A cognitive conflict risk assessment apparatus and portable device

Abstract

This invention discloses a cognitive conflict risk assessment device and a portable device. The device includes a task stimulus presentation module, a test data acquisition module, a cognitive difference calculation module, and a risk index generation module. The task stimulus presentation module provides a Stroop test task to the test subject. The test data acquisition module collects behavioral and physiological response signals generated by the test subject when completing the Stroop test task. The cognitive difference calculation module calculates the response difference values ​​of the test subject to semantic stimuli related to a preset topic and unrelated neutral semantic stimuli based on the behavioral and physiological response signals. The risk index generation module inputs the response difference values ​​into a risk assessment model optimized using the Monte Carlo method to generate and output a quantitative risk index representing the cognitive conflict level of the test subject. This invention can accurately achieve a quantitative assessment of the cognitive conflict risk level.

Description

Technical Field

[0001] This invention relates to the field of bioinformatics technology, and more particularly to a cognitive conflict risk assessment device and a portable device. Background Technology

[0002] Assessing cognitive control function, particularly an individual's conflict coping ability in the face of stimuli with emotional or addictive implications, is a crucial topic in psychology, neuroscience, and clinical medicine. Traditional assessment methods primarily rely on highly subjective psychological scales and questionnaires. While these methods are simple to implement, they have significant drawbacks: First, the assessment results are highly susceptible to the social desirability effect, meaning that participants may deliberately provide false answers for purposes such as avoiding responsibility, leading to data distortion and compromising the accuracy and reliability of the assessment. Second, these methods are static, retrospective, and subjective reports, unable to capture in real-time, objective micro-behavioral and neurophysiological changes at the moment of cognitive conflict, thus failing to reveal cognitive deficits.

[0003] To improve objectivity, the classic Stroop color-word test paradigm has been introduced as a tool for assessing cognitive flexibility. This paradigm creates cognitive conflict by asking participants to name the printed color while ignoring the semantics of words. However, the traditional Stroop paradigm often uses general affective words or color words themselves as stimuli, lacking specificity and sensitivity in particular domains, making it difficult to effectively induce specific cognitive conflict in specific populations. Furthermore, traditional analytical methods are usually limited to comparing macro-average values ​​of behavioral data, lacking the ability to simultaneously collect and deeply integrate multimodal physiological signals such as eye movements and electroencephalograms, and are even less capable of accurately probabilistically estimating individual-level risks.

[0004] In recent years, although some studies have attempted to apply computational models to cognitive assessment, most models are trained on small sample data, resulting in weak generalization ability and unstable assessment results when faced with individual differences. In particular, how to construct a robust model that can quantify uncertainty and provide individualized risk ranges from limited data remains a challenge and a gap in current technology. Therefore, there is an urgent need in this field for an integrated device that can dynamically present specific stimuli, simultaneously collect multi-dimensional response signals, and use advanced algorithm models for integrated analysis and risk assessment, in order to overcome the technical shortcomings of existing technologies, such as strong subjectivity, low specificity, and inability to quantify the level of individual cognitive conflict risk. Summary of the Invention

[0005] The present invention aims to at least partially address the technical problems in related technologies. Therefore, the first objective of the present invention is to provide a cognitive conflict risk assessment device that can quantitatively assess the level of cognitive conflict risk for a specific population in real time and objectively.

[0006] A second objective of this invention is to provide a portable device.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solution: A cognitive conflict risk assessment device, comprising: The task stimulus presentation module is used to provide the test subject with a Stroop test task, which is to respond to semantic stimulus units that are related to and unrelated to a preset topic and neutral semantic stimulus units. The test data acquisition module is connected to the task stimulus presentation module. The test data acquisition module is used to collect the behavioral response signals and physiological response signals generated by the individual under test when completing the Stroop test task. A cognitive difference calculation module is connected to the test data acquisition module. The cognitive difference calculation module is used to calculate the response difference value of the test individual to semantic stimulus units related to the preset topic and unrelated neutral semantic stimulus units based on the behavioral response signal and physiological response signal, respectively. A risk index generation module is connected to the cognitive difference calculation module. The risk index generation module is used to input the response difference value into a risk assessment model optimized by the Monte Carlo method, and generate and output a quantitative risk index that can represent the cognitive conflict level of the individual being tested.

[0008] Preferably, the task stimulus presentation module is specifically used for: The system retrieves semantic stimuli that are related to or unrelated to the preset topic, as well as neutral semantic stimuli, from a preset semantic stimulus library, and matches each semantic stimulus with an interference color to form a corresponding semantic stimulus unit.

