Cognitive impairment assessment system and method based on general artificial intelligence

Abstract

The invention discloses a cognitive impairment assessment system and method based on general artificial intelligence, and the method comprises the following steps: S1, collecting multi-modal data through a wearable device, a mobile application and an online tool; s2, carrying out preprocessing on the multi-modal data; s3, comprehensively analyzing the preprocessed data, evaluating cognitive competence, and identifying a cognitive impaired field; s4, a rehabilitation training plan is generated through a neural evolution algorithm, and the training content, sequence, difficulty and type are dynamically adjusted; s5, providing cognitive training and adjusting tasks in real time through virtual reality and augmented reality technologies; s6, predicting cognitive load and psychological state, and dynamically adjusting the intervention plan; and S7, protecting data privacy by using a block chain technology, monitoring a training effect, and generating a progress report. According to the method, the intervention plan and the privacy protection technology are dynamically adjusted through multi-modal data analysis based on general artificial intelligence, and a personalized and real-time cognitive impairment assessment and intervention scheme is provided.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly to a cognitive dysfunction assessment system and method based on general artificial intelligence. Background Art

[0002] With the acceleration of the social aging process, the incidence of cognitive dysfunction is increasing day by day. Especially in the elderly population, cognitive impairment has become a major health problem affecting the quality of human life and social resource consumption. Cognitive dysfunction usually manifests as the decline of memory, attention, judgment and other aspects, which not only seriously affects the daily life of patients, but also brings a huge burden to families and society. Therefore, early diagnosis and effective intervention become particularly important. Traditional cognitive function assessment methods mainly rely on doctors to evaluate through means such as questionnaires and behavioral tests. Although these methods can help with preliminary diagnosis, there are still many deficiencies in terms of real-time, personalization and long-term monitoring.

[0003] Existing cognitive dysfunction assessment methods mostly rely on traditional medical questionnaires and clinical behavioral tests, such as the Montreal Cognitive Assessment Scale (MoCA), Mini-Mental State Examination (MMSE), etc. These methods are relatively common in clinical applications and can judge the cognitive state of patients by quantifying the answers to questions or the completion of specific behavioral tasks. However, the limitations of traditional assessment methods are also very obvious. First of all, the assessment results of these methods are often affected by the patient's mental state at that time and cannot perform long-term and dynamic cognitive load monitoring. For example, in the early stages of some cognitive dysfunctions, patients may be able to answer questionnaires normally, but their cognitive load in daily life has already increased significantly, and this change cannot be captured in time.

[0004] Secondly, traditional assessment methods rely too much on the judgment of professional doctors and are easily affected by human factors. Different doctors may have different assessment results for the same patient, and there is a lack of standardized assessment processes. In some areas with scarce resources or insufficient human resources, patients may not be able to receive regular assessments by professional doctors, resulting in the failure to detect and intervene in the early symptoms of cognitive dysfunction in a timely manner.

[0005] In addition, existing cognitive assessment methods are difficult to achieve personalized intervention. Traditional methods usually diagnose according to unified assessment criteria and are difficult to provide personalized intervention measures according to the actual needs of different patients. Since the degree of cognitive impairment and recovery potential of each patient are different, unified intervention methods often have insignificant effects and may even cause side effects. Therefore, the lack of personalized intervention programs is a major shortcoming of existing assessment systems.

[0006] With the rapid development of artificial intelligence technology, more and more intelligent assessment tools have been proposed and began to be applied in the early diagnosis and intervention of cognitive impairment. Based on technologies such as deep learning, natural language processing, and computer vision, more and more intelligent systems can comprehensively analyze multi-modal data such as users' behaviors, voices, and facial expressions, and extract more detailed cognitive function features. However, there are still certain technical bottlenecks in the existing artificial intelligence-based cognitive function assessment methods. First of all, most of the existing methods focus on cognitive detection in a certain aspect. For example, only through facial expression analysis or speech recognition to predict cognitive impairment, while ignoring the integration of multi-modal data. The assessment of cognitive function not only depends on a single behavioral performance, but also needs to consider comprehensive factors such as cognitive load and emotional fluctuations. Therefore, the existing technologies often cannot comprehensively evaluate the cognitive state of patients, lacking comprehensiveness and accuracy.

[0007] Secondly, the existing artificial intelligence assessment systems usually rely on centralized data storage and processing methods, which are prone to cause data privacy and security problems. Especially in the medical field, the privacy protection of users is particularly important. However, the technical means in data privacy protection of the existing technologies are still insufficient. Many intelligent assessment systems use traditional centralized storage methods, storing patients' sensitive data on cloud servers, which are prone to the risks of data leakage and abuse. In order to ensure the privacy and security of patients' data, it is very necessary to adopt distributed storage and encryption technologies, but the application of the existing technologies in this regard is not yet mature.

[0008] In addition, although artificial intelligence technology has made some progress in cognitive assessment, the existing assessment systems usually lack the ability of real-time dynamic adjustment. The cognitive state and emotional changes of users are a continuous development process, and traditional assessment methods usually cannot provide the functions of real-time monitoring and dynamic adjustment. And the existing artificial intelligence assessment systems also mostly conduct assessments at fixed time points, lacking a dynamic prediction and adjustment mechanism for the long-term changes of users. How to make timely adjustments during the process of patients' cognitive state changes and provide personalized intervention plans is a major challenge for the existing technologies.

[0009] In addition, the existing artificial intelligence assessment systems usually rely on traditional data analysis methods and lack the application of intelligent contracts and blockchain technologies. Although artificial intelligence can efficiently analyze patients' multi-modal data, the security, privacy protection, and permission control of data are still problems that need to be solved urgently. Blockchain technology can effectively solve the data privacy problem through decentralized data storage and encryption processing, but it has not been widely applied in cognitive assessment systems yet.

[0010] Therefore, how to provide a cognitive impairment assessment system and method based on general artificial intelligence is an urgent problem for those skilled in the art. Summary of the Invention

[0011] An object of the present invention is to provide a cognitive impairment assessment system and method based on general artificial intelligence. The present invention makes full use of multi-modal data analysis, blockchain technology, smart contracts, and deep learning algorithms, and details the process of generating a personalized intervention plan and dynamically adjusting it according to real-time feedback by monitoring the user's cognitive state, emotional fluctuations, and training effects in real time. The system can control data access rights through smart contracts, protect the privacy of user data, and use blockchain technology for decentralized storage of data to ensure the security and immutability of data. It has the advantages of data privacy protection, personalized intervention, dynamic adjustment, and real-time feedback, and can provide accurate and reliable cognitive assessment and intervention solutions for users.

