AI-based mental health assessment system and method

Abstract

The invention discloses an AI-based mental health assessment system and method, and the method comprises the steps: authenticating and inputting a first wearable device worn by a user, and obtaining first physiological data, collected by the first wearable device worn by the user, of the user for N continuous days in recent time; obtaining first answer data filled by the user on the electronic terminal for the mental health assessment questionnaire; in the questionnaire answering period, second physiological data, collected by the wearable device in real time, of the user are obtained, and real-time data features are generated; comparing real-time data characteristics generated by the first physiological data with real-time data characteristics generated by the second physiological data, and judging whether the first physiological data and the second physiological data belong to the same user or not; if not, stopping evaluation and giving an alarm; if yes, inputting the following data into the AI evaluation model: the first answer data and the first physiological data, and outputting a psychological health state evaluation result and personalized intervention suggestions of the user. The psychological health assessment accuracy can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of mental health assessment, and in particular to an AI-based mental health assessment system and method. Background Art

[0002] Mental health assessment is a systematic tool and method that can be used to assess the user's mental state. Questionnaires are a common means of mental health assessment, but questionnaire adjustments to assess mental health generally only stay at the score of the questionnaire answers, which is prone to subjective bias and cheating by users (the actual respondents do not match the assessment goals), resulting in inaccurate assessments. The following are several common subjective biases: 1. Individuals may tend to conceal their true feelings and give answers that meet social expectations (such as denying depression or exaggerating positive performance). 2 Questionnaires usually require a review of mental states over a period of time (such as "the past two weeks"), but memories may be inaccurate or affected by recent emotions. 3. Some people lack the ability to accurately perceive their own emotions, resulting in distorted answers. Therefore, the accuracy of questionnaire surveys in assessing mental health is not high. Summary of the invention

[0003] The applicant has found through research that in questionnaire mental health assessment, the user's potential mental state can be captured through behavioral parameters (such as mouse trajectory, answering time, number of option modifications). If the psychological state expressed by the behavioral parameters is combined with the psychological state expressed by the questionnaire answers, the accuracy of the mental health assessment can be effectively improved.

[0004] In view of the above-mentioned defects of the prior art, the technical problem to be solved by the present invention is to provide an AI-based mental health assessment system and method, aiming to improve the accuracy of mental health assessment.

[0005] To achieve the above objectives, the first aspect of the present invention discloses a mental health assessment method based on AI, the method comprising: Step S1, authenticating and entering a first wearable device worn by a user, and obtaining first physiological data of the user collected by the first wearable device worn by the user for N consecutive days in the recent period; the first physiological data includes: heart rate data, blood pressure data, sleep quality data, and galvanic skin response data; Step S2, obtaining first answer data filled in by the user on the electronic terminal for the mental health assessment questionnaire; and during the questionnaire answering period, obtaining the second physiological data of the user collected in real time by the wearable device and generating real-time data features; Step S3, comparing the real-time data features generated by the first physiological data and the second physiological data to determine whether the first physiological data and the second physiological data are from the same user; if the two are not from the same user, terminating the evaluation and giving an alarm; Step S4: In response to the first physiological data and the second physiological data belonging to the same user, input the following data into the AI evaluation model: the first answering data and the first physiological data, and output the evaluation result of the user's mental health status and personalized intervention suggestions.

[0006] Optionally, step S3 includes: Step S31: Preprocess and extract features from the first physiological data to generate a first multi-dimensional feature vector; Among them, the first multi-dimensional feature vector includes a combination of at least two of the following types of features: Resting heart rate, time-domain / frequency-domain indicators of heart rate variability (HRV); Blood pressure trend feature and postural change response feature; the postural change response feature is generated by combining the triaxial acceleration data of the first wearable device with the blood pressure data; Skin conductance response amplitude, rise / fall time, and basal skin conductance level; Step S32: Input the first multi-dimensional feature vector into a machine learning model to train and generate a personalized physiological baseline model corresponding to the user; Step S33: Extract a second feature vector with the same dimension as the first multi-dimensional feature vector from the second physiological data during the questionnaire answering period; Step S34: Calculate the matching degree score between the second feature vector and the personalized physiological baseline model; Step S35: When the matching degree score exceeds a preset threshold, determine that the currently wearing user is the same person as the registered user.

[0007] Optionally, step S31 includes: Extract the mean resting heart rate and the heart rate recovery slope after exercise from the heart rate data; Extract the ratio of low-frequency power (LF) to high-frequency power (HF) (LF / HF) and the root mean square of the differences between adjacent R-R intervals (RMSSD) from the heart rate variability analysis; Combine the acceleration data to identify specific behavioral events, and extract the EDA response peak and blood pressure fluctuation amplitude within a preset time window before and after the event; The personalized physiological baseline model in step S32 is constructed using any of the following algorithms: Hidden Markov model (HMM), used to model the temporal transition probability of physiological states; Support vector data description (SVDD), used to establish the high-dimensional hypersphere boundary of user physiological characteristics; Long short-term memory network (LSTM), used to learn the multi-parameter joint temporal dependence pattern; The matching degree calculation in step S34 adopts one of the following methods: Calculate the dynamic time warping (DTW) distance between the new feature vector and the feature vector of the baseline model; Output the probability value belonging to the registered user through a pre-trained binary classifier; Measure the deviation degree of the new data relative to the distribution of the baseline model based on the Mahalanobis Distance.

[0008] Optionally, step S2 further includes: Step S21: Collect the answer content data of the first user to the mental health assessment questionnaire and the answering behavior of the first user during the answering process; wherein, the answering behavior at least includes the mouse track, the answering duration, and the number of option modifications; Step S22: Obtain the content score of the first user according to the answer content data; wherein, the content score is positively or negatively correlated with the mental health assessment of the first user; Step S23: Obtain the mouse track complexity and the proportion of invalid moves of the first user according to the mouse track; obtain the cognitive load number of the first user according to the answering duration; obtain the decision conflict degree according to the number of option modifications; obtain the attention dispersion degree of the first user according to the mouse track, the answering duration, and the number of option modifications; Step S24: Obtain the behavior score of the first user according to the mouse track complexity, the proportion of invalid moves, the cognitive load number, the decision conflict degree, and the attention dispersion degree; wherein, the behavior score is positively or negatively correlated with the mental health assessment of the first user; Step S25: Obtain the consistency score of the first user according to the content score and the behavior score; wherein, the scores of the content score and the behavior score are in the same direction as the mental health assessment of the first user; Step S26: Obtain the mental health assessment result of the first user according to the content score, the behavior score, and the consistency score.