[0009] Preferably, the device further includes a semantic stimulus library construction module connected to the task stimulus presentation module. The semantic stimulus library construction module is used to construct the preset semantic stimulus library. Specifically, the semantic stimulus library construction module is used for: A core seed lexicon should be established in advance; A corpus extraction model is used to extract corpus related to the core words in the core seed lexicon from the network; The hidden language extraction model is used to extract hidden language related to the preset topic from the captured corpus, and the related perceptual words are extracted from the hidden language corpus. The core words, cryptic terms, and perceptual words are mixed and sorted using TF-RF scoring. The words with preset values ​​before scoring are selected to construct the preset semantic stimulus library.

[0010] Preferably, the test data acquisition module includes: The behavioral data acquisition unit is used to collect the reaction time and error rate of the individual under test when completing the task of judging the color of the semantic stimulus unit as the behavioral response signal; The physiological data acquisition unit operates synchronously with the behavioral data acquisition unit. The physiological data acquisition unit is used to collect pupil diameter change data and electroencephalogram (EEG) signals of the individual under test as the physiological response signals.

[0011] Preferably, after acquiring the electroencephalogram (EEG) signal, the physiological data acquisition unit is also used to extract the P300 wave latency and N400 wave amplitude from the EEG signal.

[0012] Preferably, the cognitive difference calculation module is specifically used for: The average reaction time, error rate, average P300 latency, and average N400 amplitude of the test subjects for semantic stimulus units related to the preset topic and for neutral semantic stimulus units unrelated to the preset topic were calculated respectively. The differences between the average reaction time, error rate, average P300 latency, and average N400 amplitude were also calculated to obtain the difference values ​​of reaction time, error rate, P300 latency, and N400 amplitude.

[0013] Preferably, the risk assessment model is trained in the following manner: Obtain an initial training dataset, which includes multiple difference values ​​from multiple testers. data; Based on the initial training dataset, random sampling is performed using the Markov chain Monte Carlo method to generate a simulated dataset. The initial risk assessment model is fitted based on the simulated dataset to construct a posterior probability distribution function that can output a quantitative risk index, thereby training the risk assessment model.

[0014] Preferably, when generating a quantitative risk index, the risk index generation module is specifically used for: Based on the posterior probability distribution function, the model output value is mapped to a preset probability interval to generate the quantitative risk index.

[0015] Preferably, it also includes a model optimization module connected to the risk index generation module, the model optimization module being used for: Receive verification feedback for the quantified risk index, and adjust the parameters of the risk assessment model and the sampling parameters of the Monte Carlo method through the backpropagation algorithm.

[0016] To achieve the above objectives, a second aspect of the present invention provides a portable device integrating the aforementioned cognitive conflict risk assessment device. The task stimulus presentation module and test data acquisition module of the cognitive conflict risk assessment device are implemented through the touch screen of the portable device, and the cognitive difference calculation module and risk index generation module of the cognitive conflict risk assessment device are executed through the processor of the portable device.

[0017] This invention has at least the following technical effects: This invention provides a cognitive conflict risk assessment device and a portable device. Through the collaborative work of a task stimulus presentation module and a test data acquisition module, the device achieves objectivity and multimodality in the cognitive conflict level assessment process. The device can use AI technology to dynamically generate highly specific Stroop task sequences, effectively avoiding the cheating risks associated with subjective questionnaires. Simultaneously, the test data acquisition module collects behavioral and physiological response signals, achieving, for the first time, a comprehensive and objective quantification of cognitive conflict from a behavioral to physiological level on a single device based on multimodal data, providing a rich and reliable data foundation for analysis. Furthermore, this invention deeply processes the raw data through a cognitive difference calculation module, extracting core response difference values. These response difference values ​​eliminate interference from irrelevant variables such as the individual's basic reaction speed, and more accurately pinpoint the cognitive conflict effect triggered by specific stimuli. This significantly improves the sensitivity and specificity of the cognitive conflict risk assessment, enabling the indicators provided by this invention to more accurately reflect an individual's cognitive control function status in specific dimensions. Furthermore, this invention introduces a risk assessment model optimized using the Monte Carlo method through a risk index generation module. This enables the generation of massive simulated samples based on limited training data, thereby constructing a stable probability distribution function. The device outputs a continuous, probabilistically statistically significant quantitative risk index, which completely changes the traditional black-and-white assessment model. It can present the severity of risk in probabilistic form, achieving a refined and quantitative assessment of cognitive conflict risk, and providing a more scientific and reliable basis for subsequent decision-making.

[0018] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0019] Figure 1 This is a structural block diagram of the cognitive conflict risk assessment device according to the first embodiment of the present invention.

[0020] Figure 2 This is a structural block diagram of the cognitive conflict risk assessment device according to the second embodiment of the present invention.