[0012] The cognitive impairment assessment method based on general artificial intelligence according to an embodiment of the present invention includes the following steps:

[0013] S1. Collect multi-modal data of the user through wearable devices, mobile applications, and online quantitative assessment tools, including voice, facial expressions, behavior patterns, and physiological data;

[0014] S2. Preprocess the user's multi-modal data, remove noise, standardize it, and extract the user's key cognitive features, including language complexity, behavior regularity, emotional response characteristics, and physiological indicators;

[0015] S3. Combine the preprocessed multi-modal data of the user, introduce a cognitive impairment knowledge graph, and use an adaptive multi-modal graph neural network for comprehensive analysis to evaluate the user's cognitive ability and identify the impaired areas of cognition;

[0016] S4. Combine the user's multi-modal data and evaluation results, use a neuroevolution algorithm to generate a rehabilitation training plan, and dynamically adjust the order, difficulty, and training type of the training content according to the user's cognitive impairment situation;

[0017] S5. Through the rehabilitation training plan, combine virtual reality and augmented reality technologies to provide an immersive cognitive training environment, collect the user's virtual environment training data, and adjust the training tasks and difficulty in real time;

[0018] S6. Based on the preprocessed multi-modal data of the user and the virtual environment training data, dynamically predict the user's cognitive load and mental state, and dynamically adjust the training content, environment, and intervention intensity at different time points;

[0019] S7. Based on the dynamic adjustment results, monitor the user's training effect in real time, generate a progress report, and adjust the intervention plan according to the feedback.

[0020] Optionally, the specific steps of S3 include:

[0021] S31. Extract features from the preprocessed user multi-modal data, specifically including extracting key features from the user's speech, facial expressions, behavior patterns, and physiological data, and standardizing them into a data format suitable for input into the neural network;

[0022] S32. Introduce a cognitive impairment knowledge graph, map the multi-modal features to the nodes in the cognitive impairment knowledge graph, and perform multi-dimensional weighted sum and information fusion based on the complex relationships between the edges and nodes of the cognitive impairment knowledge graph. Adopt a multi-level graph propagation mechanism to recursively calculate the influence between nodes to obtain an enhanced feature vector:

[0023]

[0024] where, X i is the feature vector of the i-th type of modal data, A i is the weight matrix corresponding to the features of node i in the knowledge graph, α i,k is the adaptive weighting coefficient of the relationship between node i and the k-th layer of the graph, G j (X) is the graph propagation function, B j is the weighting matrix of the j-th node during the graph propagation process, X enhanced is the enhanced feature vector, K is the number of layers of graph propagation, M is the total number of nodes in the graph, N is the number of types of modal data, β j,k is the weighting coefficient of graph node j during the k-th layer of propagation;

[0025] S33. Input the enhanced feature vector into an adaptive multi-modal graph neural network, perform feature fusion through the graph convolutional layer, and dynamically adjust the weights according to the complex relationships between the nodes in the graph and the feature importance of each modality to generate a comprehensive cognitive feature vector. Through multi-layer graph convolutional operations, the network gradually propagates and fuses information to calculate the comprehensive features:

[0026]

[0027] where, represents the output feature after the (l + 1)-th layer of graph convolutional operation, σ is the activation function, A ij is the adjacency matrix between node i and node j in the graph, is the adaptive weighting coefficient between node i and node j at the l-th layer, is the weight matrix;

[0028] S34. Perform temporal modeling on the comprehensive features output by the graph neural network. Use a long short-term memory network (LSTM) with an attention mechanism and gating control to model the time series data, capture the dynamic change trend of the user's cognitive ability, and combine with the non-linear weighted gating function G of the temporal features t and the adaptive weighted coefficient α of the time step t , dynamically weight the features of each time step to enhance the sensitivity of the model to cognitive fluctuations:

[0029] H t = LSTM(F GNN,t , H t-1 , G t );

[0030]

[0031] where H t represents the hidden state output of the long short-term memory network at the time step t. ⊙ is element-wise multiplication. F weighted is the weighted comprehensive feature vector. F GNN,t is the output feature vector at the time step t. H t-1 is the hidden state output at the previous time step t - 1. T is the total number of time steps in the time series;

[0032] S35. Based on the results output by the graph neural network and temporal modeling, use a multi-layer perceptron with a hierarchical weighting mechanism to evaluate the cognitive ability and generate scores for each cognitive domain. Adopt a hierarchical weighted loss function to dynamically adjust the loss function according to the priorities of each cognitive domain and optimize the cognitive evaluation results:

[0033]

[0034] where Y i is the evaluation score of the i-th cognitive domain. Y target,i is the true score. w i is the weight of this domain. L weighted is the weighted loss function. D is the total number of cognitive domains.

[0035] Optionally, the specific steps of S4 include:

[0036] S41. According to the evaluated cognitive ability results, determine the impaired domains of the user's cognitive ability, define intervention goals for each impaired domain, and generate an intervention goal vector G = [g1, g2,..., g m , where g m represents the intervention goal of the m-th cognitive domain, and the priority of each goal is sorted by a multi-objective optimization algorithm to optimize the goal priority vector P = [p1, p2,..., p m, where p m represents the priority value of the m-th cognitive goal;

[0037] S42. According to the generated intervention target vector and target priority vector, use the neuroevolution algorithm to optimize the user's personalized intervention training plan. The combination form of the training tasks is T = [t1, t2, …, t n , where t n is the parameter of the n-th training task, including the task type, difficulty, and training goal;

[0038] S43. Introduce an adaptive feedback mechanism to adjust the generated training tasks in real time. Monitor the user's performance during training through a feedback loop, and combine the immediate performance data D = [d1, d2, …, d n . Use an adaptive neural network to dynamically adjust the training tasks, and optimize the difficulty and order of the tasks in real time:

[0039]

[0040] Among them, X t is the user input data at time step t, F t is the historical training feedback data, D t is the generated dynamic adjustment feedback data, LSTM is used to capture the long-term dependencies in time series data, ARIMA is used to capture short-term dependencies and trend changes, α i is the weighting coefficient, and n is the number of data feature dimensions;

[0041] S44. Combine the adjusted training tasks and user feedback, and globally optimize the training plan through adaptive meta-learning and multi-modal learning;

[0042] S45. According to the optimization results, integrate all training tasks and feedback into a preliminary personalized intervention plan P = [p1, p2, …, p k through a multi-modal learning method, and dynamically update the intervention strategy according to the priority and progress of each training task.

