random forest classification model was gradually optimized in descending order of feature importance. Three-fold cross-validation was used to evaluate the performance: the data was divided into a training set and a test set at a ratio of 2:1, and the cycle was verified until each fold participated in the test. Finally, the random forest classification model with the highest accuracy was selected. At the same time, the area under the curve (AUC) was calculated through the receiver operating characteristic curve (ROC) to evaluate the discrimination efficacy of the random forest classification model for NSSI patients and healthy controls, as well as for the response group and the non-response group.

[0034] It should be noted that the classification model is not limited to the random forest classification model and can be replaced by a support vector machine (SVM) classification model, a convolutional neural network (CNN) classification model, or other machine learning algorithm classification models.

[0035] Refer to Figure 3 As shown, it is the hardware architecture diagram of the non-suicidal self-injury recognition and prediction system 10 of the present invention. The system includes: an acquisition module 101, a preprocessing module 102, an analysis module 103, and an identification and prediction module 104.

[0036] The acquisition module 101 is used to collect behavioral data and electroencephalogram signals of the subject. Specifically: The acquisition module 101 collects behavioral data: In this embodiment, a visual presentation system is adopted: mainly used to present visual stimuli for testing, including: a computer equipped with a monitor, and an emotion Go / No-go task presented on the computer screen.

[0037] Grayscale face images are used as stimulus materials, including three expression types: positive (happy), negative (sad), and neutral. The size, brightness, contrast, and temporal frequency are uniformly adjusted to ensure stimulus standardization. Each face stimulus is presented for 800 milliseconds, followed by a fixation point (lasting 800 - 1200 milliseconds).

[0038] The subject needs to make a key response (Go trial) to a specific type of face stimulus and not respond to another type of stimulus (NoGo trial), and ensure the rapidity and accuracy of the response as much as possible. The task includes 4 conditions, where 75% are Go trials and 25% are NoGo trials. The stimulus combinations for each group are as follows: happy Go / neutral NoGo, neutral Go / happy NoGo, sad Go / neutral NoGo, neutral Go / sad NoGo. The test order of different conditions is random and balanced among subjects. The correct hit rate, false alarm rate, and reaction time of correct responses of the subject are recorded to evaluate the subject's attentional control and inhibitory bias for emotional stimuli.

[0039] The acquisition module 101 performs electroencephalogram (EEG) signal acquisition: In this embodiment, EEG data is acquired through a 32-channel EEG system (sampling rate: 500 Hertz (Hz)). The electrode array is arranged based on the extended international 10-20 system. The online reference electrode is the central midline (Cz), and the ground electrode is the frontal midline (Fz). During the recording process, the electrode impedance is maintained below 10 kiloohms (kΩ). Task-state EEG signals during the subject's completion of the emotional Go / No-go task and 10 minutes of resting-state EEG signals are acquired.

[0040] The preprocessing module 102 is used to preprocess the acquired EEG signals. Specifically: For all EEG signals: The preprocessing module 102 uses offline analysis to perform filtering in sequence to remove high-frequency noise and baseline drift. Bad leads are identified through visual inspection, and interpolation replacement is performed using the data of surrounding leads. Finally, the data is re-referenced to the whole-brain average reference.

[0041] For task-state EEG signals: The preprocessing module 102 centers around the stimulation event, segments them into 2.5-second time periods (1000 ms before the stimulation to 1500 ms after the stimulation), and performs independent component analysis (ICA) using the runica algorithm to identify and remove eye movement, blink, and electromyogram artifact components. After segmentation, baseline correction is performed with 200 ms before the stimulation as the baseline, and abnormal segments with amplitudes exceeding ±100 µv are removed.

[0042] For resting-state EEG signals: After removing the first and last 20 seconds of data, the preprocessing module 102 segments the filtered and re-referenced continuous signals into 4-second time periods. The ICA artifact removal process is the same as that for task-state signals. Subsequently, the entire segment of data is subtracted by the mean to eliminate the DC offset, and time periods with amplitude exceeding the limit are also removed to ensure that the data quality meets the requirements of subsequent analysis.