[0009] Optionally, step S24 includes: According to Obtain the behavior score of the first user; wherein, is the behavior score, is 4, is the mouse track complexity, is the weight of the mouse track complexity, is the proportion of invalid moves, is the weight of the invalid movement ratio, is the cognitive load number, is the weight of the cognitive load number, is the decision-making conflict degree, is the weight of the decision-making conflict degree, is the attention dispersion degree, is the weight of the attention dispersion degree.

[0010] Optionally, step S26 includes According to obtain the mental health assessment result; where is the score corresponding to the mental health assessment result, is the content score, is the weight of the content score, is the behavior score, is the weight of the behavior score, is the consistency score, is the weight of the consistency score.

[0011] Optionally, the method further includes According to the specific values of the content score, the behavior score, and the consistency score, determine the weights corresponding to the content score, the behavior score, and the consistency score; where the weight corresponding to the consistency score is positively correlated with the content score and the behavior score.

[0012] Optionally, after step S25, the method further includes: In response to the consistency score being less than the first consistency threshold, determine that the first user has cheated; where the cheating behavior includes concealing the user's own mental health condition.

[0013] Optionally, after step S21, the method further includes: In response to the answering time being less than the minimum answering time threshold, determine that the first user has answered randomly.

[0014] Optionally, in step S23, obtaining the cognitive load number of the first user according to the answering time includes: Obtain the reading speed of the first user; According to the reading speed of the first user and the answering time, obtain the cognitive load number of the first user.

[0015] The second aspect of the present invention discloses a mental health assessment system based on AI, which includes: a first physiological data acquisition unit, a second physiological data acquisition unit, a user comparison unit, and an assessment unit; The first physiological data acquisition unit is used to authenticate and input the first wearable device worn by the user, and acquire the first physiological data of the user collected by the first wearable device worn by the user in the recent consecutive N days; the first physiological data includes: heart rate data, blood pressure data, sleep quality data, and galvanic skin response data; The second physiological data acquisition unit is used to acquire the first answer data filled in by the user on the electronic terminal for the mental health assessment questionnaire; and during the questionnaire answering period, acquire the second physiological data of the user collected in real time by the wearable device and generate real-time data features; The user comparison unit is used to compare the real-time data features generated by the first physiological data and the second physiological data, and determine whether the first physiological data and the second physiological data belong to the same user; if the two are not the same user, terminate the assessment and alarm; The assessment unit is used to, in response to the first physiological data and the second physiological data belonging to the same user, input the following data into the AI assessment model: the first answer data, the first physiological data, and output the mental health status assessment result and personalized intervention suggestions of the user.

[0016] Optionally, the user comparison unit includes: a first multi-dimensional feature vector generation sub-unit, a model training generation sub-unit, a second multi-dimensional feature vector extraction sub-unit, a matching score calculation sub-unit, and a judgment sub-unit; The first multi-dimensional feature vector generation sub-unit is used to preprocess and extract features from the first physiological data to generate a first multi-dimensional feature vector; Among them, the first multi-dimensional feature vector includes a combination of at least two of the following features: Resting heart rate, time domain / frequency domain indexes of heart rate variability (HRV); Blood pressure trend characteristics and orthostatic change response characteristics; the orthostatic change response characteristics are generated by combining the three-axis acceleration data of the first wearable device with the blood pressure data; Galvanic skin response amplitude, rise / fall time, and basal galvanic skin level; The model training generation sub-unit is used to input the first multi-dimensional feature vector into a machine learning model and train to generate a personalized physiological baseline model corresponding to the user; The second multi-dimensional feature vector extraction sub-unit is used to extract a second feature vector with the same dimension as the first multi-dimensional feature vector from the second physiological data during the questionnaire answering period; The matching score calculation subunit is configured to calculate the matching degree score between the second feature vector and the personalized physiological baseline model; The determination subunit is configured to determine that the currently wearing user and the registered user are the same person when the matching degree score exceeds a preset threshold.

[0017] Optionally, the first multi-dimensional feature vector generation subunit is specifically configured to: Extract the mean resting heart rate and the post-exercise heart rate recovery slope from the heart rate data; Extract the ratio of low-frequency power (LF) to high-frequency power (HF) (LF / HF) and the root mean square of the difference between adjacent R-R intervals (RMSSD) from the heart rate variability analysis; Identify specific behavioral events in combination with the acceleration data, and extract the EDA response peak and blood pressure fluctuation amplitude within a preset time window before and after the event; The personalized physiological baseline model in the model training and generation subunit is constructed using any of the following algorithms: Hidden Markov Model (HMM), which is used to model the temporal transition probability of physiological states; Support Vector Data Description (SVDD), which is used to establish the high-dimensional hypersphere boundary of user physiological characteristics; Long Short-Term Memory Network (LSTM), which is used to learn the multi-parameter joint temporal dependence pattern; The matching degree calculation of the matching score calculation subunit adopts one of the following methods: Calculate the Dynamic Time Warping (DTW) distance between the new feature vector and the baseline model feature vector; Output the probability value belonging to the registered user through a pre-trained binary classifier; Measure the deviation degree of the new data relative to the baseline model distribution based on the Mahalanobis Distance.

[0018] Optionally, the system further includes: a data collection module, a content score obtaining module, a behavior parameter obtaining module, a behavior score obtaining module, a consistency score obtaining module, and an evaluation module; The data collection module is configured to collect the answer content data of the first user to the mental health assessment questionnaire and the answering behavior of the first user during the answering process; wherein, the answering behavior at least includes the mouse trajectory, the answering duration, and the number of option modifications; The content score obtaining module is configured to obtain the content score of the first user according to the answer content data; wherein, the content score is positively or negatively correlated with the mental health assessment of the first user; The behavior parameter acquisition module is configured to obtain the complexity of the mouse trajectory and the proportion of invalid movements of the first user according to the mouse trajectory; obtain the cognitive load number of the first user according to the answering duration; obtain the decision-making conflict degree according to the number of option modifications; and obtain the attention dispersion degree of the first user according to the mouse trajectory, the answering duration, and the number of option modifications. The behavior score acquisition module is configured to obtain the behavior score of the first user according to the complexity of the mouse trajectory, the proportion of invalid movements, the cognitive load number, the decision-making conflict degree, and the attention dispersion degree; wherein, the behavior score is positively or negatively correlated with the mental health assessment of the first user. The consistency score acquisition module is configured to obtain the consistency score of the first user according to the content score and the behavior score; wherein, the scores of the content score and the behavior score are in the same direction as the mental health assessment of the first user. The evaluation module is configured to obtain the mental health assessment result of the first user according to the content score, the behavior score, and the consistency score.