[0043] Optionally, the S44 specifically includes:

[0044] S441. Simultaneously consider cognitive load, emotional fluctuations, and task completion through a multi-objective optimization model, and combine the immediate rewards and long-term returns in adaptive meta-learning to define the overall optimization goal of the training plan;

[0045] S442. The genetic algorithm generates diverse combinations of training tasks through selection, crossover, and mutation operations, and combines a multi-modal dynamic weight adjustment mechanism to dynamically balance the fitness values of the tasks:

[0046]

[0047] Among them, F(T i ) represents the fitness value of the training plan T i , where T is the total number of time steps, M is the total number of tasks, ω j is the weighting coefficient of task j, is the immediate reward obtained by task T i at time step t, is the impact of task T i on the user's cognitive load at time step t, is the emotional fluctuation value, α1 and α2 are the balance coefficients of positive and negative impacts, and β is the emotional fluctuation sensitivity adjustment coefficient;

[0048] S443. Through the dynamic adaptive adjustment of task weights, collaborative optimization is carried out among multiple tasks:

[0049]

[0050] Among them, O AMTL represents the overall optimization goal of multi-task learning, K is the total number of tasks, T k represents the k-th task, λ k is the dynamic weighting coefficient of task k, γ is the task weight dynamic adjustment coefficient, ΔF(T k ) is the fitness change of task T k , F avg is the average fitness, and F(T k ) represents the fitness value of the training plan T k ;

[0051] S444. Through the combination of adaptive meta-learning and multi-modal learning, real-time dynamic adjustment of training tasks is carried out, and the multi-modal dynamic weight adjustment mechanism adjusts task weights based on user feedback:

[0052]

[0053] Among them, T i,new are the training task parameters after dynamic adjustment, T i,old are the previous task parameters, η is the learning rate, α m is the dynamic weight of task m in multi-modal data, is the immediate reward of the task at the current time step, are the numerical values of cognitive load and emotional fluctuation respectively, and β1 and β2 are the weight coefficients of cognitive load and emotional fluctuation;

[0054] S445. Dynamically adjust the training plan by integrating the output results of genetic algorithms, adaptive multi-task learning, and multi-modal dynamic weight adjustment mechanisms, and adjust the order, difficulty, and type of training tasks according to the feedback of different users.

[0055] Optionally, the S5 specifically includes:

[0056] S51. According to the generated personalized training plan, combine virtual reality and augmented reality technologies to construct an immersive cognitive training environment, simulating the cognitive scenarios in the user's daily life, where the difficulty of the training tasks and the environmental parameters are controlled by the training plan T = [t1, t2,..., t k , where t k represents the type, difficulty, and target time of the kth task;

[0057] S52. Real-time monitor the user's performance in the virtual environment, collect training data including task completion time, error rate, reaction time, and emotional fluctuations, and form a user performance dataset B t = [b1, b2,..., b n , where b n represents the performance data of the nth training task;

[0058] S53. Perform temporal modeling and feature extraction on the collected user performance data through a spatio-temporal convolutional neural network, and output the behavioral pattern features H t of the user in the virtual environment, where the behavioral pattern features are the feature representations at time step t, capturing the spatio-temporal characteristics of the user's cognitive load, emotional fluctuations, and behavioral responses:

[0059]

[0060] where K i is the convolutional kernel, W i is the weight of each feature channel, σ is the activation function, * represents the convolution operation, W a is the learning weight in the self-attention mechanism, Softmax() is the normalization function, and n is the number of features in the user performance dataset;

[0061] S54. According to the generated behavioral pattern features and training task objectives, use a generative adversarial network to dynamically adjust the task settings in the virtual environment. The goal is to generate training tasks that adapt to the user's cognitive state and emotional fluctuations. Specifically, the generative model G(H t , T) generates new training tasks based on the user's behavioral characteristics and training task plan, while the discriminative model D(H t , T) determines whether the generated tasks meet the cognitive needs of the current user:

[0062]

[0063] Among them, L GAN is the loss function of the generative adversarial network, T is the training task plan, is the expectation operation, and ln is the logarithmic function;

[0064] S55. Generate an optimized personalized intervention training plan based on the optimized training task parameters and feedback results, and dynamically adjust the training tasks and difficulty in real time according to real-time feedback.

[0065] Optionally, the S6 specifically includes:

[0066] S61. Construct a comprehensive user cognitive load and mental state evaluation model through the preprocessed user multi-modal data and the collected virtual environment training data. Combine the user's behavior patterns, emotional fluctuations, and cognitive performance data to perform dynamic prediction based on time series, evaluate the user's cognitive load, mental state, and emotional fluctuations, and track the user's cognitive health status in real time to accurately predict the load changes and emotional states at different time points:

[0067] C t = f(D t , E t , Q t );

[0068] Among them, C t is the cognitive load and mental state evaluation at time step t, D t is the user performance data collected in the virtual environment, E t is the user behavior data, Q t is the emotional fluctuation data, and f() is the comprehensive evaluation function;

[0069] S62. Apply dynamic time series analysis technology to the evaluated cognitive load and mental state data to model the cognitive state fluctuations of each user, continuously monitor the changing trends of cognitive load and emotions, and timely identify the changes in the user's cognitive ability or the patterns of emotional fluctuations to form the user's dynamic cognitive state trajectory:

[0070]

[0071] Among them, is the predicted cognitive load and mental state, C t-1 , C t-2 …,, C t-k are the historical cognitive load and mental state data, A t is the external influencing factor at the current time step, and g() is the time series prediction function;

[0072] S63. Based on the obtained cognitive state prediction results, combined with the user's real-time feedback, dynamically adjust the training tasks, environmental settings, and intervention intensity:

[0073]

[0074] Among them, ΔT t is the adjustment amount of the training task at time step t, α z and β u are adjustment coefficients, B t is the user performance dataset;

[0075] S64. According to the optimized training tasks and environmental settings, use the reinforcement learning algorithm to globally optimize the intervention strategy and generate personalized intervention tasks:

[0076]

[0077] Among them, γ is the discount factor, r t (θ) is the immediate reward at time step t, is the error metric between the user prediction and the actual cognitive load, λ1 and λ2 are regularization coefficients, is the L2 regularization term, F(θ is the optimization objective function of reinforcement learning, E is the expectation operation, T is the total number of steps in the training cycle, and n is the number of model parameters;

[0078] S65. According to the optimized intervention tasks and feedback results, dynamically adjust the intervention training plan in the long term, and dynamically adjust the training content, environment, and intervention intensity according to the real-time feedback.

[0079] The cognitive dysfunction assessment system based on general artificial intelligence according to the embodiment of the present invention includes the following modules:

[0080] A multimodal data acquisition module for collecting user multimodal data;

[0081] A data preprocessing and feature extraction module for preprocessing the user multimodal data and refining the key information describing the user's cognitive state;

[0082] A cognitive ability assessment module for mining the cognitive features behind the data and locating the most severely damaged area in the user's cognitive ability;

[0083] A personalized intervention plan generation module for designing a daily rehabilitation exercise plan matching the cognitive impairment condition according to the severity of the damaged ability;

[0084] A result feedback and tracking module for real-time monitoring and feedback on the user's training process and effectiveness.

[0085] The beneficial effects of the present invention are:

[0086] The cognitive dysfunction assessment system and method of the present invention based on general artificial intelligence, by introducing multi-modal data analysis, blockchain technology, smart contracts and deep learning algorithms, not only effectively solves the limitations of traditional cognitive assessment methods, but also provides a new personalized intervention solution. Compared with the prior art, the present invention achieves beneficial effects in multiple aspects.

[0087] First of all, the present invention makes full use of multi-modal data analysis. By integrating information such as the user's behavior data, emotional fluctuations, reaction time, etc., it can comprehensively evaluate the user's cognitive state and psychological burden. This multi-dimensional data collection and analysis method overcomes the deficiency of traditional assessment methods that only rely on questionnaires or single behavioral tests, and provides more accurate and comprehensive cognitive assessment results.