[0043] The analysis module 103 is used to analyze the preprocessed EEG signals. Specifically: EEG signal analysis includes three parts: event-related potential (ERP) analysis, time-frequency analysis, and resting-state EEG microstate analysis.

[0044] The analysis module 103 performs ERP analysis: First, the segmented EEG data under each condition after preprocessing is averaged to obtain the ERP waveform, and the P3 component at the Pz electrode point within the time window of 400 to 700 ms after the stimulation is focused on. The P3 component has been used in previous studies to evaluate cognitive control and impulsivity in clinical populations and is sensitive to emotional contexts.

[0045] The analysis module 103 performs time-frequency analysis: subtracting the ERP component from the single-trial EEG signal to obtain the induced power, calculating the spectral power in the 1-40 Hz frequency band (1 Hz resolution) using a sliding Hanning window (frequency-dependent time window, 3-18 cycles per frequency), and normalizing it with the power mean during the baseline period (-200 to 0 ms). Subsequently, a cluster permutation test is used to evaluate the power differences between NSSI patients and the control group in the delta (1-3 Hz), theta (4-7 Hz), alpha (8-12 Hz), and beta (13-30 Hz) frequency bands. Finally, the average power in the theta frequency band within the 300-600 ms time window after stimulation at the central midline electrodes (Cz and Fz) is extracted as the core index.

[0046] The analysis module 103 performs resting-state EEG microstate analysis: performing 2-20 Hz band-pass filtering on the resting-state EEG signal, identifying the global field power (GFP) peaks and constructing an individual microstate map: first, performing primary clustering on the individual data of the subject through the k-means algorithm, and then performing secondary clustering (2-8 classes) on the subjects in the same group based on the cross-validation criterion to distinguish four types of EEG microstate maps; after matching the microstate map to the standardized template, back-fitting it to the individual data, and extracting parameters such as the average duration, occurrence frequency, time coverage rate, and average GFP of each type of microstate to quantify the dynamic characteristics of the resting-state brain state.

[0047] The recognition and prediction module 104 is used to construct a classification model based on the collected behavioral data and the analyzed EEG signals, and identify and predict non-suicidal self-injury. Specifically: In this embodiment, to achieve the recognition and intervention effect prediction of non-suicidal self-injury (NSSI), the recognition and prediction module 104 constructs a random forest classification model based on behavioral data (such as correct rejection rate, sensitivity) and EEG signal characteristics (task-state P3 amplitude, theta wave power, and resting-state EEG microstate parameters).

[0048] The recognition and prediction module 104 uses the NSSI occurrence frequency at 1, 3, and 6 months after intervention as the core prediction index, divides the patients into a response group (no NSSI within 1 month) and a non-response group, screens the EEG signal characteristics significantly related to the follow-up index, and gradually optimizes the random forest classification model in descending order of feature importance through the forward selection strategy. The performance is evaluated using three-fold cross-validation: dividing the data into a training set and a test set at a ratio of 2:1, and cyclically validating until each fold participates in the test. Finally, the random forest classification model with the highest accuracy is selected. At the same time, the area under the curve (AUC) is calculated through the receiver operating characteristic curve (ROC) to evaluate the discrimination efficacy of the random forest classification model between NSSI patients and healthy controls, and between the response group and the non-response group.

[0049] It should be noted that the classification model is not limited to the random forest classification model and can be replaced by a support vector machine (SVM) classification model, a convolutional neural network (CNN) classification model, or other machine learning algorithm classification models.

[0050] Experimental verification: Experiment 1: In the Go / No-Go paradigm, positive or negative emotional stimuli were paired with neutral stimuli, where the proportion of Go trials was 75%, and the Go / No-Go roles of the emotional stimulus type (positive / negative) and the neutral stimulus were dynamically switched during the test. Refer to Table 1. The data of Experiment 1 further verified that the electroencephalogram P3 component and theta wave power attenuation during the task could effectively reveal the differences in neuroelectrophysiological characteristics of NSSI patients in cognitive control and emotion regulation, and accurately detect NSSI patients.

