Multi-task behavior-electroencephalogram synchronous monitoring method for schizophrenia screening
By designing a multi-task temporal loading paradigm and a multimodal fusion discrimination model, and simultaneously collecting EEG and behavioral data, the problem of insufficient accuracy in schizophrenia screening in existing technologies has been solved, achieving efficient and accurate screening for schizophrenia.
Patent Information
- Application Number
- CN202610362927.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-24
- Publication Date
- 2026-06-26
AI Technical Summary
Current technologies rely on physician consultations and scale assessments, lacking clinically applicable molecular tests or biomarkers, and fail to effectively capture the dynamic changes in the neurodynamic characteristics of the brain during long-term cognitive tasks, resulting in insufficient accuracy in schizophrenia screening.
A multi-task temporal loading paradigm was designed to simultaneously collect EEG signals and behavioral data of subjects during the execution of multiple tasks, extract multi-dimensional features, construct a multimodal fusion discrimination model, and output a schizophrenia risk index.
By fusing and dynamically correcting multidimensional features, the accuracy and robustness of schizophrenia screening have been improved, enabling efficient and precise assessment of multiple cognitive dimensions.
Smart Images
Figure CN122272021A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of schizophrenia screening, specifically to a multi-task behavioral-EEG synchronous monitoring method for schizophrenia screening. Background Technology
[0002] Schizophrenia is a severe, chronic, and complex mental disorder characterized by core symptoms such as thought disorder, perceptual abnormalities, emotional blunting, and loss of motivation. Its pathogenesis is closely related to neurophysiological dysfunctions in multiple cognitive dimensions of the brain, including sensory gating, semantic processing, and reward feedback, severely impacting patients' social functioning and quality of life. Early and accurate screening is crucial for timely intervention and improved prognosis in schizophrenia. Neuroelectrophysiological techniques, with their non-invasive, real-time, and quantifiable reflection of brain activity, have become an important research direction for objective screening of schizophrenia.
[0003] Electroencephalography (EEG) and event-related potentials (ERPs), as classic neurophysiological detection methods, can accurately capture changes in brain electrical activity during perception, cognition, and emotional processing. Specific components such as P50, MMN, N400, and FRN are highly correlated with sensory gating deficits, auditory processing abnormalities, semantic network disorder, and reward processing impairment in schizophrenia patients, providing important biological markers for objective screening of the disease. Simultaneously, subjects' behavioral performance in cognitive tasks (such as reaction time, accuracy, and false alarm rate) directly reflects their cognitive function status, including attention maintenance, thought control, and motivational stability, complementing EEG signals and further improving the objectivity and accuracy of screening. Against this backdrop, multidimensional assessment methods based on simultaneous behavioral and EEG monitoring have become an important development trend in auxiliary screening technologies for schizophrenia.
[0004] Currently, the diagnosis of this disease mainly relies on clinical symptomatology, depending on doctor's consultation and scale assessment. There is a lack of molecular tests or biomarkers that can be used clinically, which leads to subjectivity and heterogeneity in the diagnostic process. Secondly, the feature extraction of existing technologies mostly focuses on static indicators of EEG signals (such as the amplitude and latency of a single component), without considering the dynamic changes of the brain's neurodynamic characteristics during long-term cognitive tasks. However, in schizophrenic patients, the complexity of EEG signals exhibits specific dynamic evolution patterns (such as rapid decay or continuous disorder) during continuous tasks. This time-varying feature in the time dimension is an important objective marker of the disease. The lack of such features in existing technologies leads to insufficient screening features, further reducing the accuracy of screening. Summary of the Invention
[0005] To address the aforementioned technical problems, a multi-task behavioral-EEG synchronous monitoring method for schizophrenia screening is provided. This technical solution solves the problems mentioned in the background technology, which rely on doctor consultation and scale assessment and lack molecular detection or biomarkers that can be used clinically. Furthermore, the feature extraction of existing technologies mainly focuses on static indicators of EEG signals and does not consider the dynamic changes of brain neurodynamic characteristics of subjects during long-term cognitive tasks.
[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A multi-task behavioral-EEG synchronous monitoring method for schizophrenia screening includes: Based on clinical screening needs, a multi-task temporal loading paradigm is designed, which includes a sensory gating test task, a semantic and auditory oddball hybrid test task, and a reward feedback test task. Simultaneously collect EEG signals and behavioral data of subjects during the execution of a multi-task temporal loading paradigm, and preprocess the collected raw data; Multidimensional features are extracted from the preprocessed data, including EEG feature groups, behavioral feature groups, and time-varying feature groups. A multimodal fusion discriminant model is constructed, and multidimensional features are input into the multimodal fusion discriminant model to output the basic risk probability and the schizophrenia risk index.