[0088] Secondly, the present invention uses blockchain technology for data storage and management, ensuring the privacy protection and security of user data. Through the decentralized data storage method and the access control of smart contracts, the present invention effectively avoids the risks of data leakage and abuse. Data is only accessed under the condition of meeting the user's authorization, and all data operation and access records are traceable, ensuring the transparency and compliance of data processing.

[0089] In addition, the present invention manages data access permissions by introducing smart contracts, ensuring the automation and seamless connection of the entire assessment process. Smart contracts can automatically execute tasks according to predetermined rules, reducing human intervention and improving the efficiency and security of the system. The application of this mechanism greatly enhances the data privacy protection ability during the intervention process, and at the same time ensures the compliance of assessment and intervention tasks.

[0090] The present invention also has the significant advantages of personalized intervention and dynamic adjustment. By real-time monitoring the user's training effect and cognitive state changes, the system can timely adjust the task content, sequence and intervention intensity to adapt to the user's cognitive load and emotional fluctuations. This dynamic adjustment ability ensures the personalization and flexibility of the intervention plan, thereby improving the accuracy and efficiency of the intervention effect, avoiding the "one-size-fits-all" intervention plan in traditional methods, and ensuring that each user's intervention plan closely matches their cognitive needs and emotional fluctuations.

[0091] In summary, the beneficial effects of the present invention are reflected in improving the accuracy of cognitive dysfunction assessment, strengthening data privacy protection, enhancing the personalization and real-time nature of assessment and intervention, and ensuring the excellent performance of the system in terms of data security and compliance through the innovative application of blockchain and smart contract technologies. These improvements make the assessment and intervention of cognitive dysfunction more intelligent, safe and efficient, and have broad application prospects. Brief Description of the Drawings

[0092] The accompanying drawings are used to provide a further understanding of the present invention and form a part of the specification. They are used in conjunction with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the accompanying drawings:

[0093] Figure 1 is a flowchart of the cognitive impairment assessment method based on general artificial intelligence proposed by the present invention;

[0094] Figure 2 is a schematic structural diagram of the cognitive impairment assessment system based on general artificial intelligence proposed by the present invention. Detailed implementation manners

[0095] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only showing the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0096] Referring to Figure 1 , the cognitive impairment assessment method based on general artificial intelligence includes the following steps:

[0097] S1. Collect multi-modal data of the user through wearable devices, mobile applications and online quantitative assessment tools, including voice, facial expressions, behavior patterns and physiological data;

[0098] S2. Preprocess the user's multi-modal data, remove noise, standardize it, and extract the user's key cognitive features, including language complexity, behavior regularity, emotional response characteristics and physiological indicators;

[0099] S3. Combine the preprocessed multi-modal data of the user, introduce the cognitive impairment knowledge graph, and use an adaptive multi-modal graph neural network for comprehensive analysis to evaluate the user's cognitive ability and identify the impaired areas of cognition;

[0100] S4. Combine the user's multi-modal data and the evaluation results, use a neuroevolution algorithm to generate a rehabilitation training plan, and dynamically adjust the order, difficulty and training type of the training content according to the user's cognitive impairment situation;

[0101] S5. Through the rehabilitation training plan, combine virtual reality and augmented reality technologies to provide an immersive cognitive training environment, collect the user's virtual environment training data, and dynamically adjust the training tasks and difficulty in real time;

[0102] S6. Based on the preprocessed multi-modal data of the user and the virtual environment training data, dynamically predict the user's cognitive load and mental state, and dynamically adjust the training content, environment and intervention intensity at different time points;

[0103] S7. Based on the dynamic adjustment results, monitor the user's training effect in real time, generate a progress report, and adjust the intervention plan according to the feedback.

[0104] In this embodiment, the S3 specifically includes:

[0105] S31. Extract features from the preprocessed user multi-modal data, specifically including extracting key features from the user's speech, facial expressions, behavior patterns, and physiological data, and standardizing them into a data format suitable for input into the neural network;

[0106] S32. Introduce a cognitive impairment knowledge graph, map the multi-modal features to the nodes in the cognitive impairment knowledge graph, and perform multi-dimensional weighted and information fusion based on the complex relationships between the edges and nodes of the cognitive impairment knowledge graph. Adopt a multi-level graph propagation mechanism to recursively calculate the influence between nodes to obtain an enhanced feature vector:

[0107]

[0108] Among them, X i is the feature vector of the i-th type of modal data, A i is the weight matrix corresponding to the feature of node i in the knowledge graph, α i,k is the adaptive weighting coefficient of the relationship between node i and the k-th layer of the graph, G j (X) is the graph propagation function, B j is the weighting matrix of the j-th node during the graph propagation process, X enhanced is the enhanced feature vector, K is the number of layers of graph propagation, M is the total number of nodes in the graph, N is the number of types of modal data, β j,k is the weighting coefficient of graph node j during the k-th layer of propagation;

[0109] S33. Input the enhanced feature vector into an adaptive multi-modal graph neural network, perform feature fusion through the graph convolution layer, and dynamically adjust the weights according to the complex relationships between the nodes in the graph and the feature importance of each modality to generate a comprehensive cognitive feature vector. Through multi-layer graph convolution operations, the network gradually propagates and fuses information to calculate the comprehensive feature:

[0110]

[0111] Among them, represents the output feature after the (l + 1)-th layer of graph convolution operation, σ is the activation function, A ij is the adjacency matrix between node i and node j in the graph, is the adaptive weighting coefficient between node i and node j at the l-th layer, is the weight matrix;

[0112] S34. Perform temporal modeling on the comprehensive features output by the graph neural network. Use a long short-term memory network (LSTM) with an attention mechanism and gated control to model the time series data, capture the dynamic change trend of the user's cognitive ability, and combine it with the non-linear weighted gating function G of the temporal features t and the adaptive weighting coefficient α of the time step t , and dynamically weight the features of each time step to enhance the sensitivity of the model to cognitive fluctuations:

[0113] H t = LSTM(F GNN,t , H t-1 , G t );

[0114]

[0115] Among them, H t represents the hidden state output of the long short-term memory network at the time step t. ⊙ is element-wise multiplication. F weighted is the weighted comprehensive feature vector. F GNN,t is the output feature vector at the time step t. H t-1 is the hidden state output at the previous time step t - 1. T is the total number of time steps of the time series;

[0116] S35. Based on the results output by the graph neural network and temporal modeling, use a multi-layer perceptron with a hierarchical weighting mechanism to evaluate the cognitive ability and generate scores for each cognitive domain. Adopt a hierarchical weighted loss function to dynamically adjust the loss function according to the priority of each cognitive domain and optimize the cognitive evaluation results:

[0117]

[0118] Among them, Y i is the evaluation score of the i-th cognitive domain. Y target,i is the true score. w i is the weight of this domain. L weighted is the weighted loss function. D is the total number of cognitive domains.