[0051] Table 1: Effective detection of NSSI patients by electroencephalogram P3 component and theta wave power attenuation

[0052] Experiment 2: Combining task-related electroencephalogram and resting-state microstate dynamic analysis, a NSSI-specific biomarker was constructed through P3 amplitude attenuation, reduction of central midline theta wave power, and change in microstate D coverage rate, breaking through the limitations of single-modal detection. Please refer to Figure 2 , where: Figure 2 (A) and (B) in are schematic diagrams for distinguishing NSSI patients from healthy control subjects, Figure 2 (C) and (D) in are schematic diagrams for predicting self-harm behavior of NSSI patients at 1-month follow-up. The data of Experiment 2 showed that: using Go condition-theta energy, Go condition-P3 amplitude, No-go condition theta energy, and the correct rejection rate of the task could effectively identify NSSI patients (AUC = 0.8452); using the occurrence frequency of resting-state electroencephalogram microstate C, the time coverage rate of state D, and the occurrence frequency of state D could effectively predict whether patients still had self-harm behavior at 1-month follow-up (AUC = 0.8362).

[0053] Although the present invention has been described with reference to the current preferred embodiments, those skilled in the art should understand that the above preferred embodiments are only used to illustrate the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle scope of the present invention shall be included within the scope of the present invention's rights protection.

Claims

1. A method for identifying and predicting non - suicidal self - injury, characterized in that, The method includes the following steps: a. Collect behavioral data and electroencephalogram (EEG) signals from the subjects; b. Preprocess the collected EEG signals; c. Analyze the preprocessed EEG signals; d. Based on the collected behavioral data and the analyzed EEG signals, construct a classification model, and identify and predict non-suicidal self-injury.

2. The method according to claim 1, characterized in that, The step a includes: The behavioral data collection includes: recording the correct hit rate, false alarm rate, and reaction time of correct responses of the subjects, and evaluating the attention control and inhibitory bias of the subjects towards emotional stimuli; The EEG signal collection includes: collecting the task-state EEG signals of the subjects during the completion of the emotional task and 10 minutes of resting-state EEG signals.

3. The method according to claim 2, wherein The step b includes: For all EEG signals: use offline analysis to perform filtering in sequence to remove high-frequency noise and baseline drift; identify bad leads through visual inspection, and use the data of surrounding leads for interpolation and replacement, and finally re-reference the data to the whole-brain average reference; For task-state EEG signals: centered on the stimulus event, segment into 2.5-second time periods, perform independent component analysis, identify and remove eye movement, blink, and electromyogram artifact components; perform correction after segmentation, and remove abnormal segments; For resting-state EEG signals: after removing the first and last 20 seconds of data, segment the filtered and re-referenced continuous signals into 4-second time periods, perform independent component analysis, identify and remove eye movement, blink, and electromyogram artifact components, and then subtract the mean value from the whole segment of data to eliminate the DC offset, and also remove the time periods with amplitude exceeding the limit to ensure that the data quality meets the requirements of subsequent analysis.

4. The method according to claim 3, wherein The step c includes: EEG signal analysis includes: event-related potential analysis, time-frequency analysis, and resting-state EEG microstate analysis; The event-related potential analysis includes: averaging the segmented EEG data under each condition after preprocessing to obtain the event-related potential waveform; The time-frequency analysis includes: subtracting the event-related potential component from the single-trial EEG signal to obtain the induced power, calculating the spectral power in the 1-40 Hz frequency band using a sliding Hanning window, and normalizing it with the mean power of the baseline period; subsequently, use the cluster permutation test to evaluate the power differences between non-suicidal self-injury patients and the control group in the delta, theta, alpha, and beta frequency bands; finally, extract the average power of the theta frequency band within the 300-600 ms time window after stimulation at the central midline electrode as the core index; The resting-state EEG microstate analysis includes: band-pass filtering the resting-state EEG signals, identifying the global field power peak and constructing an individual microstate map; after matching the microstate map to the standardized template, back-fitting it to the individual data, and extracting the parameters for quantifying the dynamic characteristics of the resting-state brain state.