[0019] Optionally, the behavior score acquisition module is specifically configured to: According to Obtain the behavior score of the first user; wherein, is the behavior score, is 4, is the complexity of the mouse trajectory, is the weight of the complexity of the mouse trajectory, is the proportion of invalid movements, is the weight of the proportion of invalid movements, is the cognitive load number, is the weight of the cognitive load number, is the decision-making conflict degree, is the weight of the decision-making conflict degree, is the attention dispersion degree, is the weight of the attention dispersion degree.

[0020] Optionally, the evaluation module is specifically configured to: According to Obtain the mental health assessment result; wherein, is the score corresponding to the mental health assessment result, is the content score, is the weight of the content score, is the behavior score, is the weight of the behavior score, is the consistency score, is the weight of the consistency score.

[0021] Optionally, the system further includes: a first weight assignment module, The first weight assignment module is configured to determine the weights corresponding to the content score, the behavior score, and the consistency score according to the specific values of the content score, the behavior score, and the consistency score; wherein, the weight corresponding to the consistency score is positively correlated with the content score and the behavior score.

[0022] Optionally, the system further includes: a cheating determination module, The cheating determination module is configured to determine that the first user has a cheating behavior in response to the consistency score being less than a first consistency threshold; wherein the cheating behavior includes concealing one's own mental health condition.

[0023] Optionally, the system further includes: an answering behavior determination module, The answering behavior determination module is configured to determine that the first user has a random answering behavior in response to the answering duration being less than a minimum answering duration threshold.

[0024] Advantages of the present invention: 1. By using the first physiological data collected by the first wearable device during normal times and the second physiological data during answering questions, the present invention can determine whether the user is the same person. This can effectively avoid cheating or incorrect wearing, and effectively improve the accuracy of evaluation. 2. The present invention effectively combines the user's normal physiological data and answering data. Based on the appearance, it can capture the user's potential psychological state for comprehensive evaluation, improving the accuracy of evaluation. 3. The present invention collects the answering behavior of the user during the answering process and analyzes the answering behavior. The present invention captures the user's potential psychological state through the answering behavior, evaluates the user's psychological state, and obtains a behavior score; then, based on the behavior score and the content score, a consistency judgment is made to obtain a consistency score. The present invention effectively combines the content score, the behavior score, and the consistency score to comprehensively evaluate the user's mental health, avoiding the one-sidedness of a single content score evaluation, effectively reducing subjective deviation, and improving the accuracy of mental health evaluation. 4. The consistency score in the present invention evaluates the matching degree between the content score and the behavior score, avoiding excessive deviation in the comprehensive evaluation caused by the excessive divergence of the two, and further improving the accuracy of mental health evaluation. 5. The present invention determines that the first user has a cheating behavior in response to the consistency score being less than the first consistency threshold. When the consistency score is too small, it indicates that the user wants to conceal their mental health status. The present invention can discover it in time and report it for re-review. 6. The present invention determines that the first user has a random answering behavior in response to the answering duration being less than the minimum answering duration threshold. In this way, it avoids the behavior of users randomly answering questions perfunctorily. 7. The present invention also combines the individual's reading speed to obtain the cognitive load number, making the cognitive load number more accurate, and thus obtaining a more accurate behavior score.

[0025] In summary, the present invention improves the accuracy of mental health evaluation. Brief Description of the Drawings

[0026] Figure 1 is a schematic flowchart of a method for mental health evaluation based on AI provided by a specific embodiment of the present invention; Figure 2 is a schematic structural diagram of a system for mental health evaluation based on AI provided by a specific embodiment of the present invention; Figure 3 is a support vector description model diagram for user identification based on a wearable device of the present invention. Detailed Embodiments

[0027] The present invention discloses an AI-based mental health assessment system and method. Those skilled in the art can draw on the content of this article and appropriately improve the technical details for implementation. It should be particularly noted that all similar substitutions and modifications are obvious to those skilled in the art and are all considered to be included in the present invention. The method and application of the present invention have been described through preferred embodiments. Relevant personnel can obviously make changes or appropriate modifications and combinations to the methods and applications described herein without departing from the content, spirit, and scope of the present invention to implement and apply the technology of the present invention.

[0028] After research by the applicant, it is found that: in questionnaire-based mental health assessment, by capturing the user's potential mental state through behavioral parameters (such as mouse trajectory, answering time, number of option modifications), if the mental state expressed by the behavioral parameters can be combined with the mental state expressed by the questionnaire answers, the accuracy of mental health assessment can be effectively improved.

[0029] Therefore, an embodiment of the present invention provides an AI-based mental health assessment method, as Figures 1 - 3 shown, the method includes: Step S1, authenticate and input the first wearable device worn by the user, and obtain the first physiological data of the user collected by the first wearable device worn by the user in the recent consecutive N days.

[0030] The first physiological data includes: heart rate data, blood pressure data, sleep quality data, and galvanic skin response data.

[0031] It should be noted that the first physiological data can reflect the user's usual mental health state from the side.

[0032] Step S2, obtain the first answering data filled in by the user on the electronic terminal for the mental health assessment questionnaire; and during the questionnaire answering period, obtain the second physiological data of the user collected by the wearable device in real time and generate real-time data features.

[0033] It should be noted that the user needs to continuously wear the first wearable device so that the first wearable device can collect the first physiological data and the second physiological data.

[0034] Step S3, compare the real-time data features generated from the first physiological data and the second physiological data to determine whether the first physiological data and the second physiological data belong to the same user; if the two are not the same user, terminate the assessment and alarm.