[0021] Figure 3This is a structural block diagram of the cognitive conflict risk assessment device according to the third embodiment of the present invention.

[0022] Figure 4 This is a structural block diagram of a portable device according to an embodiment of the present invention. Detailed Implementation

[0023] The following describes this embodiment in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the invention, and should not be construed as limiting the invention.

[0024] A cognitive conflict risk assessment device and portable device according to this embodiment are described below with reference to the accompanying drawings.

[0025] Figure 1 This is a schematic diagram of the cognitive conflict risk assessment device according to an embodiment of the present invention. Figure 1 As shown, the cognitive conflict risk assessment device 100 includes a task stimulus presentation module 10, a test data acquisition module 20, a cognitive difference calculation module 30, and a risk index generation module 40 connected in sequence. The task stimulus presentation module 10 provides the test subject with a Stroop test task, which involves responding to semantic stimuli related to a preset topic, as well as neutral semantic stimuli. The test data acquisition module 20, connected to the task stimulus presentation module, collects the behavioral and physiological response signals generated by the test subject when completing the Stroop test task. The cognitive difference calculation module 30, connected to the test data acquisition module 20, calculates the response difference values ​​of the test subject to semantic stimuli related to a preset topic and neutral semantic stimuli, respectively, based on the behavioral and physiological response signals. The risk index generation module 40, connected to the cognitive difference calculation module 30, inputs the response difference values ​​into a risk assessment model optimized using the Monte Carlo method, generating and outputting a quantitative risk index representing the cognitive conflict level of the test subject.

[0026] In this embodiment, the preset topic can be a nitrous oxide addiction test, meaning the above technical solution can achieve a cognitive conflict risk assessment corresponding to nitrous oxide addiction. It should be noted that the following will focus on nitrous oxide addiction as the preset topic to elaborate on this technical solution in detail.

[0027] In one embodiment of the present invention, the task stimulus presentation module 10 is specifically used to: call semantic stimuli related to and unrelated to the preset topic and neutral semantic stimuli from the preset semantic stimulus library, and match an interference color for each semantic stimulus to form a corresponding semantic stimulus unit.

[0028] Specifically, semantic stimuli related to nitrous oxide addiction, such as "nitrous oxide tank", "inhalation device", "nitrous oxide experience", "nitrous oxide inhalation", etc., can be called from the preset semantic stimulus library. At the same time, neutral semantic stimuli unrelated to nitrous oxide addiction, such as "table", "chair", "stool", "tissue paper", etc., are called, and then a interference color is matched to the picture corresponding to each word. In this embodiment, a color rendering algorithm is used to match three interference colors, red, yellow, and blue, to the two types of semantic stimuli, forming two types of experimental task groups, namely two types of semantic stimulus units, related semantic-interference color and neutral semantic-interference color. Each group contains 120 semantic stimulus units for testing to ensure data validity.

[0029] In an embodiment of the present invention, as Figure 2 shown, the device further includes a semantic stimulus library construction module 50, which is connected to the task stimulus presentation module 10. The semantic stimulus library construction module 50 is used to construct a preset semantic stimulus library through an AI text analysis technology screening method. Specifically, the semantic stimulus library construction module 50 is used to: establish a core seed word library in advance; use a corpus scraping model to scrape the corpus related to the core words in the core seed word library from the network; use a jargon extraction model to extract the jargons related to the preset theme from the scraped corpus, and extract the feeling words related to the jargons from the jargon corpus; mix the core words, jargons and feeling words, and sort them using TF-RF scoring, and select the vocabulary with the top preset values to construct the preset semantic stimulus library.

[0030] Specifically, first, semantic stimuli related to the nitrous oxide theme are screened through AI text analysis technology. For example, a core seed word library of themes such as "nitrous oxide", "N2O", "balloon" is constructed first, and BERTopic (a natural language processing model for topic clustering) is used to cluster and scrape Weibo, Tieba, and e-commerce reviews, and high-emotion polarity sentences are filtered; then RoBERTa-CRF (a natural language processing model that combines context semantic understanding and sequence label annotation) is trained to identify jargons such as "blowing balloons", "cream gun", etc., and combine dependency syntax to extract feeling words related to the jargons, such as "refreshed", "high", etc. Sort by TF-RF (term frequency-inverse document frequency) scoring, retain the top 5% high-frequency semantic stimuli, and then manually review the toxicity to output a light quantifier table for constructing the preset semantic stimulus library.