[0007] Preferably, the design of the multi-task temporal loading paradigm based on clinical screening needs specifically includes: The first stage is a sensory gating test task, in which paired click stimuli are played to the subjects. Each pair of clicks consists of a first click and a second click with a first preset time interval. The interval between the pairs of clicks is a second preset time. At the same time, the subjects are asked to silently count the total number of pairs of clicks played. The second phase is a mixed semantic and auditory oddball test task, in which a series of word stimuli are played continuously to the subjects, and the subjects are required to press a key when they hear a specific category of words; Among them, the standard stimulus is a frequently occurring common word, and the deviation stimulus includes two categories: physical deviation stimulus and semantic deviation stimulus. The physical deviation stimulus is a high-pitched word that changes the speech frequency, and the semantic deviation stimulus is an abnormal word that is unrelated to the context. The third stage is a reward feedback test task, in which a simplified game interface is presented to the subjects, and three results are given through visual feedback: win, lose, and draw. At the same time, the reaction time of the subjects from the presentation of the stimulus to pressing the button to place a bet is recorded.
[0008] Preferably, the simultaneous acquisition of EEG signals and behavioral data of the subject during the execution of the multi-task temporal loading paradigm, and the preprocessing of the acquired raw data specifically include: A high-density EEG acquisition device was used to record the EEG signals of the subjects during the performance of multiple tasks, and the subjects' key press reaction time, key press accuracy, missed key presses and false alarms were recorded simultaneously. The acquired raw EEG signals are preprocessed, including: removing electrooculography and electromyography artifacts using independent component analysis, removing bad channels and repairing them by neighborhood interpolation, filtering with a bandpass filter, and eliminating signal drift using a baseline correction method. Based on the stimulus presentation timestamp and key press timestamp, the preprocessed EEG signals are precisely aligned with the behavioral event tags to form a synchronized dataset.
[0009] Preferably, the extraction of EEG feature sets from the preprocessed data specifically includes: Calculate the P50 suppression ratio based on paired click sound task data; Based on semantic and auditory oddball task data, the peak value of the differential wave induced by the deviation stimulus at the Fz electrode was selected as the MMN amplitude, and the time from the stimulus onset to the peak value was selected as the MMN latency. Based on semantic and auditory oddball task data, calculate the average N400 amplitude of the central parietal electrode within the [a,b] time window after stimulation; Based on the reward feedback task data, calculate the average amplitude of the feedback-related negative wave difference wave induced by output feedback and win feedback within the [c,d] time window after feedback; Based on data from the entire mission period, the average phase-locking values of the prefrontal and temporal electrodes in the [e,f] frequency band were calculated.
[0010] Preferably, the extraction of behavioral feature groups from the preprocessed data specifically includes: Based on semantic and auditory oddball task data, the number of incorrect key presses made by the subjects on non-target stimuli was counted, and the total number of non-target stimuli was also counted. The false alarm rate is calculated based on the number of incorrect key presses made by the subject on non-target stimuli and the total number of non-target stimuli. Based on the reward feedback task data, the reaction time of the subjects in all valid key press attempts was statistically analyzed, and the standard deviation and mean of the reaction time were calculated. Calculate the coefficient of variation for reaction time based on the standard deviation and mean of the reaction time; Based on data from the entire task period, the number of missed responses by subjects on all target stimuli was counted, and the total number of target stimuli was also counted. The overall task dropout rate is calculated based on the number of missed responses by the subjects to all target stimuli and the total number of target stimuli.
[0011] Preferably, the extraction of time-varying feature groups from the preprocessed data specifically includes: Divide the entire task time period into several consecutive windows of equal length; Calculate the sample entropy value of the EEG signal within each window; Linear regression fitting was performed on the midpoint of the window time and the corresponding sample entropy value to obtain the regression slope, which was then used as the EEG signal complexity attenuation coefficient.