[0035] In this specific embodiment, step S3 includes: Step S31, perform preprocessing and feature extraction on the first physiological data to generate a first multi-dimensional feature vector; Among them, the first multi-dimensional feature vector includes a combination of at least the following two types of features: Time-domain / frequency-domain indicators of resting heart rate and heart rate variability (HRV) related to heart rate data; Blood pressure trend characteristics and orthostatic change response characteristics related to blood pressure data; the orthostatic change response characteristics are generated from the triaxial acceleration data of the first wearable device combined with blood pressure data; Skin conductance response amplitude, rise / fall time, and basal skin conductance level related to skin conductance response data; Step S32: Input the multi-dimensional feature vector into a machine learning model to train and generate a personalized physiological baseline model for the corresponding user; Step S33: Extract a second feature vector with the same dimension as the first multi-dimensional feature vector from the second physiological data during questionnaire answering; Step S34: Calculate the matching degree score between the second feature vector and the personalized physiological baseline model; Step S35: When the matching degree score exceeds a preset threshold, determine that the currently wearing user is the same person as the registered user.

[0036] It is worth mentioning that, to avoid misjudgment, in this technical solution, it is also considered to combine triaxial acceleration data and target resting heart rate and heart rate variability; actually, it takes into account the state of the user before answering the questionnaire (whether in motion). Based on the data in the motion state of the first physiological data, it is determined whether the user was in motion before answering the questionnaire, effectively avoiding misjudgment.

[0037] Furthermore, step S31 includes: Extract the mean resting heart rate and the post-exercise heart rate recovery slope from the heart rate data; Extract the ratio of low-frequency power (LF) to high-frequency power (HF) (LF / HF) and the root mean square of the differences between adjacent R-R intervals (RMSSD) from heart rate variability analysis; Combine acceleration data to identify specific behavioral events, and extract the EDA response peak and blood pressure fluctuation amplitude within a preset time window before and after the event; The personalized physiological baseline model in step S32 is constructed using any of the following algorithms: Hidden Markov Model (HMM), used to model the temporal transition probability of physiological states; Support Vector Data Description (SVDD), used to establish the high-dimensional hypersphere boundary of user physiological characteristics; Long Short-Term Memory Network (LSTM), used to learn the multi-parameter joint temporal dependence pattern; The matching degree calculation in step S34 is performed using one of the following methods: Calculate the Dynamic Time Warping (DTW) distance between the new feature vector and the feature vector of the baseline model; Output the probability value of belonging to the registered user through the pre-trained binary classifier; The Mahalanobis distance is used to measure the deviation of new data from the baseline model distribution.

[0038] In addition, the personalized physiological baseline model also includes a model adaptive update mechanism: When M consecutive verifications are successful, the new feature vector is merged into the original baseline model according to the decay weight; When the match score is below the warning threshold but above the rejection threshold, a short period of high-frequency data acquisition is triggered to recalibrate the model.

[0039] It should be noted that artificial intelligence can be used to accurately judge users and effectively avoid cheating.

[0040] Step S4, in response to the first physiological data and the second physiological data being of the same user, input the following data into the AI evaluation model: the first answer data, the first physiological data, and output the user's mental health status evaluation results and personalized intervention suggestions.

[0041] In this specific embodiment, step S2 also includes: Step S21: Collect the first user's answer content data to the mental health assessment questionnaire and the first user's answering behavior during the answering process.

[0042] Among them, the answering behavior at least includes the mouse trajectory, answering time and the number of times the option is modified.

[0043] It should be noted that when users are anxious or inattentive, the complexity of the mouse trajectory and the proportion of invalid movements will increase. When users have psychological problems, they may have delayed thinking, which will significantly prolong the time it takes to answer questions, or when they deliberately avoid questions due to psychological factors, the time it takes to answer questions related to negative emotions will be significantly prolonged. When the user group lacks the ability to accurately perceive their own emotions, options may be modified multiple times. Therefore, the user's mental health status can be captured by analyzing these behaviors.

[0044] In this specific embodiment, after step S21, the method further includes: In response to the answering time being less than the minimum answering time threshold, it is determined that the first user has made random answers.

[0045] It should be noted that the questions in the questionnaire generally require reading before answering, but some users will randomly choose in order to quickly get through the questions, which will cause the content score to be too distorted. Therefore, this embodiment can effectively prevent the occurrence of such a situation by judging whether the user has such behavior based on the length of time spent answering the questions.

[0046] Step S22: Obtain the content score of the first user based on the answer content data.

[0047] Among them, the content score is positively or negatively correlated with the mental health assessment of the first user.

[0048] It should be noted that depending on the content of the questionnaire, for some questionnaires, the higher the score, the healthier the psychology, while for some questionnaires, the higher the score, the less healthy the psychology. It needs to be determined according to the actual situation. However, in any case, there is a monotonic relationship between the score and the mental health status.

[0049] Step S23: Obtain the mouse trajectory complexity and the proportion of invalid movements of the first user according to the mouse trajectory; obtain the cognitive load number of the first user according to the answering duration; obtain the decision-making conflict degree according to the number of option modifications; obtain the attention dispersion degree of the first user according to the mouse trajectory, the answering duration, and the number of option modifications.

[0050] It should be noted that when the answering duration is longer, it indicates that the user's cognitive ability is declining (possibly caused by psychological factors), and thus the cognitive load number is larger. The cognitive load number is defined by specific values. The more the number of option modifications, the greater the decision-making conflict degree, and the decision-making conflict degree is defined by specific values. And the above various behaviors may all be caused by attention. When the attention is more scattered, the answering duration, the mouse trajectory, and the number of option modifications will all increase, and psychological problems are also likely to cause more scattered attention.

[0051] In this specific embodiment, obtaining the mouse trajectory complexity and the proportion of invalid movements of the first user according to the mouse trajectory includes: Obtain the mouse trajectory complexity of the first user according to the mouse trajectory; Determine the effective movement distance according to the initial position and the final position of the mouse; obtain the total movement distance according to the mouse trajectory; obtain the proportion of invalid movements according to the effective movement distance and the total movement distance.

[0052] It should be noted that when the user has problems such as anxiety, impatience, and inattention, the mouse trajectory complexity and the proportion of invalid movements are likely to increase. In the embodiment of the present invention, the mouse trajectory complexity and the effective movement distance are defined by specific values.