[0119] In this embodiment, the specific steps of S4 include:

[0120] S41. According to the evaluated cognitive ability results, determine the impaired domains of the user's cognitive ability, define intervention goals for each impaired domain, and generate an intervention goal vector G = [g1, g2,..., g m . Among them, g m represents the intervention goal of the m-th cognitive domain, and the priority of each goal is sorted by a multi-objective optimization algorithm to optimize the goal priority vector P = [p1, p2,..., p m, where p m represents the priority value of the m-th cognitive goal;

[0121] S42. According to the generated intervention target vector and target priority vector, use the neuroevolution algorithm to optimize the user's personalized intervention training plan. The combination form of the training tasks is T = [t1, t2, …, t n , where t n is the parameter of the n-th training task, including the task type, difficulty, and training goal;

[0122] S43. Introduce an adaptive feedback mechanism to adjust the generated training tasks in real time. Through the feedback loop, monitor the user's performance during training in real time, and combine the immediate performance data D = [d1, d2, …, d n , and use the adaptive neural network to dynamically adjust the training tasks, and optimize the difficulty and order of the tasks in real time:

[0123]

[0124] where, X t is the user input data at time step t, F t is the historical training feedback data, D t is the generated dynamic adjustment feedback data, LSTM is used to capture the long-term dependencies in time series data, ARIMA is used to capture short-term dependencies and trend changes, α i is the weighting coefficient, and n is the number of data feature dimensions;

[0125] S44. Combine the adjusted training tasks and user feedback, and globally optimize the training plan through adaptive meta-learning and multi-modal learning;

[0126] S45. According to the optimization results, integrate all training tasks and feedback into a preliminary personalized intervention plan P = [p1, p2, …, p k through the multi-modal learning method, and dynamically update the intervention strategy according to the priority and progress arrangement of each training task.

[0127] In this embodiment, the S44 specifically includes:

[0128] S441. Simultaneously consider cognitive load, emotional fluctuations, and task completion through a multi-objective optimization model, and combine the immediate rewards and long-term returns in adaptive meta-learning to define the overall optimization goal of the training plan;

[0129] S442. The genetic algorithm generates diverse combinations of training tasks through selection, crossover, and mutation operations, and combines the multi-modal dynamic weight adjustment mechanism to dynamically balance the fitness values of the tasks:

[0130]

[0131] Among them, F(T i ) represents the fitness value of the training plan T i , T is the total number of time steps, M is the total number of tasks, ω j is the weighted coefficient of task j, is the immediate reward obtained by task T i at time step t, is the impact of task T i on the user's cognitive load at time step t, is the emotional fluctuation value, α1 and α2 are the balance coefficients of positive and negative impacts, and β is the emotional fluctuation sensitivity adjustment coefficient;

[0132] S443. Through the dynamic adaptive adjustment of task weights, collaborative optimization is carried out among multiple tasks:

[0133]

[0134] Among them, O AMTL represents the overall optimization goal of multi-task learning, K is the total number of tasks, T k represents the k-th task, λ k is the dynamic weighted coefficient of task k, γ is the task weight dynamic adjustment coefficient, ΔF(T k ) is the fitness change amount of task T k , F avg is the fitness mean, F(T k ) represents the fitness value of the training plan T k ;

[0135] S444. Through the combination of adaptive meta-learning and multi-modal learning, real-time dynamic adjustment of training tasks is carried out, and the multi-modal dynamic weight adjustment mechanism adjusts task weights based on user feedback:

[0136]

[0137] Among them, T i,new are the training task parameters after dynamic adjustment, T i,old are the previous task parameters, η is the learning rate, α m is the dynamic weight of task m in multi-modal data, is the immediate reward of the task at the current time step, are the numerical values of cognitive load and emotional fluctuation respectively, and β1 and β2 are the weight coefficients of cognitive load and emotional fluctuation;

[0138] S445. Dynamically adjust the training plan based on the output results of the comprehensive genetic algorithm, adaptive multi-task learning, and multi-modal dynamic weight adjustment mechanism, and adjust the order, difficulty, and type of training tasks according to the feedback of different users.

[0139] In this embodiment, the S5 specifically includes:

[0140] S51. According to the generated personalized training plan, combine virtual reality and augmented reality technologies to construct an immersive cognitive training environment, simulating the cognitive scenarios in the user's daily life, where the difficulty of training tasks and environmental parameters are controlled by the training plan T = [t1, t2, …, t k , where t k represents the type, difficulty, and target time of the k-th task;

[0141] S52. Real-time monitor the user's performance in the virtual environment, collect training data including task completion time, error rate, reaction time, and emotional fluctuations, and form a user performance dataset B t = [b1, b2, …, b n , where b n represents the performance data of the n-th training task;

[0142] S53. Perform temporal modeling and feature extraction on the collected user performance data through a spatio-temporal convolutional neural network, and output the behavioral pattern features H t of the user in the virtual environment, where the behavioral pattern features are the feature representations at time step t, capturing the spatio-temporal characteristics of the user's cognitive load, emotional fluctuations, and behavioral responses:

[0143]

[0144] where K i is the convolutional kernel, W i is the weight of each feature channel, σ is the activation function, * represents the convolution operation, W a is the learning weight in the self-attention mechanism, Softmax() is the normalization function, and n is the number of features in the user performance dataset;

[0145] S54. According to the generated behavioral pattern features and training task objectives, use a generative adversarial network to dynamically adjust the task settings in the virtual environment. The goal is to generate training tasks that adapt to the user's cognitive state and emotional fluctuations. Specifically, the generative model G(H t , T) generates new training tasks based on the user's behavioral characteristics and training task plan, while the discriminant model D(H t , T) determines whether the generated tasks meet the cognitive needs of the current user:

[0146]

[0147] Among them, L GAN is the loss function of the generative adversarial network, T is the training task plan, is the expectation operation, and ln is the logarithmic function;

[0148] S55. Generate an optimized personalized intervention training plan based on the optimized training task parameters and feedback results, and dynamically adjust the training tasks and difficulty in real time according to real-time feedback.

[0149] In this embodiment, the S6 specifically includes:

[0150] S61. Construct a comprehensive user cognitive load and mental state evaluation model through the preprocessed user multi-modal data and the collected virtual environment training data. Combine the user's behavior pattern, emotional fluctuation, and cognitive performance data to perform dynamic prediction based on time series, evaluate the user's cognitive load, mental state, and emotional fluctuation, and track the user's cognitive health status in real time to accurately predict the load change and emotional state at different time points:

[0151] C t = f(D t , E t , Q t );

[0152] Among them, C t is the cognitive load and mental state evaluation at time step t, D t is the user performance data collected in the virtual environment, E t is the user behavior data, Q t is the emotional fluctuation data, and f() is the comprehensive evaluation function;

[0153] S62. Apply dynamic time series analysis technology to the evaluated cognitive load and mental state data to model the cognitive state fluctuation of each user, continuously monitor the change trends of cognitive load and emotion, and timely identify the change of user cognitive ability or the pattern of emotional fluctuation to form the user's dynamic cognitive state trajectory:

[0154]