[0012] Preferably, the construction of the multimodal fusion discriminant model, which involves inputting multidimensional features into the model and outputting the basic risk probability and schizophrenia risk index, specifically includes: A multimodal fusion discriminant model is constructed, which includes a first-layer feature-level fusion classifier and a second-layer decision-level fusion regulator. First-level feature-level fusion classifier: Combines EEG feature groups and behavioral feature groups to form a feature vector, inputs it into a support vector machine classifier, and outputs the basic risk probability; The second-level decision-level fusion regulator introduces the attenuation coefficient in the time-varying feature group as a regulation factor to dynamically correct the basic risk probability and output the schizophrenia risk index. Based on the final risk index and the degree of abnormality of each characteristic indicator, an abnormality dimension radar chart is generated. The radar chart includes four dimensions: sensory gating, semantic processing, reward motivation, and brain network connectivity.
[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention provides a multi-task behavioral-EEG synchronous monitoring method for schizophrenia screening. It simultaneously assesses multiple cognitive dimensions, such as sensory gating, semantic processing, and reward feedback, through a multi-task temporal loading paradigm, overcoming the limitation of single-task methods that can only assess a single dimension. It introduces the entropy decay coefficient of EEG signal samples as a time-varying feature, quantifying the dynamic neural changes of subjects during the task and compensating for the shortcomings of traditional methods that only focus on static features. It deeply integrates behavioral indicators such as false alarm rate and reaction time coefficient of variation as core diagnostic features with EEG characteristics, fully leveraging the value of behavioral data in reflecting clinical manifestations such as impulsivity and motivational stability. Furthermore, it employs a multimodal fusion discriminant model, using time-varying features to dynamically correct the baseline risk probability, effectively solving the problem of test result distortion caused by subject state fluctuations, and improving the accuracy, robustness, and clinical applicability of schizophrenia screening. Attached Figure Description
[0014] Figure 1This is a flowchart of a multi-task behavioral-EEG synchronous monitoring method for schizophrenia screening according to the present invention; Figure 2 This is a flowchart of the present invention for extracting EEG feature groups from preprocessed data; Figure 3 This is a flowchart of the present invention for extracting behavioral feature groups from preprocessed data; Figure 4 This is a flowchart of the present invention for extracting time-varying feature groups from preprocessed data; Figure 5 This is a structural diagram of the electronic device proposed in this invention; Figure 6 This is a schematic diagram of the structure of the computer-readable storage medium proposed in this invention. Detailed Implementation
[0015] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.
[0016] Reference Figure 1 As shown, a multi-task behavioral-EEG simultaneous monitoring method for schizophrenia screening includes: Based on clinical screening needs, a multi-task temporal loading paradigm is designed, which includes a sensory gating test task, a semantic and auditory oddball hybrid test task, and a reward feedback test task. Simultaneously collect EEG signals and behavioral data of subjects during the execution of a multi-task temporal loading paradigm, and preprocess the collected raw data; Multidimensional features are extracted from the preprocessed data, including EEG feature groups, behavioral feature groups, and time-varying feature groups. A multimodal fusion discriminant model is constructed, and multidimensional features are input into the multimodal fusion discriminant model to output the basic risk probability and the schizophrenia risk index.
[0017] This can be explained by the fact that schizophrenia is not a single cognitive dysfunction, but rather a systemic impairment across multiple cognitive dimensions, including sensory gating, semantic processing, and reward motivation. Therefore, a single-task paradigm can only assess one dimension and cannot fully capture the patient's pathological characteristics. This scheme designs a multi-task temporal loading paradigm to sequentially induce neurophysiological responses related to the aforementioned dimensions within a continuous timeframe, providing a data foundation for multidimensional assessment. EEG signals objectively reflect the duration and intensity of the brain's neural response to stimuli, while behavioral data (such as key press reaction time, accuracy, and false alarm rate) directly reflect the subject's cognitive executive function. These two aspects corroborate and complement each other, providing a more complete characterization of the subject's cognitive state. Specifically, EEG characteristics (such as P50 inhibition ratio, MMN amplitude, N400 amplitude, FRN differential waveform, and phase lock value) correspond to neural processing functions in different cognitive dimensions; behavioral characteristics (false alarm rate, reaction time coefficient of variation, and task dropout rate) reflect impulsivity, motivational stability, and attention maintenance ability; and time-varying characteristics (sample entropy decay coefficient) capture the neurodynamic changes in subjects during long-term tasks. The combination of these three elements achieves a comprehensive characterization from static to dynamic, and from neural to behavioral aspects. By constructing a multimodal fusion discriminant model, multidimensional features are organically integrated to output a risk index and anomaly dimension radar chart. This model can comprehensively assess the degree of abnormality in each dimension and provide intuitive visualization results, assisting clinicians in quickly locating the cognitive impairment characteristics of subjects, thereby achieving efficient and accurate schizophrenia screening.