[0053] In this specific embodiment, in step S23, obtaining the cognitive load number of the first user according to the answering duration includes: Obtain the reading speed of the first user; Obtain the cognitive load number of the first user according to the reading speed and the answering duration of the first user.

[0054] It should be noted that different users have different reading speeds. Therefore, by collecting the reading speed and comprehensively obtaining the cognitive load number based on the reading speed, the cognitive load number can be made more accurate.

[0055] In this specific embodiment, simple common sense questions can be set before the questionnaire for the user to answer in order to collect the user's reading speed.

[0056] Step S24: Obtain the behavior score of the first user according to the mouse trajectory complexity, the proportion of invalid movements, the cognitive load number, the decision-making conflict degree, and the attention dispersion degree.

[0057] Among them, the behavior score is positively or negatively correlated with the mental health assessment of the first user.

[0058] It should be noted that the behavior score and the content score are the same. In any case, there is a monotonic relationship between the score and the mental health condition.

[0059] Step S25: Obtain the consistency score of the first user according to the content score and the behavior score.

[0060] Among them, the scores of the content score and the behavior score are positively correlated with the mental health assessment of the first user.

[0061] In this specific embodiment, after step S25, the method further includes: In response to the consistency score being less than the first consistency threshold, it is determined that the first user has cheated; where the cheating behavior includes concealing one's own mental health condition.

[0062] It should be noted that when the consistency is too small, it indicates that the first user has a problem of concealing their own mental health condition, and it can be discovered in time and a pre-plan can be made.

[0063] It should be noted that the positive correlation means that when the content score is higher, the user's mental health is better, and when the behavior score is higher, the user's mental health is better. Or when the content score is higher, the user's mental health is worse, and when the behavior score is higher, the user's mental health is worse.

[0064] Step S26: Obtain the mental health assessment result of the first user according to the content score, the behavior score, and the consistency score.

[0065] In the first specific embodiment, step S24 includes: According to Obtain the behavior score of the first user; where is the behavior score, is 4, is the mouse trajectory complexity, is the weight of the mouse trajectory complexity, is the proportion of invalid moves, is the weight of the proportion of invalid moves, is the cognitive load number, is the weight of the cognitive load number, is the decision conflict degree, is the weight of the decision conflict degree, is the attention dispersion degree, is the weight of the attention dispersion degree.

[0066] In the first specific embodiment, step S26 includes According to obtain a mental health assessment result; wherein, is the score corresponding to the mental health assessment result, is the content score, is the weight of the content score, is the behavior score, is the weight of the behavior score, is the consistency score, is the weight of the consistency score.

[0067] In the first specific embodiment, the method further includes According to the specific scores of the content score, behavior score and consistency score, determine the weights corresponding to the content score, behavior score and consistency score; wherein, the weight corresponding to the consistency score is positively correlated with the content score and behavior score.

[0068] In the embodiment of the present invention, the first physiological data collected by the first wearable device during normal times and the second physiological data during answering questions can be used to determine whether the user is the same person. This can effectively avoid cheating or incorrect wearing, and effectively improve the evaluation accuracy.

[0069] Through the above embodiments, it is found that: the embodiments of the present invention effectively combine the user's normal physiological data and answering data, and based on the appearance, can capture the user's potential mental state for comprehensive evaluation, improving the evaluation accuracy.

[0070] Embodiments of the present invention collect the answering behaviors of users during the answering process and analyze the answering behaviors. Embodiments of the present invention capture the potential psychological state of users through answering behaviors, evaluate the psychological state of users, and obtain behavior scores; then, according to the behavior scores and content scores, consistency judgment is performed to obtain consistency scores. Embodiments of the present invention effectively combine content scores, behavior scores, and consistency scores to comprehensively evaluate the mental health of users, avoiding the one-sidedness of single content score evaluation, effectively reducing subjective bias, and improving the accuracy of mental health evaluation.

[0071] The consistency score in the embodiments of the present invention evaluates the matching degree between the content score and the behavior score, avoiding the excessive deviation of the comprehensive evaluation caused by the excessive deviation of the two, and further improving the accuracy of mental health evaluation.

[0072] Embodiments of the present invention determine that the first user has cheating behavior in response to the consistency score being less than the first consistency threshold. When the consistency score is too small, it indicates that the user wants to conceal their mental health status. Embodiments of the present invention can discover it in time and report it for review.

[0073] Embodiments of the present invention determine that the first user has random answering behavior in response to the answering duration being less than the minimum answering duration threshold. In this way, the behavior of users answering randomly perfunctorily is avoided.

[0074] Embodiments of the present invention also combine the reading speed of individuals to obtain the cognitive load number, making the cognitive load number more accurate, and then obtaining a more accurate behavior score.

[0075] In summary, embodiments of the present invention improve the accuracy of mental health evaluation.

[0076] Based on the above mental health evaluation method, embodiments of the present invention also provide an AI-based mental health evaluation system, as Figure 2 shown, the system includes: a first physiological data acquisition unit 201, a second physiological data acquisition unit 202, a user comparison unit 203, and an evaluation unit 204; The first physiological data acquisition unit 201 is used to authenticate and input the first wearable device worn by the user, and acquire the first physiological data of the user collected by the first wearable device worn by the user in the recent continuous N days; the first physiological data includes: heart rate data, blood pressure data, sleep quality data, galvanic skin response data; The second physiological data acquisition unit 202 is used to acquire the first answering data filled in by the user on the electronic terminal for the mental health evaluation questionnaire; and during the questionnaire answering period, acquire the second physiological data of the user collected by the wearable device in real time and generate real-time data features; A user comparison unit 203 is configured to compare the real-time data features generated from the first physiological data and the second physiological data to determine whether the first physiological data and the second physiological data belong to the same user. If they do not belong to the same user, the evaluation is terminated and an alarm is issued. An evaluation unit 204 is configured to, in response to the first physiological data and the second physiological data belonging to the same user, input the following data into the AI evaluation model: the first answering data and the first physiological data, and output an evaluation result of the user's mental health status and personalized intervention suggestions.