[0031] Among them, TF represents the term frequency, that is, the frequency of a word appearing in the nitrous oxide-related corpus. RF represents the inverse document frequency, which measures the specificity of a word to the nitrous oxide theme. If a word such as "feeling dizzy" appears frequently in the nitrous oxide corpus but rarely in ordinary corpora such as news, its RF value is high. TF-RF represents the comprehensive score of term frequency-inverse document frequency, that is, to find those words that appear frequently in the discussion of nitrous oxide and are also discriminative for other topics. It can filter out high-frequency but meaningless words such as "of" and "is", and can also highlight words with strong specificity such as "feeling dizzy". Further, only the top 5% of the words with the highest TF-RF scores are retained to ensure that the selected words are all highly representative high-purity stimulus words. Finally, let domain experts such as psychologists and doctors conduct manual review on the final word list selected by the machine, and finally form a core word list with appropriate quantity and reliable quality for the Stroop experiment.

[0032] In an embodiment of the present invention, the test data acquisition module 20 includes a behavioral data acquisition unit and a physiological data acquisition unit. The behavioral data acquisition unit is used to collect the reaction time and error rate of the待测个体 when completing the color judgment task of the semantic stimulus unit as the behavioral response signal; the physiological data acquisition unit runs synchronously with the behavioral data acquisition unit, and the physiological data acquisition unit is used to collect the pupil diameter change data and electroencephalogram signal of the待测个体 as the physiological response signal. After collecting the electroencephalogram signal, the physiological data acquisition unit is further used to extract the P300 wave latency and N400 wave amplitude from the electroencephalogram signal.

[0033] Specifically, before the待测个体 performs the Stroop experiment, a basic test is first performed to exclude the influence of other factors such as the shape factor of the semantic stimulus unit on cognitive conflict. For example, first judge the shape of the word picture, such as quadrilateral, hexagon, octagon, press the blue button for quadrilateral, the green button for hexagon, and the red button for octagon. After the basic test, semantic stimulus units related and unrelated to the preset theme and neutral semantic stimulus units are presented. Among them, while presenting different shapes, the text content of the two types of stimulus units includes the first type, that is, words related to the preset theme screened by AI text analysis technology, such as "nitrous oxide tank", "inhalation device", "nitrous oxide feeling", "nitrous oxide inhalation", etc., and the second type of neutral words such as "table", "chair", "stool", "tissue paper", etc. Thus, this embodiment can solve the problem of insufficient pertinence and sensitivity in a specific field through AI text analysis technology, can effectively induce specific cognitive conflict in specific populations, and achieve the risk assessment of cognitive conflict in specific populations.

[0034] It should be noted that the "待测个体" in the above translation is a placeholder that needs to be replaced with the specific term in the original context.The task group consisting of relevant semantic-distractor colors related to the preset theme comprises 120 nitrous oxide-related words randomly paired with three colors. The task group consisting of neutral semantic-distractor colors unrelated to the preset theme comprises 120 neutral words randomly paired with three colors.

[0035] Furthermore, semantic stimulus units, or word images, are rapidly displayed on the device screen. The individual's task is to press buttons as quickly and accurately as possible to indicate the color displayed by the semantic stimulus unit; for example, pressing the R button for red and the Y button for yellow. Then, pupil diameter changes, reaction time, and error rate are recorded using eye-tracking sensors and response buttons. Additionally, an EEG cap is used to collect EEG signals from the frontal and parietal lobes of the brain (brain regions responsible for higher cognitive control and attention allocation, respectively).

[0036] In this embodiment, behavioral data includes reaction time and error rate. Reaction time is the time (in milliseconds) from the appearance of a word to the individual pressing the correct color key. A longer reaction time indicates stronger cognitive conflict and greater difficulty in processing. The error rate is the proportion of incorrect color keys pressed; a higher error rate indicates poorer cognitive control. Physiological data includes pupil diameter changes and neural data, i.e., electroencephalogram (EEG) signals. Pupil diameter changes reflect cognitive load and emotional arousal; pupils dilate during conflict tasks. In EEG signals, P300 waves are typically associated with attention allocation and information evaluation; a prolonged P300 wave latency indicates a slower information processing speed. N400 waves are typically associated with semantic conflict processing; an increased N400 wave amplitude indicates greater difficulty in integrating semantic information.

[0037] In this embodiment, the reaction time, error rate, and pupil diameter changes of the individual being tested during the color judgment task are recorded by eye tracking. The amplitude and latency data of N400 and P300 waves are captured by the collected EEG signals from the frontal and parietal lobes. Then, the multimodal data are denoised and feature extracted in real time by AI algorithm, which can avoid the delay and error of manual processing.

[0038] In one embodiment of the present invention, the cognitive difference calculation module is specifically used to: calculate the average reaction time, error rate, average P300 wave latency, and average N400 wave amplitude of the test individual for semantic stimulus units related to a preset topic, and the average reaction time, error rate, average P300 wave latency, and average N400 wave amplitude of neutral semantic stimulus units unrelated to the preset topic, and calculate the differences between the average reaction time, error rate, average P300 wave latency, and average N400 wave amplitude, respectively, to obtain the difference values ​​of reaction time, error rate, P300 wave latency, and N400 wave amplitude.