[0018] The design of the multi-task temporal loading paradigm based on clinical screening needs specifically includes: The first stage is a sensory gating test task, in which paired click stimuli are played to the subjects. Each pair of clicks consists of a first click and a second click with a first preset time interval. The interval between the pairs of clicks is a second preset time. At the same time, the subjects are asked to silently count the total number of pairs of clicks played. The second phase is a mixed semantic and auditory oddball test task, in which a series of word stimuli are played continuously to the subjects, and the subjects are required to press a key when they hear a specific category of words; Among them, the standard stimulus is a frequently occurring common word, and the deviation stimulus includes two categories: physical deviation stimulus and semantic deviation stimulus. The physical deviation stimulus is a high-pitched word that changes the speech frequency, and the semantic deviation stimulus is an abnormal word that is unrelated to the context. The third stage is a reward feedback test task, in which a simplified game interface is presented to the subjects, and three results are given through visual feedback: win, lose, and draw. At the same time, the reaction time of the subjects from the presentation of the stimulus to pressing the button to place a bet is recorded.
[0019] Understandably, mental illnesses often manifest as functional abnormalities across multiple cognitive dimensions, and a single-task paradigm is insufficient to comprehensively assess the cognitive impairment characteristics of patients. This approach designs a multi-task sequential loading paradigm to sequentially induce neuroelectrophysiological responses across multiple cognitive dimensions, including sensory gating, auditory-physical bias processing, semantic bias processing, and reward-feedback processing, thereby achieving a systematic and multidimensional assessment of cognitive function. Specifically: The first phase is a sensory gating test task, which uses paired click stimuli to induce the P50 component and assess the brain’s ability to inhibit repetitive sensory information. This indicator is a core biological marker of schizophrenia. The second stage is a mixed semantic and auditory oddball test task. Physical deviation stimulation induces mismatch negative waves (MMN) to reflect early auditory processing function, while semantic deviation stimulation induces N400 components to reflect higher semantic processing integration ability. Both target characteristic damage of diseases such as schizophrenia, autism and Alzheimer's disease, respectively. The third stage is the reward feedback test task, which uses the feedback of winning and losing in a game to induce feedback-related negative waves (FRN), and combines reaction time analysis to evaluate reward processing and decision-making functions. This indicator has important value in the identification of depression and addictive behaviors. Through the sequential loading of these three-stage tasks, this solution can simultaneously collect multidimensional neuroelectrophysiological indicators related to various mental illnesses, providing a data foundation for subsequent analysis and screening. It should be noted that the first preset duration is usually 500ms, used to evaluate the P50 inhibition ratio; the second preset duration is usually 8-10 seconds, to avoid mutual interference between the stimuli before and after. It should be noted that this solution focuses on schizophrenia screening as its core application scenario. The specific categories of words mentioned can be selected based on the clinical characteristics of schizophrenia, using either threatening or neutral terms. However, the specific categories of words provided in this solution are not limited to schizophrenia screening and can also be extended to the screening of other mental disorders. The specific word types and content can be adaptively adjusted according to the screening objectives.
[0020] The simultaneous acquisition of EEG signals and behavioral data from subjects during the execution of the multi-task temporal loading paradigm, and the preprocessing of the acquired raw data specifically include: A high-density EEG acquisition device was used to record the EEG signals of the subjects during the performance of multiple tasks, and the subjects' key press reaction time, key press accuracy, missed key presses and false alarms were recorded simultaneously. The acquired raw EEG signals are preprocessed, including: removing electrooculography and electromyography artifacts using independent component analysis, removing bad channels and repairing them by neighborhood interpolation, filtering with a bandpass filter, and eliminating signal drift using a baseline correction method. Based on the stimulus presentation timestamp and key press timestamp, the preprocessed EEG signals are precisely aligned with the behavioral event tags to form a synchronized dataset.