[0155] Among them, is the predicted cognitive load and mental state, C t-1 , C t-2 …,, C t-k are the historical cognitive load and mental state data, A t is the external influencing factor at the current time step, and g() is the time series prediction function;

[0156] S63. Based on the obtained cognitive state prediction results, combined with the user's real-time feedback, dynamically adjust the training tasks, environmental settings, and intervention intensity:

[0157]

[0158] Among them, ΔT t is the adjustment amount of the training task at time step t, α z and β u are adjustment coefficients, B t is the user performance dataset;

[0159] S64. According to the optimized training tasks and environmental settings, use the reinforcement learning algorithm to globally optimize the intervention strategy and generate personalized intervention tasks:

[0160]

[0161] Among them, γ is the discount factor, r t (θ) is the immediate reward at time step t, is the error metric between the user prediction and the actual cognitive load, λ1 and λ2 are regularization coefficients, is the L2 regularization term, F(θ is the optimization objective function of reinforcement learning, E is the expectation operation, T is the total number of steps in the training cycle, and n is the number of model parameters;

[0162] S65. According to the optimized intervention tasks and feedback results, make long-term dynamic adjustments to the intervention training plan, and dynamically adjust the training content, environment, and intervention intensity according to the real-time feedback.

[0163] Reference Figure 2 , a cognitive dysfunction assessment system based on general artificial intelligence, includes the following modules:

[0164] Multimodal data acquisition module, used to collect user multimodal data;

[0165] Data preprocessing and feature extraction module, used to preprocess the user multimodal data and extract the key information describing the user's cognitive state;

[0166] Cognitive ability assessment module, used to mine the cognitive features behind the data and locate the most severely damaged area in the user's cognitive ability;

[0167] Personalized intervention plan generation module, used to design a daily rehabilitation exercise plan that matches the cognitive impairment status according to the severity of the damaged ability;

[0168] Result feedback and tracking module, used to monitor and feedback the user's training process and effectiveness in real time.

[0169] Example 1:

[0170] To verify the feasibility of the present invention in implementation, the present invention is applied to a certain hospital. In its daily work, the hospital adopts traditional cognitive function assessment methods, and patients mainly rely on doctors' diagnoses and conventional cognitive scales. However, these traditional methods have many limitations. They cannot monitor patients' cognitive load and emotional fluctuations in real time, nor can they perform personalized interventions according to patients' individual differences. To solve these problems, the hospital introduced a cognitive dysfunction assessment system based on general artificial intelligence and conducted more accurate assessments and interventions on patients through this system.

[0171] In the neurology outpatient department of this hospital, patients usually need to undergo a cognitive function assessment once. The assessment cycle of traditional methods is once every three months. Traditional assessment tools can only provide cognitive assessment results in the short term, ignoring the changes in cognitive status at different time points. The cognitive problems of many patients develop gradually, and the initial changes are relatively hidden, making it difficult for traditional tests to capture these subtle changes. In addition, patients' emotional fluctuations and changes in cognitive load are also key factors affecting their cognitive function, but traditional assessments cannot monitor the changes in these factors in real time, thus affecting the effectiveness of interventions.

[0172] To solve this series of problems, the hospital deployed a cognitive dysfunction assessment system based on general artificial intelligence. The core functions of the system include: monitoring in real time the changes in patients' cognitive status, emotional fluctuations, and cognitive load, and dynamically adjusting the intervention plan based on this data. The clinical doctors in the hospital input patients' multimodal data (such as behavioral data, emotional fluctuation data, completion of training tasks, etc.) into the system. The system generates personalized cognitive function assessment reports through deep learning algorithms, reinforcement learning techniques, and blockchain technology, and dynamically adjusts the intervention plan according to patients' cognitive status and emotional fluctuations.

[0173] Specifically, in the implementation scenario of this hospital, the system analyzes and models by collecting data such as patients' daily performance data, emotional fluctuations, and reaction time, using spatio-temporal convolutional neural network (ST-CNN) and deep reinforcement learning (DRL) algorithms. After being encrypted, patients' multimodal data is stored through blockchain technology to ensure the security and privacy of the data. During the entire assessment process, the system dynamically adjusts the intervention intensity and task content according to real-time feedback to ensure that each patient's intervention plan always adapts to their cognitive needs and emotional fluctuations.

[0174] Specifically for a 60-year-old male patient, Mr. Zhang, during the initial diagnosis of cognitive impairment, traditional assessments showed that his score on the MMSE scale was 24, falling within the range of mild cognitive impairment. However, the traditional method could not capture the gradual increase in his cognitive load in daily life, nor could it identify in a timely manner the impact of his mood fluctuations on cognitive load. Therefore, the hospital introduced Mr. Zhang into a cognitive impairment assessment system based on general artificial intelligence for further intervention.

[0175] With the help of the system, Mr. Zhang's cognitive function was monitored in real time. By combining data on his mood fluctuations and changes in cognitive load, the system provided a dynamically updated assessment report. Before the first intervention, the system analyzed and found that Mr. Zhang's cognitive load in daily life was gradually increasing and his mood fluctuations were significant. Based on these analysis results, the system customized a personalized intervention plan for Mr. Zhang, including optimizing cognitive training tasks and adjusting the intensity of intervention, and ensured the privacy protection of all data through smart contracts.

[0176] Mr. Zhang's progress report showed that in the first month after the start of the intervention, his cognitive load decreased significantly, and his mood fluctuations were effectively controlled. Compared with traditional assessments, the system can capture the slightest changes in the patient's cognitive load and adjust the intervention strategy in real time during the training process. Through continuous monitoring by the system, doctors can accurately judge the intervention effect and make timely adjustments, rather than relying on traditional assessments every three months.

[0177] Through the artificial intelligence-based intervention system, the hospital can achieve more efficient and accurate cognitive function assessment. Mr. Zhang's cognitive training plan was adjusted in real time during the intervention process, and the intervention effect was significantly improved. The system provided a personalized cognitive intervention plan for the patient, successfully avoiding the "one-size-fits-all" general intervention strategy.

[0178] Table 1 Comparison of assessment data under traditional assessment and artificial intelligence-based assessment system

[0179]

[0180]

[0181] Based on the data comparison in the above table, it can be clearly seen that the cognitive impairment assessment system based on general artificial intelligence has significantly improved the personalized intervention effect compared with traditional assessment methods. Although Mr. Zhang scored 24 in the traditional assessment, indicating that he was in the range of mild cognitive impairment, the traditional method could not quantify mood fluctuations and cognitive load, nor did it have the ability of real-time dynamic adjustment, which made it difficult to track and optimize the intervention effect.

[0182] Under the artificial intelligence-based assessment system, the system can accurately evaluate the changes in Mr. Zhang's cognitive state by monitoring emotional fluctuations and cognitive load in real time. The emotional fluctuation index and cognitive load index are 12 and 15 respectively, indicating an obvious burden. After one month of intervention through the adjustment of intelligent algorithms, Mr. Zhang's cognitive load decreased significantly (-6), and the emotional fluctuation decreased (-4), which proves the effectiveness of the system in dynamically adjusting the intervention intensity.