[0077] Optionally, the user comparison unit 203 includes: a first multi-dimensional feature vector generation sub-unit, a model training generation sub-unit, a second multi-dimensional feature vector extraction sub-unit, a matching score calculation sub-unit, and a determination sub-unit. The first multi-dimensional feature vector generation sub-unit is configured to perform preprocessing and feature extraction on the first physiological data to generate a first multi-dimensional feature vector. Wherein, the first multi-dimensional feature vector includes a combination of at least two of the following types of features: Resting heart rate related to heart rate data, time domain / frequency domain indexes of heart rate variability (HRV); Blood pressure trend features and postural change response features related to blood pressure data; the postural change response features are generated by combining the three-axis acceleration data of the first wearable device with the blood pressure data. Skin conductance response amplitude, rise / fall time, and basal skin conductance level related to skin conductance response data; The model training generation sub-unit is configured to input the multi-dimensional feature vector into a machine learning model to train and generate a personalized physiological baseline model for the corresponding user. The second multi-dimensional feature vector extraction sub-unit is configured to extract a second feature vector with the same dimension as the first multi-dimensional feature vector from the second physiological data during questionnaire answering. The matching score calculation sub-unit is configured to calculate the matching degree score between the second feature vector and the personalized physiological baseline model. The determination sub-unit is configured to determine that the currently wearing user and the registered user are the same person when the matching degree score exceeds a preset threshold.

[0078] Optionally, the first multi-dimensional feature vector generation sub-unit is specifically configured to: Extract the mean resting heart rate and the heart rate recovery slope after exercise from the heart rate data; Extract the ratio of low-frequency power (LF) to high-frequency power (HF) (LF / HF) and the root mean square of the differences between adjacent R-R intervals (RMSSD) from heart rate variability analysis; Identify specific behavioral events in combination with acceleration data, and extract the EDA response peak and blood pressure fluctuation amplitude within a preset time window before and after the event; The personalized physiological baseline model in the model training and generation subunit is constructed using any of the following algorithms: Hidden Markov Model (HMM), which is used to model the temporal transition probability of physiological states; Support Vector Data Description (SVDD), which is used to establish the boundary of a high-dimensional hypersphere for the user's physiological characteristics; Long Short-Term Memory Network (LSTM), which is used to learn the multi-parameter joint temporal dependence pattern; The matching degree calculation in the matching score calculation subunit adopts one of the following methods: Calculate the Dynamic Time Warping (DTW) distance between the new feature vector and the feature vector of the baseline model; Output the probability value belonging to the registered user through a pre-trained binary classifier; Based on the Mahalanobis Distance, measure the deviation of the new data from the distribution of the baseline model.

[0079] Optionally, the system further includes: a data collection module, a content score obtaining module, a behavior parameter obtaining module, a behavior score obtaining module, a consistency score obtaining module, and an evaluation module; The data collection module is used to collect the answer content data of the first user to the mental health assessment questionnaire and the answering behavior of the first user during the answering process; wherein, the answering behavior at least includes the mouse trajectory, the answering duration, and the number of option modifications; The content score obtaining module is used to obtain the content score of the first user according to the answer content data; wherein, the content score is positively or negatively correlated with the mental health assessment of the first user; The behavior parameter obtaining module is used to obtain the mouse trajectory complexity and the proportion of invalid movements of the first user according to the mouse trajectory; obtain the cognitive load number of the first user according to the answering duration; obtain the decision conflict degree according to the number of option modifications; obtain the attention dispersion degree of the first user according to the mouse trajectory, the answering duration, and the number of option modifications; The behavior score obtaining module is used to obtain the behavior score of the first user according to the mouse trajectory complexity, the proportion of invalid movements, the cognitive load number, the decision conflict degree, and the attention dispersion degree; wherein, the behavior score is positively or negatively correlated with the mental health assessment of the first user; The consistency score obtaining module is used to obtain the consistency score of the first user according to the content score and the behavior score; wherein, the scores of the content score and the behavior score are in the same direction as the mental health assessment of the first user; The evaluation module is used to obtain the mental health assessment result of the first user according to the content score, the behavior score, and the consistency score.

[0080] Optionally, the behavior score obtaining module is specifically used for: According to obtain the behavior score of the first user; wherein, is the behavior score, is 4, is the mouse trajectory complexity, is the weight of the mouse trajectory complexity, is the proportion of invalid moves, is the weight of the proportion of invalid moves, is the cognitive load number, is the weight of the cognitive load number, is the decision-making conflict degree, is the weight of the decision-making conflict degree, is the attention dispersion degree, is the weight of the attention dispersion degree.

[0081] Optionally, the evaluation module is specifically configured to: according to obtain the mental health evaluation result; wherein, is the score corresponding to the mental health evaluation result, is the content score, is the weight of the content score, is the behavior score, is the weight of the behavior score, is the consistency score, is the weight of the consistency score.

[0082] Optionally, the system further includes: a first weight allocation module, The first weight allocation module is used to determine the weights corresponding to the content score, the behavior score, and the consistency score according to the specific values of the content score, the behavior score, and the consistency score; wherein, the weight corresponding to the consistency score is positively correlated with the content score and the behavior score.

[0083] Optionally, the system further includes: a cheating judgment module, The cheating judgment module is used to judge that the first user has a cheating behavior in response to the consistency score being less than the first consistency threshold; wherein the cheating behavior includes concealing one's own mental health condition.

[0084] Optionally, the system further includes: an answering behavior judgment module, The answering behavior judgment module is used to judge that the first user has a random answering behavior in response to the answering duration being less than the minimum answering duration threshold.

[0085] In an embodiment of the present invention, by using the first physiological data collected by the first wearable device during normal times and the second physiological data during answering questions, it can be determined whether the user is the same person. This can effectively avoid cheating or incorrect wearing, and effectively improve the evaluation accuracy.

[0086] In an embodiment of the present invention, the physiological data of the user during normal times and the answering data are effectively combined. Based on the appearance, the potential psychological state of the user can be captured for comprehensive evaluation, improving the evaluation accuracy.

[0087] In an embodiment of the present invention, the answering behavior of the user during the answering process is collected and analyzed. In an embodiment of the present invention, the potential psychological state of the user is captured through the answering behavior, the psychological state of the user is evaluated to obtain a behavior score; then, according to the behavior score and the content score, a consistency judgment is made to obtain a consistency score. In an embodiment of the present invention, the content score, the behavior score, and the consistency score are effectively combined to comprehensively evaluate the mental health of the user, avoiding the one-sidedness of a single content score evaluation, effectively reducing the subjective deviation, and improving the accuracy of mental health evaluation.