[0021] This can be explained by the fact that EEG signals are highly susceptible to interference from physiological artifacts such as eye movements, blinking, and electromyography (EMG). If these artifacts are not effectively removed, they will severely affect the accuracy of subsequent feature extraction. Independent component analysis (ICA) decomposes multi-channel EEG signals into statistically independent components, which can accurately identify and remove artifact components representing EEG and EMG, while preserving the brain's true neural electrical activity. The bandpass filter is set to 0.5-45Hz because key electrophysiological indicators related to mental illness are mainly distributed in this frequency band: the P50 component is located around 50ms, which is a high-frequency band; event-related potentials such as MMN and N400 are located between 100-500ms, which is a mid-to-low-frequency band; and the theta wave energy changes related to reward processing are mainly concentrated in 4-7Hz. Too wide a frequency band will introduce unnecessary noise such as power frequency and electromagnetic interference, while too narrow a frequency band will filter out effective neural electrical signals. Bad channel removal and interpolation can ensure the data integrity of high-density EEG channels. Precise alignment of behavioral event markers with EEG signals is a key aspect of this approach. Simultaneously with each stimulus, a synchronous trigger pulse is sent to the EEG acquisition system, marking the precise start time of the stimulus. When the subject presses a key, the system also records the timestamp of the key press. Through precise alignment of the trigger pulse and the timestamp, the EEG segment corresponding to each stimulus can be accurately extracted, providing a unified and reliable time reference for subsequent event-related potential superposition averaging and feature extraction. Simultaneous recording of behavioral data such as missed key presses and false alarms also provides a complete behavioral basis for subsequent assessment of the subject's task engagement and cognitive function.
[0022] Reference Figure 2 As shown, the extraction of EEG feature sets from the preprocessed data specifically includes: Calculate the P50 suppression ratio based on paired click sound task data; Based on semantic and auditory oddball task data, the peak value of the differential wave induced by the deviation stimulus at the Fz electrode was selected as the MMN amplitude, and the time from the stimulus onset to the peak value was selected as the MMN latency. Based on semantic and auditory oddball task data, calculate the average N400 amplitude of the central parietal electrode within the [a,b] time window after stimulation; Based on the reward feedback task data, calculate the average amplitude of the feedback-related negative wave difference wave induced by output feedback and win feedback within the [c,d] time window after feedback; Based on data from the entire mission period, the average phase-locking values of the prefrontal and temporal electrodes in the [e,f] frequency band were calculated.
[0023] This approach is used to extract multidimensional indicators from synchronized data that can quantify the core pathological characteristics of schizophrenia. Specifically: The P50 inhibition ratio is a classic indicator for assessing sensory gating function. In healthy individuals, due to the normal operation of the sensory gating mechanism, the P50 inhibition ratio is significantly lower than the preset theoretical baseline value. However, in patients with schizophrenia, due to impaired sensory gating function, the P50 inhibition ratio is significantly increased, and its value approaches or even exceeds the theoretical baseline value. MMN and N400 reflect the brain's ability to automatically process changes in physical stimuli and semantic content, respectively. A decrease in MMN amplitude indicates a decline in the brain's ability to automatically detect changes in the environment, which is consistent with the patient's behavior of being slow to respond to or overly sensitive to external stimuli. An abnormal N400 indicates a disorder in the activation of the semantic network, which is directly related to the patient's clinical symptoms of incoherent speech and confused thinking. Feedback-related negative waves and theta phase synchronization values reflect reward processing function and brain network integration function, respectively. Abnormal amplitude of feedback-related negative waves indicates functional abnormality of the reward circuit, corresponding to loss of pleasure in negative symptoms. Theta phase synchronization value reflects the strength of functional connectivity between the prefrontal and temporal lobes, and its reduction indicates impaired brain network integration function. The P50 suppression ratio = the amplitude of the P50 component induced by the second click sound / the amplitude of the P50 component induced by the first click sound; It should be noted that the P50 amplitude is taken as the positive peak value within a specific time window after the first and second click stimuli. This time window is determined by adding or subtracting a certain number of standard deviations from the average latency of the population. The intervals [a,b], [c,d], and [e,f] are determined based on data statistics, such as the 3σ criterion.
[0024] Reference Figure 3 As shown, the extraction of behavioral feature groups from the preprocessed data specifically includes: Based on semantic and auditory oddball task data, the number of incorrect key presses made by the subjects on non-target stimuli was counted, and the total number of non-target stimuli was also counted. The false alarm rate is calculated based on the number of incorrect key presses made by the subject on non-target stimuli and the total number of non-target stimuli. Based on the reward feedback task data, the reaction time of the subjects in all valid key press attempts was statistically analyzed, and the standard deviation and mean of the reaction time were calculated. Calculate the coefficient of variation for reaction time based on the standard deviation and mean of the reaction time; Based on data from the entire task period, the number of missed responses by subjects on all target stimuli was counted, and the total number of target stimuli was also counted. The overall task dropout rate is calculated based on the number of missed responses by the subjects to all target stimuli and the total number of target stimuli.