[0039] In this embodiment, by deeply processing the raw data, the core response difference value is extracted. The response difference value eliminates the interference of irrelevant variables such as the basic reaction speed of the subjects, and can more accurately pinpoint the cognitive conflict effect caused by specific stimuli. This greatly improves the sensitivity and specificity of cognitive conflict risk assessment, making the indicators provided in this embodiment more realistically reflect the cognitive control function status of individuals in specific dimensions.

[0040] After obtaining multiple differential values, these values ​​are combined with multiple clinical indicators and input into a risk assessment model to conduct a risk assessment of the level of cognitive conflict. Specifically, this assessment can be performed by outputting a quantitative risk index.

[0041] In one embodiment of the present invention, the risk assessment model is trained in the following manner: an initial training dataset is obtained, which includes multiple differential value data of multiple testers; based on the initial training dataset, random sampling is performed using the Markov chain Monte Carlo method to generate a simulated dataset; the initial risk assessment model is fitted based on the simulated dataset to construct a posterior probability distribution function that can output a quantitative risk index, so as to train the risk assessment model.

[0042] Specifically, the risk index generation module is used to generate a quantitative risk index by mapping the model output value to a preset probability interval based on the posterior probability distribution function.

[0043] Furthermore, such as Figure 3 As shown, the device also includes a model optimization module 60, which is connected to the risk index generation module 40. The model optimization module 60 is used to: receive verification feedback for the quantitative risk index, and adjust the parameters of the risk assessment model and the sampling parameters of the Monte Carlo method through the backpropagation algorithm.

[0044] In this embodiment, the risk assessment model construction or training method is as follows: the first step is variable selection and weight allocation; the second step is MCM (Monte Carlo method) simulation and probability distribution construction; and the third step is model validation and AI optimization.

[0045] Specifically, the first step involves selecting 120 research subjects, including 40 nitrous oxide addicts, 40 individuals with a history of nitrous oxide exposure but not yet addicted, and 40 healthy individuals with no history of exposure. Before conducting the Stroop experiment, basic tests are performed. After the basic tests, the Stroop experiment is conducted, and individual cognitive index data (variable values) are obtained from the 120 research subjects. These individual cognitive index data—reaction time variation, error rate variation, N400 amplitude variation, and P300 latency variation—are used as core input variables. Then, combined with clinical indicators such as nitrous oxide usage frequency, duration of use, and withdrawal reaction intensity, demographic and clinical data, an AI feature importance analysis algorithm is used to determine the weights of each variable. Cognitive indicators account for 60% of the weight, and clinical indicators account for 40%, ensuring that the risk assessment model considers both cognitive mechanisms and actual usage.

[0046] The second step involved using individual cognitive index data from 120 research subjects as the initial training dataset. Then, based on the initial training dataset, 10,000 sets of simulated data were generated through random sampling using an MCM Markov chain. Finally, an AI fitting algorithm was used to construct the posterior probability distribution function of the degree of nitrous oxide addiction, i.e., the level of cognitive conflict, such as mild, moderate, and severe. The threshold for mild cognitive conflict was set to 0.3–0.5, the threshold for moderate cognitive conflict was set to 0.5–0.7, and the threshold for severe cognitive conflict was set to >0.7. The predicted results of the iteratively optimized model showed a greater than 85% agreement with the clinical diagnosis.

[0047] The third step involves selecting another 60 new participants and grouping them for model testing. Prediction accuracy, sensitivity, and specificity are calculated. If the accuracy is below 80%, the variable weights and MCM sampling parameters are adjusted using an AI backpropagation algorithm until the model performance stabilizes, forming an assessment model that can output quantitative risk indices such as addiction probability and severity in real time, providing a quantitative tool for clinical diagnosis. It should be noted that the MCM simulation is only used in the second step to generate simulated data to construct the posterior probability distribution function; it does not participate in weight determination. Specifically, the MCM sampling parameters refer to the proposed distribution step size and variance, the initial value of the Markov chain, and the Markov chain number. The proposed distribution indicates where the next random sample point should be.