[0088] In an embodiment of the present invention, the consistency score evaluates the matching degree between the content score and the behavior score, avoiding the over-deviation of the comprehensive evaluation caused by the excessive deviation of the two, and further improving the accuracy of mental health evaluation.

[0089] In an embodiment of the present invention, in response to the consistency score being less than the first consistency threshold, it is determined that the first user has a cheating behavior. When the consistency score is too small, it indicates that the user wants to conceal their mental health status. In an embodiment of the present invention, it can be discovered in time and reported for review.

[0090] In an embodiment of the present invention, in response to the answering duration being less than the minimum answering duration threshold, it is determined that the first user has a random answering behavior. In this way, the behavior of the user answering randomly perfunctorily can be avoided.

[0091] In an embodiment of the present invention, the reading speed of the individual is also combined to obtain the cognitive load number, making the cognitive load number more accurate, and further obtaining a more accurate behavior score.

[0092] To illustrate the user identity verification process based on heart rate, blood pressure, and galvanic skin response data, the following supplements the specific data input format and examples.

[0093] a. Original data input format The system needs to collect the sequential physiological data of the user for N consecutive days (such as 7 days), and record the timestamp, data type, and auxiliary information (such as action status). The following is a typical input format (taking JSON as an example): { "user_id": "U20250512001", / / Unique user identifier "device_id": "WATCH-007", / / Wearable device ID "data": { "timestamp": "2025-05-10 08:00:00", / / Timestamp (year-month-day hour:minute:second) "data_type": "heart_rate", / / Data type (heart rate) "value": 72, / / Heart rate value (BPM, beats per minute) "state": "rest" / / Activity state (rest / exercise / sleep, etc.) }, { "timestamp": "2025-05-10 08:05:00", "data_type": "blood_pressure", / / Blood pressure "systolic": 120, / / Systolic blood pressure (mmHg) "diastolic": 80, / / Diastolic blood pressure (mmHg) "state": "rest" }, { "timestamp": "2025-05-10 08:10:00", "data_type": "eda", / / Electrodermal activity (EDA) "value": 1.2, / / Skin conductivity (μS, microsiemens) "state": "rest" }, { "timestamp": "2025-05-10 09:30:00", "data_type": "accelerometer", / / Triaxial acceleration (for assisting in judging activity) "x": 0.1, "y": 0.2, "z": 9.8 / / Acceleration components (m / s²) }] [[ID=54}} b. Specific data examples (excerpts of the first physiological data for 7 consecutive days) The following are the physiological data of a certain user in the resting state for 3 consecutive days (N = 7 days, only the first 3 days are shown) for generating a personalized physiological baseline model: 1. Heart rate data (resting state) (Note: The actual data volume is more than this data) Table 1. Schematic table of heart rate data Example of feature extraction: Mean resting heart rate: (72 + 75 + 68 + 70 + 73 + 71) / 6 ≈ 71.8 BPM; Heart rate variability (HRV): The root mean square of the differences between adjacent R-R intervals (RMSSD) is 5.2 ms (reflecting the stability of autonomic nerve activity).

[0094] 2. Blood pressure data (resting state) Table 2. Schematic table of blood pressure data Example of feature extraction: Blood pressure trend feature: Mean systolic blood pressure is 120 mmHg, standard deviation ±1.5 mmHg (high stability); Orthostatic change response (combined with acceleration data): When the user stands up from a sitting position (acceleration z-axis changes from 9.8 → 8.5 m / s²), the systolic blood pressure rises by 5 mmHg (120 → 125), and the diastolic blood pressure rises by 3 mmHg (80 → 83), which conforms to the normal orthostatic blood pressure change pattern.

[0095] 3. Electrodermal activity (EDA) data (resting state) Table 3. Schematic table of electrodermal activity data Example of feature extraction: Mean basal electrodermal level: 1.3 μS; Amplitude of skin electrical response (peak value during emotional fluctuation - basal value): For example, at 15:00, 1.5 - 1.2 = 0.3 μS (mild tension); Rise time (time from basal value to peak value): Approximately 30 seconds (conforming to the normal emotional activation speed).

[0096] c. Specific implementation of data comparison When the user answers the questionnaire, the system collects the second physiological data in real time (with the same format as the first physiological data), extracts the feature vectors of the same dimension (such as the mean resting heart rate, RMSSD of HRV, blood pressure trend, body position response, basic EDA level, etc.), and calculates the matching degree with the personalized baseline model (such as the time series pattern trained by LSTM).

[0097] Comparison result: Characteristics of the second physiological data of the real user: Resting heart rate 70 BPM (baseline mean 71.8, deviation +2.5%), HRV-RMSSD 5.0 ms (baseline 5.2, deviation -3.8%), blood pressure trend 121 / 81 mmHg (baseline 120 / 80, deviation +0.8%), basic EDA 1.2 μS (baseline 1.3, deviation -7.7%); Matching degree score (Mahalanobis distance): 0.85 (threshold set to 0.8), determined to be the same user. For example Figure 3 It is a support vector description model diagram for user identification based on wearable devices. Among them, test point A is within the boundary of the high-dimensional hypersphere of the registered user's physiological characteristics, and it is determined that the wearing user and the registered user are the same user; while test point B is outside the boundary of the high-dimensional hypersphere of the registered user's physiological characteristics, and it is determined to be an abnormal user.

[0098] In summary, the embodiment of the present invention improves the accuracy of mental health assessment.

[0099] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the element.

[0100] Each embodiment in this specification is described in a related manner. The same or similar parts between each embodiment can be referred to each other, and the differences between each embodiment and other embodiments are emphasized. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0101] The above are only the preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention are all included in the protection scope of the present invention.

Claims

1. An AI-based mental health assessment method, characterized in that, The method includes: Step S1: Authenticate the first wearable device worn by the user and obtain the first physiological data of the user collected by the first wearable device worn by the user in the recent consecutive N days; the first physiological data includes: heart rate data, blood pressure data, sleep quality data, and galvanic skin response data; Step S2: Obtain the first answer data filled in by the user on the electronic terminal for the mental health assessment questionnaire; and during the questionnaire answering period, obtain the second physiological data of the user collected by the wearable device in real time and generate real-time data features; Step S3: Compare the real-time data features generated by the first physiological data and the second physiological data to determine whether the first physiological data and the second physiological data belong to the same user; if they do not belong to the same user, terminate the assessment and alarm; Step S4: In response to the first physiological data and the second physiological data belonging to the same user, input the following data into the AI assessment model: the first answer data, the first physiological data, and output the mental health status assessment result and personalized intervention suggestions of the user.