[0025] This can be explained by the fact that traditional EEG studies often only use behavioral data as quality control indicators, while this protocol uses behavioral characteristics as one of the core diagnostic features. The false report rate reflects the degree of impulsivity and thought disorder of the subjects under uncertain conditions. Even in cases where EEG indicators are atypical, a high false report rate may itself indicate a thought disorder subtype of schizophrenia. The coefficient of variation of reaction time reflects the subjects' motivational stability and attention maintenance ability. An excessively large coefficient of variation suggests diminished willpower and emotional apathy. The overall task dropout rate reflects the subjects' attention maintenance ability and task participation. A high dropout rate suggests inattention and insufficient initiative, which are typical behavioral manifestations of negative symptoms of schizophrenia. The false alarm rate = (Number of incorrect key presses / Total number of non-target stimuli) × 100%; The coefficient of variation during reaction time = standard deviation during reaction time / mean during reaction time; The dropout rate is calculated as: (Number of missed responses / Total number of target stimuli) × 100%, where " / " represents division.
[0026] Reference Figure 4 The extraction of time-varying feature groups from the preprocessed data, as shown, specifically includes: Divide the entire task time period into several consecutive windows of equal length; Calculate the sample entropy value of the EEG signal within each window; Linear regression fitting was performed on the midpoint of the window time and the corresponding sample entropy value to obtain the regression slope, which was then used as the EEG signal complexity attenuation coefficient.
[0027] This can be explained by the fact that the neurodynamic characteristics of schizophrenic patients may change significantly during prolonged tasks. Some patients exhibit a rapid decrease in EEG complexity, suggesting that the brain is gradually entering a rigid state; while others exhibit persistently high complexity, indicating that their thinking remains in a state of confusion. This method can quantify this dynamic change process by calculating the decay coefficient of sample entropy, providing supplementary information in the time dimension for screening. The expression for the attenuation coefficient is:
[0028] In the formula, This is the attenuation coefficient for the complexity of the EEG signal. Let be the midpoint of time for the i-th window, be the sample entropy value for the i-th window, and n be the total number of windows; Sample entropy is a core indicator for measuring the regularity and complexity of time series data. A higher value indicates a more irregular signal and a more disordered information processing capacity in the brain; a lower value indicates a more rigid signal and decreased neuroplasticity in the brain. This is achieved by calculating the attenuation coefficient. It can quantitatively assess the rate of change in EEG complexity of subjects throughout the task: if A significantly negative value with an excessively large absolute value suggests rapid brain fatigue or functional decline in the subject; if... A value close to zero or significantly positive indicates that the subject's brain network is in a state of disorder or unable to adapt to the task. These dynamic characteristics are important objective evidence for assisting in the screening of schizophrenia.
[0029] The construction of the multimodal fusion discriminant model, which inputs multidimensional features into the model and outputs a basic risk probability and a schizophrenia risk index, specifically includes: A multimodal fusion discriminant model is constructed, which includes a first-layer feature-level fusion classifier and a second-layer decision-level fusion regulator. First-level feature-level fusion classifier: Combines EEG feature groups and behavioral feature groups to form a feature vector, inputs it into a support vector machine classifier, and outputs the basic risk probability; The second-level decision-level fusion regulator introduces the attenuation coefficient in the time-varying feature group as a regulation factor to dynamically correct the basic risk probability and output the schizophrenia risk index. Based on the final risk index and the degree of abnormality of each characteristic indicator, an abnormality dimension radar chart is generated. The radar chart includes four dimensions: sensory gating, semantic processing, reward motivation, and brain network connectivity.