5. The method according to claim 4, wherein The step d includes: Construct a classification model based on behavioral data and EEG signal features; Through the forward selection strategy, gradually optimize the classification model in descending order of feature importance, use three-fold cross-validation to evaluate the performance, and finally select the classification model with the highest accuracy; Evaluate the discrimination efficacy of the classification model for non-suicidal self-injury patients and healthy control groups, responders and non-responders.

6. A non-suicidal self-harm identification and prediction system, characterized in that, The system includes a collection module, a preprocessing module, an analysis module, and an identification and prediction module, where: The collection module is used to collect behavioral data and electroencephalogram (EEG) signals of the subject; The preprocessing module is used to preprocess the collected EEG signals; The analysis module is used to analyze the preprocessed EEG signals; The identification and prediction module is used to construct a classification model based on the collected behavioral data and the analyzed EEG signals, and identify and predict non-suicidal self-injury.

7. The system according to claim 6, characterized in that, The specific functions of the collection module are as follows: The collection of behavioral data includes: recording the correct hit rate, false alarm rate, and reaction time of correct responses of the subject, and evaluating the attention control and inhibitory bias of the subject to emotional stimuli; The collection of EEG signals includes: collecting the task-state EEG signals of the subject during the completion of the emotional task and 10 minutes of resting-state EEG signals.

8. The system according to claim 7, wherein The specific functions of the preprocessing module are as follows: For all EEG signals: perform filtering offline in sequence to remove high-frequency noise and baseline drift; identify bad leads through visual inspection, and use the data of surrounding leads for interpolation and replacement, and finally re-reference the data to the whole-brain average reference; For task-state EEG signals: centered on the stimulus event, segment into 2.5-second time periods, perform independent component analysis, identify and remove eye movement, blink, and electromyogram artifact components; perform correction after segmentation, and remove abnormal segments; For resting-state EEG signals: after removing the first and last 20 seconds of data, segment the filtered and re-referenced continuous signal into 4-second time periods, perform independent component analysis, identify and remove eye movement, blink, and electromyogram artifact components, and then subtract the mean value of the whole segment of data to eliminate the DC offset, and also remove the time periods with amplitude exceeding the limit to ensure that the data quality meets the requirements of subsequent analysis.

9. The system according to claim 8, wherein The specific functions of the analysis module are as follows: EEG signal analysis includes: event-related potential analysis, time-frequency analysis, and resting-state EEG microstate analysis; The event-related potential analysis includes: averaging the segmented EEG data under each condition after preprocessing to obtain the event-related potential waveform; The time-frequency analysis includes: subtracting the event-related potential component from the single-trial EEG signal to obtain the induced power, calculating the spectral power in the 1-40 Hz frequency band using a sliding Hanning window, and normalizing it with the mean power of the baseline period; subsequently, using a cluster permutation test to evaluate the power differences between non-suicidal self-injury patients and the control group in the delta, theta, alpha, and beta frequency bands; finally, extracting the average power of the theta frequency band within the 300-600 ms time window after stimulation of the central midline electrode as the core index; The resting-state EEG microstate analysis includes: band-pass filtering the resting-state EEG signal, identifying the global field power peak and constructing an individual microstate map; after matching the microstate map to a standardized template, inverse-fitting it to the individual data, and extracting the parameters for quantifying the dynamic characteristics of the resting-state brain state.

10. The system according to claim 9, wherein, The specific functions of the identification and prediction module are as follows: Construct a classification model based on behavioral data and EEG signal characteristics; By using the forward selection strategy, the classification model is gradually optimized step by step in descending order of feature importance. Three-fold cross-validation is used to evaluate the performance, and finally the classification model with the highest accuracy is selected; Evaluate the discrimination efficacy of the classification model between patients with non-suicidal self-injury and healthy controls, and between responders and non-responders.

Images