2. The AI-based mental health assessment method according to claim 1, wherein, The step S3 includes: Step S31: Preprocess and extract features from the first physiological data to generate a first multi-dimensional feature vector; Among them, the first multi-dimensional feature vector includes a combination of at least two of the following features: Resting heart rate, time domain / frequency domain indicators of heart rate variability (HRV); Blood pressure trend features and postural change response features; the postural change response features are generated by combining the triaxial acceleration data of the first wearable device with the blood pressure data; Galvanic skin response amplitude, rise / fall time, and basal galvanic skin level; Step S32: Input the first multi-dimensional feature vector into a machine learning model and train to generate a personalized physiological baseline model corresponding to the user; Step S33: Extract a second feature vector with the same dimension as the first multi-dimensional feature vector from the second physiological data during the questionnaire answering period; Step S34: Calculate the matching degree score between the second feature vector and the personalized physiological baseline model; Step S35: When the matching degree score exceeds a preset threshold, determine that the currently worn user and the registered user are the same person.

3. The AI-based mental health assessment method according to claim 2, wherein The step S31 includes: Extract the resting heart rate mean and the heart rate recovery slope after exercise from the heart rate data; Extract the ratio of low-frequency power (LF) to high-frequency power (HF) (LF / HF) and the root mean square of the differences between adjacent R-R intervals (RMSSD) from the heart rate variability analysis; Identify specific behavioral events in combination with acceleration data, and extract the EDA response peak and blood pressure fluctuation amplitude within a preset time window before and after the event; The personalized physiological baseline model in the step S32 is constructed using any of the following algorithms: Hidden Markov Model (HMM), used to model the time series transition probability of physiological states; Support Vector Data Description (SVDD), used to establish the high-dimensional hypersphere boundary of user physiological characteristics; Long Short-Term Memory Network (LSTM), used to learn the multi-parameter joint time series dependence pattern; The matching degree calculation in the step S34 adopts one of the following methods: Calculate the dynamic time warping (DTW) distance between the new feature vector and the baseline model feature vector; Output the probability value of belonging to the registered user through the pre-trained binary classifier; The Mahalanobis distance is used to measure the deviation of new data from the baseline model distribution.

4. The AI-based mental health assessment method according to claim 1, wherein The step S2 further comprises: Step S21, collecting the first user's answer content data to the mental health assessment questionnaire and the first user's answering behavior during the answering process; wherein the answering behavior at least includes the mouse track, answering time, and number of option modifications; Step S22: obtaining a content score of the first user according to the answer content data; wherein the content score is positively correlated or negatively correlated with the mental health assessment of the first user; Step S23, according to the mouse trajectory, obtaining the mouse trajectory complexity and invalid movement ratio of the first user; according to the answering time, obtaining the cognitive load of the first user; according to the number of option modifications, obtaining the decision conflict degree; according to the mouse trajectory, the answering time and the number of option modifications, obtaining the attention dispersion of the first user; Step S24, obtaining a behavior score of the first user according to the mouse trajectory complexity, the invalid movement ratio, the cognitive load number, the decision conflict degree, and the attention dispersion; wherein the behavior score is positively correlated or negatively correlated with the mental health assessment of the first user; Step S25: obtaining a consistency score of the first user according to the content score and the behavior score; wherein the values of the content score and the behavior score are positively correlated with the mental health assessment of the first user; Step S26: Obtain a mental health assessment result of the first user according to the content score, the behavior score, and the consistency score.

5. The AI-based mental health assessment method according to claim 4, wherein The step S24 comprises: according to Obtain the behavior score of the first user; wherein, is the behavior score, is 4, is the mouse trajectory complexity, is the weight of the mouse trajectory complexity, is the proportion of invalid moves, is the weight of the proportion of invalid moves, is the cognitive load number, is the weight of the cognitive load number, is the decision conflict degree, is the weight of the decision conflict degree, is the attention dispersion degree, is the weight of the attention dispersion degree.

6. The AI-based mental health assessment method according to claim 4, characterized in that, The step S26 includes: according to Obtain the mental health assessment result; wherein, is the score corresponding to the mental health assessment result, is the content score, is the weight of the content score, is the behavior score, is the weight of the behavior score, is the consistency score, is the weight of the consistency score.

7. The AI-based mental health assessment method according to claim 6, wherein The method further comprises According to the specific scores of the content score, the behavior score and the consistency score, the weights corresponding to the content score, the behavior score and the consistency score are determined; wherein the weight corresponding to the consistency score is positively correlated with the content score and the behavior score.

8. The AI-based mental health assessment method according to claim 6, wherein After step S25, the method further includes: In response to the consistency score being less than a first consistency threshold, it is determined that the first user has cheated; wherein the cheating behavior includes concealing one's own mental health condition.

9. The AI-based mental health assessment method according to claim 6, wherein After step S21, the method further includes: In response to the answering time being less than a minimum answering time threshold, it is determined that the first user has given random answers.

10. An AI-based mental health assessment system, characterized in that, The system comprises: a first physiological data acquisition unit, a second physiological data acquisition unit, a user comparison unit and an evaluation unit; The first physiological data acquisition unit is used to authenticate the first wearable device worn by the user who enters the system, and acquire the first physiological data of the user collected by the first wearable device worn by the user in the recent consecutive N days; the first physiological data includes: heart rate data, blood pressure data, sleep quality data, and galvanic skin response data; The second physiological data acquisition unit is used to acquire the first answering data filled in by the user on the electronic terminal for the mental health assessment questionnaire; and during the questionnaire answering period, acquire the second physiological data of the user collected by the wearable device in real time and generate real-time data features; The user comparison unit is used to compare the real-time data features generated by the first physiological data and the second physiological data, and determine whether the first physiological data and the second physiological data belong to the same user; if the two are not the same user, terminate the assessment and give an alarm; The evaluation unit is used to, in response to the first physiological data and the second physiological data belonging to the same user, input the following data into the AI evaluation model: the first answering data, the first physiological data, and output the mental health status evaluation result and personalized intervention suggestions of the user.

Images