[0030] This can be explained by the fact that the first layer of feature fusion uses a support vector machine as a classifier, which has the advantage of good classification performance for small samples and high-dimensional data. The EEG feature group and the behavioral feature group together constitute a multi-dimensional characterization of the subject's functional state. The support vector machine finds the optimal classification hyperplane, organically combines the two types of features, and outputs a basic risk probability between 0 and 1. The second-level decision-making fusion incorporates time-varying features as a moderating factor. In actual clinical testing, some subjects may experience poor behavioral data quality due to tension, fatigue, or lack of cooperation. In such cases, relying solely on the risk probability output from the first level may lead to inaccuracies. This approach dynamically adjusts the baseline risk probability by monitoring the trend of EEG complexity changes throughout the task: if a subject experiences a rapid decrease or continuous abnormal fluctuation in EEG complexity in the latter half of the task, indicating abnormal neurodynamic characteristics, the risk index is appropriately increased; if a subject maintains stable EEG complexity throughout the task, even if the baseline risk probability is slightly high, the risk index can be appropriately decreased. This effectively solves the practical problem of variable subject states and varying levels of cooperation in psychiatric examinations. The expression for the basic risk probability is:
[0031] In the formula, P is the basic risk probability, and f SVM Here, A is the decision function of the support vector machine, and B is the vector of EEG feature groups and B is the vector of behavioral feature groups. The expression for the schizophrenia risk index is:
[0032] In the formula, P final This is a risk index for schizophrenia. To adjust the weighting coefficients, This is the normalized EEG complexity attenuation coefficient; It should be noted that the adjustment weight coefficient Based on historical or experimental data, the grid search optimization on the training set is performed using cross-validation to determine the optimal screening accuracy.
[0033] Furthermore, the method according to the embodiments of this application can also be achieved by means of... Figure 5 The architecture of the electronic device shown is used to implement this. For example... Figure 5 As shown, the electronic device 500 may include a bus 501, one or more CPUs 502, a read-only memory (ROM) 503, a random access memory (RAM) 504, a communication port 505 connected to a network, an input / output component 506, a hard disk 507, etc. The storage device in the electronic device 500, such as the ROM 503 or the hard disk 507, may store a multi-task behavioral-EEG synchronous monitoring method for schizophrenia screening provided in this application. The electronic device 500 may also include a user interface 508. Of course, Figure 5 The architecture shown is merely exemplary and can be omitted as needed when implementing different devices. Figure 5 One or more components in the illustrated electronic device.
[0034] Figure 6 This is a schematic diagram of a computer-readable storage medium structure provided in one embodiment of this application. Figure 6 The diagram illustrates a computer-readable storage medium 600 according to one embodiment of this application. The computer-readable storage medium 600 stores computer-readable instructions. When executed by a processor, the computer-readable instructions can perform a multi-task behavioral-EEG synchronous monitoring method for schizophrenia screening according to an embodiment of this application, as described with reference to the above figures. The storage medium 600 includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.
[0035] In summary, the advantages of this invention are as follows: by synchronously collecting EEG and behavioral data through a multi-task temporal loading paradigm, extracting static and dynamic multi-dimensional features, and using a multi-modal fusion discriminant model for dynamic correction, it achieves efficient, accurate, and objective screening of the core cognitive dimensions of schizophrenia, significantly improving the sensitivity and specificity of the screening.
[0036] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection of the present invention is defined by the appended claims and their equivalents.
Claims
1. A multi-task behavior-brain electroencephalogram synchronous monitoring method for schizophrenia screening, characterized in that, include: Based on clinical screening needs, a multi-task temporal loading paradigm is designed, which includes a sensory gating test task, a semantic and auditory oddball hybrid test task, and a reward feedback test task. Simultaneously collect EEG signals and behavioral data of subjects during the execution of a multi-task temporal loading paradigm, and preprocess the collected raw data; Multidimensional features are extracted from the preprocessed data, including EEG feature groups, behavioral feature groups, and time-varying feature groups. A multimodal fusion discriminant model is constructed, and multidimensional features are input into the multimodal fusion discriminant model to output the basic risk probability and the schizophrenia risk index.
2. A multi-task behavior-EEG synchronous monitoring method for schizophrenia screening according to claim 1, characterized in that, The design of the multi-task temporal loading paradigm based on clinical screening needs specifically includes: The first stage is a sensory gating test task, in which paired click stimuli are played to the subjects. Each pair of clicks consists of a first click and a second click with a first preset time interval. The interval between the pairs of clicks is a second preset time. At the same time, the subjects are asked to silently count the total number of pairs of clicks played. The second phase is a mixed semantic and auditory oddball test task, in which a series of word stimuli are played continuously to the subjects, and the subjects are required to press a key when they hear a specific category of words; Among them, the standard stimulus is a frequently occurring common word, and the deviation stimulus includes two categories: physical deviation stimulus and semantic deviation stimulus. The physical deviation stimulus is a high-pitched word that changes the speech frequency, and the semantic deviation stimulus is an abnormal word that is unrelated to the context. The third stage is a reward feedback test task, in which a simplified game interface is presented to the subjects, and three results are given through visual feedback: win, lose, and draw. At the same time, the reaction time of the subjects from the presentation of the stimulus to pressing the button to place a bet is recorded.