[0048] More specifically, the weights of the differential value variables and clinical indicator variables in the model are first determined using individual cognitive and clinical indicator data from 120 individuals. Then, 10,000 sets of simulated data are generated through random sampling of the individual cognitive and clinical indicator data from the 120 individuals using an MCM Markov chain. Based on these 10,000 sets of simulated data, the addiction probability of each set of simulated data is obtained by outputting a probability distribution function constructed based on the pre-determined variable weights and algorithm structure. A posterior probability distribution function that can classify the degree of addiction is obtained through statistical fitting of the addiction probability of each set of simulated data. This posterior probability distribution function is the constructed risk assessment model, and the prediction results of the risk assessment model must have a consistency >85% with the clinical diagnosis. The model is tested with several new subjects. If the accuracy is lower than 80%, the variable weights and MCM sampling parameters are adjusted using an AI backpropagation algorithm until the model performance is stable, forming an assessment model that can output the addiction probability and degree in real time. Test indicators include prediction accuracy, sensitivity, and specificity, where accuracy is the proportion of correct predictions. Sensitivity is the ability to correctly identify addicts. Specificity is the ability to correctly identify non-addicts.

[0049] This embodiment introduces a risk assessment model optimized using the Monte Carlo method through a risk index generation module. This model can generate a massive number of simulated samples based on limited training data, thereby constructing a stable probability distribution function. The device outputs a continuous, probabilistically statistically significant quantitative risk index, which completely changes the traditional black-and-white assessment model. It can present the severity of risk in probabilistic form, achieving a refined and quantitative assessment of cognitive conflict risk, and providing a more scientific and reliable basis for subsequent decision-making.

[0050] Furthermore, when the preset theme is nitrous oxide addiction, the corresponding risk assessment model output can be used as a basis to design a tiered and precise intervention plan to achieve closed-loop management of assessment-intervention-feedback.

[0051] The first step is the tiered design of the intervention program. Specifically, it employs cognitive-behavioral intervention, which combines mindfulness and transcranial direct current stimulation with an AI reminder system. Online mindfulness and transcranial direct current stimulation anti-relapse courses correct cognitive biases regarding nitrous oxide, while the AI ​​system sends anti-relapse reminders based on the subjects' daily behavioral data, providing categorized interventions for mild, moderate, and severely addicted patients.

[0052] The second step is dynamic feedback on the intervention's effectiveness. After the intervention, the participants' addiction level is reassessed using the Stroop experiment and risk assessment model. If the predicted probability of mild addicts drops to <0.3 after the intervention, the intervention is considered effective and the participants enter the follow-up phase. If the probability of moderate or severe addicts does not decrease or increases, the reasons for the intervention failure are identified using AI attribution analysis algorithms, and the intervention plan is adjusted and continued.

[0053] The third step is the iterative optimization of the intervention program. Intervention data from all participants is collected, and AI clustering analysis is used to identify intervention sensitivities in different addiction groups. The tiered intervention strategy is continuously optimized to reduce overall intervention effectiveness and create a scalable, precise intervention system for nitrous oxide addiction.

[0054] Furthermore, the present invention also provides a portable device.

[0055] Figure 4 This is a structural block diagram of a portable device according to an embodiment of the present invention. Figure 4 As shown, the portable device 1000 integrates the aforementioned cognitive conflict risk assessment device 100. The task stimulus presentation module 10 and test data acquisition module 20 in the cognitive conflict risk assessment device 100 are implemented through the touch screen of the portable device 1000. The cognitive difference calculation module 30 and risk index generation module 40 in the cognitive conflict risk assessment device 100 are executed through the processor of the portable device 1000.

[0056] Specifically, AI-powered Stroop testing tasks can be embedded into mini-programs on 1000 portable devices such as mobile phones and tablets, enabling self-testing to be completed in 90 seconds. The risk assessment model outputs a 95% confidence interval risk value in real time, achieving large-scale rapid screening in hidden scenarios such as communities and schools, and promoting addiction prevention from clinical to "zero-level prevention." For the risk assessment model, a Monte Carlo Bayesian deep temporal network (MC-BDTN) can be constructed, sampling 50,000 times in parallel within a 0.3s window. This automatically decouples reaction time, micro-expressions, and eye-tracking noise, improving the signal-to-noise ratio by 5 times, and achieving a highly robust quantitative assessment of cognitive conflict levels with small sample sizes.

[0057] For assessments conducted via mobile app, the risk assessment model can be trained first, followed by real-time individual assessments. Specifically, for the topic of nitrous oxide addiction, 60 nitrous oxide users and 60 control subjects can be recruited from provincial drug rehabilitation centers and universities. A 90-second adaptive Stroop task is pushed to the mobile app. A vocabulary database is dynamically generated from 100,000 social media corpora using GPT-4, resulting in 120 highly stimulating cue words. Reaction time and accuracy are recorded, along with mindfulness values ​​collected simultaneously from a mindfulness ring at a sampling rate of 1kHz. Secondly, wavelet denoising and eye-tracking drift correction are performed on the original millisecond sequences, and multidimensional physiological and cognitive features are extracted for training the risk assessment model.