[0183] Different from the fixed tasks in traditional assessments, the artificial intelligence system can dynamically adjust training tasks according to real-time feedback to ensure personalized intervention. The order, content, and difficulty of each task will be adjusted according to the patient's cognitive needs, thereby improving the intervention effect. The real-time feedback ability ensures the accuracy and effectiveness of the intervention, avoiding the lag of the traditional method of evaluating once every three months.

[0184] In addition, traditional assessment methods may rely on centralized storage, posing a risk of data leakage. In contrast, the artificial intelligence assessment system ensures data encryption storage and decentralized management through blockchain technology, guaranteeing data security and privacy protection.

[0185] In summary, the artificial intelligence-based cognitive dysfunction assessment system has obvious advantages in personalized intervention, real-time feedback, dynamic adjustment, and data privacy protection, and can provide more efficient and accurate cognitive intervention, improving the treatment effect and user experience.

[0186] The above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes should be covered within the protection scope of the present invention.

Claims

1. A method for assessing cognitive dysfunction based on general artificial intelligence, characterized in that: The steps include: S1. Collect multimodal user data, including voice, facial expressions, behavioral patterns, and physiological data, through wearable devices, mobile applications, and online quantitative assessment tools; S2. Preprocess the user's multimodal data, remove noise, standardize, and extract the user's key cognitive features, including language complexity, behavioral regularity, emotional response characteristics, and physiological indicators; S3. Combine the pre-processed user multimodal data, introduce the cognitive impairment knowledge graph, and use the adaptive multimodal graph neural network for comprehensive analysis to evaluate the user's cognitive ability and identify the areas of cognitive impairment; S4. Combining user multimodal data and evaluation results, using neuroevolutionary algorithms to generate rehabilitation training plans, dynamically adjusting the order, difficulty and type of training content according to the user's cognitive impairment; S5. Through rehabilitation training programs, combined with virtual reality and augmented reality technologies, an immersive cognitive training environment is provided, the user's virtual environment training data is collected, and the training tasks and difficulty are adjusted in real time; S6. Based on the pre-processed user multimodal data and virtual environment training data, dynamically predict the user's cognitive load and psychological state, and dynamically adjust the training content, environment and intervention intensity at different time points; S7. Based on the dynamic adjustment results, monitor the user's training effect in real time, generate progress reports and adjust the intervention plan according to the feedback.

2. The method for evaluating cognitive dysfunction based on general artificial intelligence according to claim 1, characterized in that: The S3 specifically includes: S31, extracting features from the pre-processed user multimodal data, specifically extracting key features from the user's voice, facial expression, behavior pattern, and physiological data, and standardizing them into a data format suitable for input into a neural network; S32. Introduce the cognitive impairment knowledge graph, map the multimodal features to the nodes in the cognitive impairment knowledge graph, perform multi-dimensional weighting and information fusion based on the complex relationship between the edges and nodes of the cognitive impairment knowledge graph, adopt a multi-level graph propagation mechanism, recursively calculate the influence between nodes, and obtain an enhanced feature vector: Among them, X i is the eigenvector of the i-th modal data, A i is the weight matrix corresponding to the features of node i in the knowledge graph, α i,k is the adaptive weighting coefficient of the relationship between node i and the kth layer of the graph, G j (X) is the graph propagation function, B j is the weight matrix of the jth node in the graph propagation process, X enhanced is the enhanced feature vector, K is the number of layers of graph propagation, M is the total number of nodes in the graph, N is the number of types of modal data, and β j,k is the weight coefficient of graph node j in the k-th layer propagation process; S33. Input the enhanced feature vector into the adaptive multimodal graph neural network, perform feature fusion through the graph convolution layer, and dynamically adjust the weights according to the complex relationship between nodes in the graph and the feature importance of each modality to generate a comprehensive cognitive feature vector. Through multi-layer graph convolution operations, the network gradually propagates and fuses information to calculate the comprehensive features: in, represents the output feature after the l+1th layer of graph convolution operation, σ is the activation function, A ij is the adjacency matrix between node i and node j in the graph, is the adaptive weight coefficient of node i and node j in layer l, is the weight matrix; S34, conduct time series modeling on the comprehensive features of the graph neural network output, use the long short-term memory network LSTM with attention mechanism and gate control to model the time series data, capture the dynamic trend of user cognitive ability, and combine the nonlinear weighted gate function G of the time series features t and the adaptive weighting coefficient α of the time step t , dynamically weighting the features at each time step to enhance the model’s sensitivity to cognitive fluctuations: H t =LSTM(F GNN,t ,H t-1 ,G t ); Among them, H t represents the hidden state of the long short-term memory network, the output at time step t, ⊙ is the element-by-element multiplication, F weighted is the weighted comprehensive feature vector, F GNN,t is the output feature vector at time step t, H t-1 is the hidden state output at the previous time step t-1, and T is the total number of steps in the time series; S35. Based on the results of graph neural network and time series modeling output, a multi-layer perceptron with a hierarchical weighting mechanism is used to evaluate cognitive ability and generate scores for each cognitive domain. A hierarchical weighted loss function is used to dynamically adjust the loss function according to the priority of each cognitive domain to optimize the cognitive evaluation results: Among them, Y i is the assessment score of the i-th cognitive domain, Y target,i is the real score, w i is the weight of the field, L weighted is the weighted loss function, and D is the total number of cognitive domains.

3. The method for evaluating cognitive dysfunction based on general artificial intelligence according to claim 1, characterized in that: The S4 specifically includes: S41. According to the cognitive ability assessment results, determine the user's cognitive impairment areas, define intervention targets for each impaired area, and generate an intervention target vector G = [g1, g2, ..., g m ], where g m represents the intervention target of the mth cognitive domain, and the priority of each target is sorted by the multi-objective optimization algorithm, optimizing the target priority vector P = [p1, p2, ..., p m ], where p m represents the priority value of the mth cognitive target; S42. According to the generated intervention target vector and target priority vector, the user's personalized intervention training plan is optimized using a neuroevolutionary algorithm. The combination of training tasks is T = [t1, t2, ..., t n ], where t n The parameters of the nth training task, including task type, difficulty, and training goal; S43, introduce an adaptive feedback mechanism to adjust the generated training tasks in real time, monitor the user's performance in training in real time through the feedback loop, and combine the real-time performance data D = [d1, d2, ..., d n ], using adaptive neural networks to dynamically adjust training tasks and optimize the difficulty and order of tasks in real time: Among them, X t For user input data at time step t, F t is the historical training feedback data, D t The generated dynamic adjustment feedback data, LSTM is used to capture the long-term dependencies in time series data, ARIMA is used to capture short-term dependencies and trend changes, α i is the weighting coefficient, n is the number of data feature dimensions; S44, combining the adjusted training tasks and user feedback, globally optimizing the training plan through adaptive meta-learning and multimodal learning; S45. Based on the optimization results, all training tasks and feedback are integrated into a preliminary personalized intervention plan P = [p1, p2, ..., p k ] and dynamically updates the intervention strategy based on the priority and schedule of each training task.