3. The multi-task behavioral-EEG synchronous monitoring method for schizophrenia screening according to claim 2, characterized in that, The simultaneous acquisition of EEG signals and behavioral data from subjects during the execution of a multi-task temporal loading paradigm, and the preprocessing of the acquired raw data specifically include: A high-density EEG acquisition device was used to record the EEG signals of the subjects during the performance of multiple tasks, and the subjects' key press reaction time, key press accuracy, missed key presses and false alarms were recorded simultaneously. The acquired raw EEG signals are preprocessed, including: removing electrooculography and electromyography artifacts using independent component analysis, removing bad channels and repairing them by neighborhood interpolation, filtering with a bandpass filter, and eliminating signal drift using a baseline correction method. Based on the stimulus presentation timestamp and key press timestamp, the preprocessed EEG signals are precisely aligned with the behavioral event tags to form a synchronized dataset.
4. The multi-task behavioral-EEG synchronous monitoring method for schizophrenia screening according to claim 3, characterized in that, The extraction of EEG feature sets from the preprocessed data specifically includes: Calculate the P50 suppression ratio based on paired click sound task data; Based on semantic and auditory oddball task data, the peak value of the differential wave induced by the deviation stimulus at the Fz electrode was selected as the MMN amplitude, and the time from the stimulus onset to the peak value was selected as the MMN latency. Based on semantic and auditory oddball task data, calculate the average N400 amplitude of the central parietal electrode within the [a,b] time window after stimulation; Based on the reward feedback task data, calculate the average amplitude of the feedback-related negative wave difference wave induced by output feedback and win feedback within the [c,d] time window after feedback; Based on data from the entire mission period, the average phase-locking values of the prefrontal and temporal electrodes in the [e,f] frequency band were calculated.
5. The multi-task behavioral-EEG synchronous monitoring method for schizophrenia screening according to claim 4, characterized in that, The extraction of behavioral feature groups from the preprocessed data specifically includes: Based on semantic and auditory oddball task data, the number of incorrect key presses made by the subjects on non-target stimuli was counted, and the total number of non-target stimuli was also counted. The false alarm rate is calculated based on the number of incorrect key presses made by the subject on non-target stimuli and the total number of non-target stimuli. Based on the reward feedback task data, the reaction time of the subjects in all valid key press attempts was statistically analyzed, and the standard deviation and mean of the reaction time were calculated. Calculate the coefficient of variation for reaction time based on the standard deviation and mean of the reaction time; Based on data from the entire task period, the number of missed responses by subjects on all target stimuli was counted, and the total number of target stimuli was also counted. The overall task dropout rate is calculated based on the number of missed responses by the subjects to all target stimuli and the total number of target stimuli.
6. The multi-task behavioral-EEG synchronous monitoring method for schizophrenia screening according to claim 5, characterized in that, The extraction of time-varying feature groups from the preprocessed data specifically includes: Divide the entire task time period into several consecutive windows of equal length; Calculate the sample entropy value of the EEG signal within each window; Linear regression fitting was performed on the midpoint of the window time and the corresponding sample entropy value to obtain the regression slope, which was then used as the EEG signal complexity attenuation coefficient.
7. The multi-task behavioral-EEG synchronous monitoring method for schizophrenia screening according to claim 6, characterized in that, The construction of the multimodal fusion discriminant model, which inputs multidimensional features into the model and outputs a basic risk probability and a schizophrenia risk index, specifically includes: A multimodal fusion discriminant model is constructed, which includes a first-layer feature-level fusion classifier and a second-layer decision-level fusion regulator. First-level feature-level fusion classifier: Combines EEG feature groups and behavioral feature groups to form a feature vector, inputs it into a support vector machine classifier, and outputs the basic risk probability; The second-level decision-level fusion regulator introduces the attenuation coefficient in the time-varying feature group as a regulation factor to dynamically correct the basic risk probability and output the schizophrenia risk index. Based on the final risk index and the degree of abnormality of each characteristic indicator, an abnormality dimension radar chart is generated. The radar chart includes four dimensions: sensory gating, semantic processing, reward motivation, and brain network connectivity.
8. An electronic device, characterized in that, include: At least one processor; And, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform a multi-task behavioral-EEG synchronous monitoring method for schizophrenia screening as described in any one of claims 1-7.
9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the multi-task behavioral-EEG synchronous monitoring method for schizophrenia screening as described in any one of claims 1-7.