[0058] For the risk assessment model, a bottom layer can be constructed using a bidirectional LSTM (Long Short-Term Memory) network to capture temporal dependencies, a middle layer introducing Dropout (random deactivation), and a top layer using MCM for 50,000 parallel samplings to estimate the posterior probability distribution of the parameters. Then, for the individual being tested, the risk assessment model outputs the individual's cognitive conflict risk probability and 95% confidence interval in real time, and provides an interpretable report based on the SHAP (Shapley Sum of Explanations) value.

[0059] Specifically, for example, a Stroop test on a test subject might have 100 time points. Each time point contains multiple individual cognitive indicators, such as changes in reaction time and pupil diameter. For instance, each time point might include color judgment tests performed on two experimental groups to obtain individual cognitive indicator data for each time point. This yields a multi-dimensional time series for the test subject. A bidirectional LSTM can analyze the entire data sequence of the test subject from left to right and then from right to left. When analyzing the data point at t=5s, the LSTM not only analyzes the current pupil dilation but also identifies the appearance of the word "overwhelmed" at t=3s and predicts a significant key press delay in the test subject at t=7s.

[0060] Through this analysis, the model identifies cognitive conflict at t=5s. The model integrates these signals scattered across different time points into a highly condensed feature vector representing the level of cognitive conflict in the tested individual. This vector could be an array of numbers, such as [0.85, 1.45, 0.12, -0.89, ...]. Since multiple time points out of 100 may be identified as having cognitive conflict, the average value of each dimension across all time points with cognitive conflict, or the value with the largest amplitude, can be used as the final dimension value within the feature vector. A quantitative risk index is then calculated based on the final dimension values ​​within the feature vector. For example, 0.85 could indicate the intensity of the N400 wave induced by nitrous oxide, 1.45 could indicate the reaction time delay caused by cognitive conflict, 0.12 could indicate the correlation between pupil dilation and craving, and -0.89 could indicate the degree of inhibition of prefrontal cortex control.

[0061] In this embodiment, during model training, Dropout randomly shuts down a portion of neurons in the network to prevent overfitting and enhance generalization ability.

[0062] Furthermore, MCM performs 50,000 rapid simulations in an extremely short time (0.3 seconds). Each simulation contains two random operations: First, for example, perturb the model parameters by randomly sampling a set of possible parameter values ​​near the predicted values ​​of the model parameters. For instance, if the model assumes that the weight of reaction time delay is 0.6, the first simulation might sample a value of 0.59, the second a value of 0.61, and so on.

[0063] Second, for example, noise is added to the data of the individuals being tested. That is, considering that the data itself has measurement errors, a little random noise is also added to the original data.

[0064] Furthermore, the calculation is repeated, that is, for each set of perturbed parameters and noisy data, a forward propagation calculation is performed to obtain a temporary addiction tendency score, and finally the statistical results are used to obtain a quantitative risk index, a number between 0 and 1.

[0065] For example, the first simulation result is a score of 0.68; the second simulation result is a score of 0.72; ...; the 50,000th simulation result is a score of 0.65. Constructing a probability distribution by plotting these 50,000 scores as a histogram will form a probability distribution. This distribution represents the best estimate of the individual's addiction risk after considering all uncertainties. Finally, according to the probability distribution, if the score is > 0.7 in 36,500 out of 50,000 simulations, it indicates that the individual has a high risk probability of 36,500 / 50,000 = 73%, meaning there is a 73% probability of high cognitive conflict risk, which is the quantitative risk index.

[0066] A 95% confidence interval is calculated by cutting off 2.5% from each end of the distribution, leaving 95% of the scores in the middle. For example, [0.66, 0.78] means there is a 95% confidence that the individual's true risk score falls within this range.

[0067] Furthermore, the SHAP technology analyzes which input feature, such as N400 amplitude or reaction time, contributes most to the final high-risk score across 50,000 simulations. For example, the report might show: "In this assessment, the abnormal increase in N400 amplitude was the primary factor leading to the high-risk assessment, contributing 45%."

[0068] Finally, the model is embedded into the judicial drug rehabilitation platform to achieve terminal reasoning in 3 seconds. Additionally, an 8-week mindfulness + transcranial direct current stimulation (TCD) intervention can be designed, using a mixed-effects model to assess the coupling between model prediction and efficacy, forming a closed loop of "detection-intervention-reassessment".

[0069] In summary, this invention integrates AI, the classic Stroop paradigm, and Monte Carlo Method (MCM) to transform the task of quantifying cognitive conflict risk into an interpretable multimodal learning problem. It utilizes deep learning to extract the microscopic spatiotemporal features of Stroop interference effects, and uses MCM to probabilistically extrapolate individual impulses and cravings, achieving an integrated assessment of cognition, behavior, and brain networks. This invention can output a quantitative risk index and intervention targets within 30 seconds, providing a low-cost, transferable screening tool for clinical use.