4. The method for assessing cognitive dysfunction based on general artificial intelligence according to claim 3, characterized in that: The S44 specifically includes: S441. The overall optimization goal of the training plan is defined by taking into account cognitive load, emotional fluctuations and task completion through a multi-objective optimization model, combining immediate rewards and long-term returns in adaptive meta-learning; S442, Genetic Algorithm generates a variety of training task combinations through selection, crossover and mutation operations, and combines the multi-modal dynamic weight adjustment mechanism to dynamically balance the fitness value of the task: Among them, F(T i ) represents the training plan T i The fitness value, T is the total number of time steps, M is the total number of tasks, ω j is the weight coefficient of task j, For task T i The immediate reward obtained at time step t, For task T i The impact on user cognitive load at time step t, is the emotion fluctuation value, α1, α2 are the balance coefficients of positive and negative influences, and β is the emotion fluctuation sensitivity adjustment coefficient; S443. Collaborative optimization among multiple tasks through dynamic adaptive adjustment of task weights: Among them, O AMTL represents the overall optimization goal of multi-task learning, K is the total number of tasks, T k represents the kth task, λ k is the dynamic weighting coefficient of task k, γ is the dynamic adjustment coefficient of task weight, ΔF(T k ) is task T k The fitness change, F avg is the mean fitness, F(T k ) represents the training plan T k The fitness value of S444. Through the combination of adaptive meta-learning and multimodal learning, the training tasks are adjusted dynamically in real time. The multimodal dynamic weight adjustment mechanism adjusts the task weights based on user feedback: Among them, T i,new are the dynamically adjusted training task parameters, is the previous task parameter, η is the learning rate, α m is the dynamic weight of task m in multimodal data, is the immediate reward of the task at the current time step, are the values ​​of cognitive load and emotional fluctuation, β1 and β2 are the weight coefficients of cognitive load and emotional fluctuation respectively; S445. By integrating the output results of genetic algorithms, adaptive multi-task learning and multimodal dynamic weight adjustment mechanisms, the training plan is dynamically adjusted, and the order, difficulty and type of training tasks are adjusted according to the feedback from different users.

5. The method for evaluating cognitive dysfunction based on general artificial intelligence according to claim 1, characterized in that: The S5 specifically includes: S51. Based on the generated personalized training plan, virtual reality and augmented reality technologies are combined to build an immersive cognitive training environment to simulate the cognitive scenarios in the user's daily life. The difficulty of the training task and the environmental parameters are determined by the training plan T = [t1, t2, ..., t k ] control, where t k Indicates the type, difficulty, and target time of the kth task; S52, real-time monitoring of the user's performance in the virtual environment, collecting training data including task completion time, error rate, reaction time, and emotional fluctuations, forming a user performance data set B t =[b1,b2,…,b n ], where b n represents the performance data of the nth training task; S53, perform time series modeling and feature extraction on the collected user performance data through spatiotemporal convolutional neural network, and output the user's behavior pattern features in the virtual environment H t , where the behavior pattern feature is the feature representation at time step t, capturing the spatiotemporal characteristics of the user's cognitive load, emotional fluctuations, and behavioral responses: Among them, K i is the convolution kernel, W i is the weight of each feature channel, σ is the activation function, * represents the convolution operation, W a is the learning weight in the self-attention mechanism, Softmax() is the normalization function, and n is the number of features in the user performance dataset; S54, based on the generated behavior pattern characteristics and training task objectives, the task settings in the virtual environment are dynamically adjusted using a generative adversarial network, the goal being to generate training tasks that are compatible with the user's cognitive state and emotional fluctuations. Specifically, the generative model G(H t ,T) generates new training tasks according to the user's behavioral characteristics and training task plan, and the discriminant model D(H t ,T) determines whether the generated task is suitable for the current user's cognitive needs: Among them, L GAN is the loss function of the generative adversarial network, T is the training task plan, is the expected operation, ln is the logarithmic function; S55. Generate an optimized personalized intervention training plan based on the optimized training task parameters and feedback results, and dynamically adjust the training tasks and difficulty in real time based on the real-time feedback.

6. The method for evaluating cognitive dysfunction based on general artificial intelligence according to claim 1, characterized in that: The S6 specifically includes: S61. A comprehensive user cognitive load and psychological state assessment model is constructed through pre-processed user multimodal data and collected virtual environment training data. Combined with the user's behavior pattern, emotional fluctuations and cognitive performance data, dynamic prediction based on time series is performed to assess the user's cognitive load, psychological state and emotional fluctuations, track the user's cognitive health status in real time, and accurately predict the load changes and emotional states at different time points: C t =f(D t ,E t ,Q t ); Among them, C t is the cognitive load and mental state evaluation at time step t, D t The user performance data collected in the virtual environment, E t is user behavior data, Q t is the emotional fluctuation data, f() is the comprehensive evaluation function; S62. Apply dynamic time series analysis technology to the assessed cognitive load and psychological state data, model the fluctuation of each user's cognitive state, continuously monitor the trend of cognitive load and emotion, timely identify the changes in the user's cognitive ability or the pattern of emotional fluctuations, and form the user's dynamic cognitive state trajectory: in, is the predicted cognitive load and mental state, C t-1 ,C t-2 …,,C t-k is the historical cognitive load and mental state data, A t is the external influencing factor of the current time step, and g() is the time series prediction function; S63. Based on the obtained cognitive state prediction results and in combination with the user's real-time feedback, dynamically adjust the training tasks, environment settings and intervention intensity: Where, ΔT t is the training task adjustment amount at time step t, α z and β u is the adjustment coefficient, B t Presenting datasets to users; S64. Based on the optimized training tasks and environment settings, the reinforcement learning algorithm is used to globally optimize the intervention strategy and generate personalized intervention tasks: Among them, γ is the discount factor, r t (θ) is the immediate reward at time step t, is the error measure between the user's prediction and the actual cognitive load, λ1 and λ2 are regularization coefficients, is the L2 regularization term, F(θ is the optimization objective function of reinforcement learning, E is the expected operation, T is the total number of steps in the training cycle, and n is the number of model parameters; S65. Based on the optimized intervention tasks and feedback results, the intervention training plan is adjusted dynamically over the long term, and the training content, environment and intervention intensity are adjusted dynamically based on real-time feedback.

7. Based on the cognitive dysfunction assessment system based on general artificial intelligence, the cognitive dysfunction assessment method based on general artificial intelligence according to any one of claims 1 to 5 is characterized in that: Includes the following modules: Multimodal data collection module, used to collect user multimodal data; Data preprocessing and feature extraction module, used to preprocess user multimodal data and extract key information describing the user's cognitive state; The cognitive ability assessment module is used to mine the cognitive features behind the data and locate the areas where the user's cognitive ability is most severely damaged; A personalized intervention plan generation module is used to design a daily rehabilitation exercise plan that matches the cognitive impairment status according to the severity of the impaired abilities; The result feedback and tracking module is used to monitor and provide feedback on the user's training process and results in real time.

Images