[0070] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0071] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A cognitive conflict risk assessment device, characterized in that, include: The task stimulus presentation module is used to provide the test subject with a Stroop test task, which is to respond to semantic stimulus units that are related to and unrelated to a preset topic and neutral semantic stimulus units. The test data acquisition module is connected to the task stimulus presentation module. The test data acquisition module is used to collect the behavioral response signals and physiological response signals generated by the individual under test when completing the Stroop test task. A cognitive difference calculation module is connected to the test data acquisition module. The cognitive difference calculation module is used to calculate the response difference value of the test individual to semantic stimulus units related to the preset topic and unrelated neutral semantic stimulus units based on the behavioral response signal and physiological response signal, respectively. A risk index generation module is connected to the cognitive difference calculation module. The risk index generation module is used to input the response difference value into a risk assessment model optimized by the Monte Carlo method, and generate and output a quantitative risk index that can represent the cognitive conflict level of the individual being tested.

2. The apparatus as claimed in claim 1, characterized in that, The task stimulus presentation module is specifically used for: The system retrieves semantic stimuli that are related to or unrelated to the preset topic, as well as neutral semantic stimuli, from a preset semantic stimulus library, and matches each semantic stimulus with an interference color to form a corresponding semantic stimulus unit.

3. The apparatus as described in claim 2, characterized in that, It also includes a semantic stimulus library construction module, connected to the task stimulus presentation module. This semantic stimulus library construction module is used to construct the preset semantic stimulus library. Specifically, the semantic stimulus library construction module is used for: A core seed lexicon should be established in advance; A corpus-grabbing model is used to extract corpus related to the core words in the core seed lexicon from the network; The hidden language extraction model is used to extract hidden language related to the preset topic from the captured corpus, and the related perceptual words are extracted from the hidden language corpus. The core words, cryptic terms, and perceptual words are mixed and sorted using TF-RF scoring. The words with preset values ​​before scoring are selected to construct the preset semantic stimulus library.

4. The apparatus as claimed in claim 1, characterized in that, The test data acquisition module includes: The behavioral data acquisition unit is used to collect the reaction time and error rate of the individual under test when completing the task of judging the color of the semantic stimulus unit as the behavioral response signal; The physiological data acquisition unit operates synchronously with the behavioral data acquisition unit. The physiological data acquisition unit is used to collect pupil diameter change data and electroencephalogram (EEG) signals of the individual under test as the physiological response signals.

5. The apparatus as described in claim 4, characterized in that, After acquiring the electroencephalogram (EEG) signal, the physiological data acquisition unit is also used to extract the P300 wave latency and N400 wave amplitude from the EEG signal.

6. The apparatus as claimed in claim 1, characterized in that, The cognitive difference calculation module is specifically used for: The average reaction time, error rate, average P300 latency, and average N400 amplitude of the test subjects for semantic stimulus units related to the preset topic and for neutral semantic stimulus units unrelated to the preset topic were calculated respectively. The differences between the average reaction time, error rate, average P300 latency, and average N400 amplitude were also calculated to obtain the difference values ​​of reaction time, error rate, P300 latency, and N400 amplitude.

7. The apparatus as claimed in claim 1, characterized in that, The risk assessment model is trained in the following way: Obtain an initial training dataset, which includes multiple difference values ​​from multiple testers. data; Based on the initial training dataset, random sampling is performed using the Markov chain Monte Carlo method to generate a simulated dataset. The initial risk assessment model is fitted based on the simulated dataset to construct a posterior probability distribution function that can output a quantitative risk index, thereby training the risk assessment model.

8. The apparatus as claimed in claim 7, characterized in that, The risk index generation module, when generating a quantitative risk index, is specifically used for: Based on the posterior probability distribution function, the model output value is mapped to a preset probability interval to generate the quantitative risk index.

9. The apparatus as claimed in claim 1, characterized in that, It also includes a model optimization module, connected to the risk index generation module, the model optimization module being used for: The system receives verification feedback on the quantified risk index and adjusts the parameters of the risk assessment model and the sampling parameters of the Monte Carlo method using a backpropagation algorithm.

10. A portable device, characterized in that, The device integrates a cognitive conflict risk assessment device as described in any one of claims 1 to 9, wherein the task stimulus presentation module and test data acquisition module of the cognitive conflict risk assessment device are implemented through the touch screen of the portable device, and the cognitive difference calculation module and risk index generation module of the cognitive conflict risk assessment device are executed through the processor of the portable device.

Images